International journal of computer integrated manufacturing , tập 23, số 4, 2010

This research aims to facilitate a new engineering environment to build and configure machines from reusable smart modules, b concurrent engineering between product, process and control e

Planning in concert: A logistics platform for production networks

Jo´zsef Va´nczaa,b*, Pe´ter Egriaand Da´vid Karnoka,b

a

Computer and Automation Research Institute, Hungarian Academy of Sciences, Hungary;bDepartment of Manufacturing Science

and Technology, Budapest University of Technology and Economics, Hungary(Received 15 March 2009; final version received 18 January 2010)

In this paper the authors consider supply planning in a production network as a distributed effort for matchingfuture demand and supply by relying on asymmetric and partly uncertain information Even though decisions aremade autonomously and locally, partners should act in a concerted way For approaching the two main conflictinggoals of a high service level and low overall costs throughout the network, there is a need for a specific coordinationmedia for managing the intentions and interactions of the partners Starting from the design principles, the authorsdescribe a logistics platform that provides a complex service for communicating and evaluating all relevantinformation that may influence the operation of supply channels A particular interest is in coordinating a focalsupply network that produces customised mass products The implementation technologies of the system areoutlined together with the first lessons of the deployed application

Keywords: production networks; planning; coordination; cooperation

1 Introduction

The global behaviour of production networks emerge

from the interaction of local intentions and actions of

the partners Supply planning is considered in a

production network as a distributed effort for

match-ing future demand and supply under continuously

changing conditions Even though decisions are made

locally in an autonomous manner, partners in a

production network should act in a concerted way

Given their business goals, market and the production

environment that all evolve in time, partners have to

reason over their future courses of actions by

considering to some extent also the others’ situations

and intentions The problem is a distributed planning

problem: network members would like to exercise

control over some future events by relying on all kind

of information they have at hand Some of this

information can be considered certain, such as that

relating to products, production technologies, resource

capabilities, or sales histories An essential part of the

accessible information is, however, incomplete and

uncertain, such as those items capturing forecasted

demand, or expected resource and material

availabil-ity Industry strategists and academics – while

sketch-ing alternative trajectories for technological and

organisational developments – agree alike that

resol-ving incompleteness and uncertainty by proper

in-formation exchange is a matter of survival for any

production network that is to operate under volatilemarket conditions

However, uncertainty and the lack of information isonly one side of the coin; different, though equally hardproblems ensue from the plethora of information Whenpreparing the ground for informed planning decisions,

an enormous amount of behaviour related – i.e.,dynamic – data must be handled, synchronised, cleared,filtered, aggregated and archived The decision com-plexity of planning processes can but grow with theextension of input data, which is in sharp conflictwith the requirement of giving timely, almost instantresponses to queries during interactive planningsessions

Production informatics has well-proven proaches to handle uncertainty and structural com-plexity Aggregation merges detailed information onproducts, orders, demand forecasts, production pro-cesses, resource capacities, and time Various planningproblems – such as production scheduling, productionplanning or master planning – are formulated bymerging more and more details on longer and longerhorizons (Pochet and Wolsey 2006) Solutions aregenerated in a hierarchical planning process wherehigher-level solutions provide constraints to lower-level problems Hence, according to their horizon anddetail, plans can have strategic, tactical or operationaldimensions At the same level, decomposition

ap-*Corresponding author Email: vancza@sztaki.hu

Vol 23, No 4, April 2010, 297–307

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2010 Taylor & Francis

DOI: 10.1080/09511921003630092

separates planning problems into easier-to-solve

sub-problems This is the case when, e.g., on the tactical

level production planning is separated from supply or

distribution planning

The evolution of planning functions in production

management resulted in a generic hierarchical planning

matrix (Fleischmann and Meyr 2003, Stadtler 2005)

Figure 1 shows typical planning functions on the

strategic (long-term), tactical (medium-term) and

operational (short-term) level organised along the

main flow of information and materials These

func-tions are more or less common at each node of a

production network, though, of course, manifest

themselves in different forms and complexity

Within an enterprise the coordination of segregated

planning modules is a grave problem in itself

(Fleischmann and Meyr 2003, McKay and Wiers

2003) However, this issue is even more critical in a

production network where proper information

ex-change is the primary precondition of the

collabora-tion between the partners (Maropoulos et al 2006)

While the theoretical aspects of coordination and

cooperation in production networks have been

inves-tigated extensively for a long time, these studies have

paid due attention neither to the differentiation of

planning functions, nor to the underlying causes

Here, one can but remark that the surprisingly low

number of deployed manufacturing applications of

agent technologies – even on the ‘ideal’ field of supply

network management (Monostori et al 2006) – is also

a symptom of this lack of focus

The authors are motivated in bridging the gap

between the theory and practice of coordinated

planning in production networks The particular

back-ground to this work is a national industry–academia

R&D project that is aimed at improving the

perfor-mance of a network that produces customised mass

products (Va´ncza and Egri 2006, Monostori et al.2009) The network is woven around a focal manu-facturer by suppliers of components and packagingmaterials The manufacturer produces in an averageseveral million units per week from a mix of thousands

of low-tech electronics products Some of the productsare sold by retailers under their own labels what makesthe market situation extremely uncertain and complex

As Immelt (2006) remarks, ‘[If] you want to seesomething risky, try selling a lightbulb to a big-boxretailer.’ Against all these uncertainties, exploitingeconomies of scale of mass production technology is

a must

In this paper, a so-called logistics platform ispresented that was developed and deployed to facilitatecooperation of partners in this production network.Departing from the design rationale the system isdescribed, together with its main concepts andimplementation technologies Finally, application ex-periences and generalisations of the underlying modelare summarised

2 Problem statement2.1 Scope and objectivesProduction networks are considered as legally sepa-rated enterprises that are linked by material, informa-tion and financial flows They produce value in theform of products or services for the ultimate customer.The market increasingly demands products that arecustomised, yet available with shorter delivery times.Hence, the greatest pressures are time compression,customisation and cost reduction While any network

as a whole is driven by the overall objectives to meetcustomer demand at the possible minimal productionand logistic costs, the efficiency of operations and theeconomical use of material, energy and resourceshinges on the local decisions of the autonomouspartners These decisions are made necessarily byrelying on asymmetric and uncertain information Ingeneral, the network is coordinated, if the

service level of the overall network can beguaranteed at a predefined, reasonably highvalue, and

total production and logistics costs along thesupply channels are reduced to the minimum.Whenever order lead times acceptable by the marketare typically shorter than the actual production leadtimes, these main criteria are in conflict: owing touncertain market conditions, inventories (of compo-nents, packaging materials, products) are inevitable toprovide service at the required level On the otherhand, low costs can be achieved only with larger lot

Figure 1 Matrix of strategic, tactical and operational

planning functions

sizes, which involve, again, higher product and

component inventories as well as increased

work-in-process Though, if in the future demand unexpectedly

ceases for a product then accumulated inventories

become obsolete

The key to coordinated planning is just to master

such essential conflict situations time and again, in a

robust and reliable way The authors’ goal is to

develop such methods that are applicable also under

practical conditions When doing so, one has to face

issues related to the lack, uncertainty, inconsistency as

well as the abundance of information that might affect

the efficient operation of a production network

2.2 Related work

Research of coordinated supply planning goes back to

inventory management where the main questions are

when to order and how much to order However,

coordinating even a two-echelon supply chain via

orders is not really possible, because optimal

produc-tion quantities (and periods) depend also on factors –

such as set-up and production costs, resource

capa-cities – that are known to the supplier only Further

on, a supplier that serves several customers at a time

may exploit economies of scale by aggregating distinct

demands By placing orders, the customer intrudes into

the planning process of the supplier Typically,

centralised channel coordination models have been

developed where one of the partners had all the

information available to make optimal decisions The

centralised approach is though hardly realisable in

practice owing to the legally separated supply chain

partners Instead, upstream planning is the most

widespread form of collaboration at the moment

(Albrecht 2010)

Since in reality no partner can control the chain,

let alone a complete network, there is an increased

interest in decentralised control, both in deterministic

and stochastic settings (Cachon 2003) Actual

investi-gations take mostly the approaches of theories of

games and economy with asymmetric information In a

real network there is always an information gap

between the partners: the supplier is familiar with the

production and setup cost for the components, while

the end product manufacturer (in the customer’s role)

can estimate better the finished goods demand This

demand is distorted by the internal planning processes:

normally, master plans are generated which are further

refined into production plans and schedules In the

meantime, lot sizing decisions are made and parallel

component demands are aggregated As a result, the

actual component demand forecast can hardly be

related to the original finished good forecast (see

Figure 2) Although in the age of electronic

information exchange this gap could be bridged easily,partners do not have incentives for sharing privatebusiness information

On the practical side, there are various tion sharing solutions that support order processing

informa-in production networks One example is the lised SupplyOn platform for automotive and manu-facturing industries, developed by several Europeanautomotive part suppliers (SupplyOn 2007) It facil-itates information sharing between numerous plan-ning tasks in the fields of engineering, sourcing,logistics and quality management It uses ElectronicData Interchange (EDI) as a basic format, but alsosupports WebEDI, which technically requires onlyInternet access

centra-A similar approach called myOpenFactory (Schuh

et al 2008) proposes a centralised information sharingagency It is based on a standardised, industry–neutraldata and process model, and provides commercialservice with the implemented system The data format

is designed to be open and flexible, therefore it isspecified by XML schemas Just like in the previouscase, this solution is also confined to order processingand monitoring

2.3 Roadmap to cooperationNowadays, various paths to coordinated – and even-tually, even to cooperative – planning in productionnetworks are under extensive investigations Togetherwith related researches (Li and Wang 2007), for suchstudies the authors suggest the following generalroadmap:

(1) Development of processes and establishment ofmedia for sharing information about the actualand expected situations, as well as of the futureintentions (i.e., plans) of autonomous networkpartners

(2) At each level of aggregation, development ofpowerful local planners and schedulers whichare able to solve the richer, extended models.(3) Design and set up of incentive mechanisms thatfacilitate truthful information exchange, thesharing of risks and benefits, and cooperativebehaviour, after all

Figure 2 Transition of demand forecasts

In earlier research (Va´ncza and Egri 2006, Va´ncza and

Egri 2008, Monostori et al 2009), the authors

addressed all the above issues; some results are

summed up in Section 7 The sequel to this works

focused on the first issue, i.e., information exchange

3 Specification of the logistics platform (LP)

3.1 Design rationale

In order to make both the conflicting objectives of

coordinated planning manageable, it is suggested

detaching them: while the service level should be

tackled on the short-term (where information

concern-ing demand is almost certain), cost-efficient production

should be the main concern of medium-term planning

The two levels should be coupled by proper inventory

management Consequently, there is a need to share

and fit demand and supply plans on several levels

of aggregation, both in the medium and in the

short-term This matching should be performed time and

again, on a rolling horizon, with some tolerance to

eventual inconsistencies between aggregated and

de-tailed plans

However, as regards of making planning decisions,

the system should be passive and leave all planning

decisions to the actual (or future) local planning

modules The LP anticipates and calls the attention

of planners to conflicting situations (exceptions), but

does not make decisions on behalf of them Hence, it

provides only an interface both within and between

enterprises, for making better informed and timely

decisions

3.2 Requirements towards the LP

The functional requirements toward the logistics

plat-form are as follows:

The LP should link the planning functions of a

manufacturer and its suppliers both at the levels

of tactical and operational planning (i.e.,

pro-duction planning and scheduling, transportation

planning) This sharing makes the situation

symmetric, i.e., both partners along a channel

have access to the same information, at the same

time, in the same way

The LP should be flexible in supporting a wide

range of supply methods, from traditional

purchase orders via vendor-managed inventories

to coordinated supply

It should monitor near-time component supply

and give feedback information to both

aggrega-tion levels

It should support the evaluation and analysis of

the performance of supply channels, taking the

perspectives both of the customer and thesupplier Special regard is needed for anticipatingfuture situations where demand and supplymismatch

The LP should keep the privacy of individualcustomer-supplier relationships

Finally, the system should be prepared for ananytime, asynchronous and concurrent usage by

a large number of planners representing allpartners in a supply network

These functional requirements lead to the followingmain services of the system:

User authorisation and the management ofaccess rights and user profiles

System administration services for filling in andmaintaining the system’s master data

General navigation that provides easy access toall information on both levels

Automatic updating the system with dynamicdata on a regular basis, according to a givenprotocol

Checking the consistency of data, matchingsupply and demand plans both on the mediumand the short-term, anticipating shortage situa-tions and checking the fulfilment of inventoryhandling regulations, generating alerts and warn-ings While checking is automatic, alerts andwarnings are to be processed by the users Measuring, reporting and archiving pastperformance of channels, users and partners.Generating aggregate reports for specific pur-poses (such as capacity planning, transportationplanning)

The main requirement concerning the applied tion and communication technologies is that theservices of the LP can be accessible via the Internet

informa-Of course, the system should keep the privacy of eachcustomer-supplier relationship and comply even withthe most rigorous security requirements of thepartners Finally, as far as possible, it should use theexisting information resources and avoid storing andhandling data in a redundant way

4 System designFollowing the main design principles, the logisticsplatform is hierarchical and consists of two levels (seeFigure 3)

On the scheduling level, the supplier meetsthe exact, short-term component demand of thecustomer This demand is generated from the

actual daily production schedule of the customer

in form of call-offs and can be satisfied only by

direct, just-in-time delivery from an inventory

This short-term demand of the customer is

responded by the supplier’s actual delivery

schedule Decisions are made on a daily basis,

on a horizon of 1 to 2 weeks With this short

look-ahead, demand uncertainty is hedged by

safety stocks

On the planning level, the supplier has to make

preparations for satisfying the short-term

de-mand of the customer Hence, the supplier

receives medium-term demand forecasts of

com-ponents from the customer, together with some

information about the reliability of forecasts

Managing inventories, deciding about the

peri-ods and optimal lot sizes of production is the

supplier’s responsibility The demand forecast of

the customer should be acknowledged by the

supplier, either by simply accepting/rejecting it or

by responding with an appropriate production

plan On this level, decisions can be made in a

longer (even weekly) cycle

The LP is organised around the concept of the supply

channel Each channel is defined by a customer, a

supplier and a particular component that is delivered

through the channel There could be multiple channels

between the same partners, and the same component

may arrive through parallel channels Each channel is

assigned to at least one planner on both sides Planners

can handle only channels which are under their

authority There are two basic channel types:

Purchase Order channels for capturing the

traditional process, where supply is controlled

by the orders of the customer

Coordinated channels for managing supply out orders, giving more responsibility to thesuppliers

with-For each channel, there is a complex inventory posed of the in-transit, consignment, as well as of theon-hand inventories at the supplier and the customer.The LP keeps track of the inventory items on a dailybasis However, as Figure 3 shows, inventory is apassive element whose level is influenced by the localdecisions of the supplier (who builds up the inventory)and the customer (who consumes the inventory) The

com-LP collects, presents, analyses and aggregates allrelevant information concerning the future, presentand past of the channels Hence, each channel hasvarious dynamic future-related information, such asforecasted demand, open orders, scheduled demand(generated by the customer), production plan anddelivery schedule (generated by the supplier) Depart-ing from the actual inventories, projected inventoriesare calculated both in the medium and short term, andinventory statuses are evaluated from both partners’point of view

The channels can be controlled by the customeraccording to particular inventory handling rules.Minimum coverage rules guarantee that the produc-tion schedule of the customer can be executed even inface of uncertainties Hence, safety stock requirementsare expressed in this way Maximal coverage ruleswarn the supplier from overfilling the customer’sinventory In general, amount of the on-hand inven-tory at the customer should always be between theminimal and the maximal required quantities Rulescan be given either in terms of past average daily usage(backward coverage), fixed amounts, or future fore-casted demand (forward coverage) For PurchaseOrder channels, the LP checks also whether ordersreally cover the scheduled demand

The actual warnings and alerts are of distinctpriorities; e.g., short term shortage situations mustcertainly be avoided, while overfilling the inventory

in some distant period is far from being critical The

LP supports the planners in sorting and filteringthese exceptions according to type and priority;hence they can focus on the most critical cases first

In order to initiate informed decisions, the LPpresents statistical data about past usage of thecomponents, their substituting materials and otherdetails

The past performance of channels is evaluatedfrom both perspectives Since performance evaluation

is essential for coordinated and cooperative planning,these measures are also discussed in some detail (seeSection 6) Results can be aggregated and presented forany past period

Figure 3 Information flow through the LP

Summing up, the LP provides an interface between

various planning functions of the customer and the

supplier Decisions have to be made locally, but the

various checks – whether production schedule of

the customer is really served by a delivery schedule

of the supplier, or safety stocks are really sufficient –are executed within the LP

5 Implementation and system integrationThe developed and deployed version of the LP followsthe focal structure of the supply network where it isapplied The LP is a Java Enterprise Edition (EE) webapplication built on the customer’s proprietary webapplication framework This framework manages thedatabase connection pool, the request dispatching andcorporate Single Sign On (SSO) authentication (seeFigure 4) The application can be accessed from thecustomer’s intranet as well as from the externalsuppliers through the Virtual Private Network (VPN)

of the customer

Each user has an associated list of channels whichhe/she can see and modify This allows the privacy to beretained between different suppliers On a channel,every assigned user can read the same data, but users atthe sides of the customer and supplier have differentmodification rights For example, the supplier’s usercan modify the delivery schedule for the componentwhile the customer’s user cannot On the other wayaround, the inventory checking rules of a channel can

be set solely at the customer’s side (see Figure 5)

Figure 5 Detailed scheduling level data of a channel

Figure 4 Implementation technologies of the LP

The web application collects data from legacy

systems either via

direct Java Database Connectivity (JDBC) data

access to the customer’s scheduling and planning

systems, or

XML-based (eXtensible Markup Language)

data exchange with the suppliers’ and customer’s

enterprise resource planning (ERP) systems

The XML-based data exchange with the suppliers can

be automatic by using Secure-SOAP (Simple Object

Access Protocol) services built into the web application

or direct XML file upload in which case the

logged-in user’s account is used for the data validation

context

The data and information acquisition process

works in three different ways: periodically scheduled,

event-based and ad hoc Currently, a job is running

every morning right after the customer’s scheduling

system has generated its new production schedule

(and, subsequently, its scheduled component

de-mand) This job collects also the actual inventory

data as well as the material planning and forecast

data (from the customer’s planning system) During

the day, the customer’s operators can change the

initial schedule by hand and this change is propagated

automatically to the LP In the general case, the web

application’s administrator can trigger any time a

complete resynchronisation of the LP with the related

systems

The web application’s report and input screens are

designed for maximum data and access security by

utilising:

user roles, page level access check and object

access checks;

client- (JavaScript) and server-side form

valida-tion and data integrity checks; and

anti-SQL injection and Cross Site Scripting

(XSS) techniques by using only JDBC’s

Prepar-edStatement and HTML-encoding of all user

entered text before presentation

Beyond guaranteeing security, another primary design

goal was to make the access of the vast amount of data

behind the LP fast and filterable The speed

require-ment was achieved by using in-memory object caching

technology for critical data such as actual inventory

levels Filtering is a key feature in the application,

because each user can have hundreds of assigned

channels, but space and time restrictions allow them to

operate only on a small subset at a time Therefore

each user can define his/her own set of filters which he/

she can use later on in any situation The filters which

are logical constructs of 5property – value set4 pairsbelong to the personal profile of the users Note thatfilters are used also for collecting basic and generatingaggregate values for a set of channels, such as forevaluating the overall performance of a supplier who isresponsible for a number of channels

6 Performance evaluationThe traditional, order-based supply has standardperformance evaluation techniques that measure thefulfilment of orders on the one hand, and the inventory

or overall logistics costs on the other hand (Hon 2005)

In cooperative supply however, the responsibilities aremore complex because the decisions of any partnermay propagate throughout the network Certain kinds

of the uncertainties that emerge from an unpredictablemarket environment or from the system’s propertiesare hard to avoid (for a categorisation, see (Mula et al.2006)) However, in a production network the othermembers’ operation is also a source of uncertaintywhich is often charged by factors that are, in fact,unnecessary For example, when the supply planners ofthe customer are measured by the material shortage,then they tend to inflate the demand and forward toooptimistic plans towards suppliers On the contrary, ifthe planners are rewarded for over-performing theirplans, then they deliberately underestimate the demandand share too pessimistic plans with the suppliers Thesuppliers can be aware of the biases and may notaccept the demand information without critique Theyrather tinker the forecasts based on their own pastexperiences, but this cannot completely restore thequality of the forecast This way the network as awhole operates far away from its main objectives: whilethe service level is corrupted (and is to be restored timeand again by urgent orders at the price of increasedsystem nervousness and additional costs), large, evenobsolete inventories may also accumulate All in all,the selfish distortion of information necessarily de-creases the performance of the network Hence, in the

LP the performance of all partners is measured with akind of deviation between their promised and actualactivities

6.1 Measuring plan imprecisionThe customer’s main criterion is the deviation of themedium-term plans it shares with the supplier from itsactual component usage Hence, plan imprecision ismeasured by the difference between the plan and itsexecution On a rolling horizon, plans are generatedperiod by period with some look-ahead Since theseplans are overlapping, measuring plan deviation is notevident

One option is to use the standard measurement of

forecast errors that are generally some variants of the

This formula measures the difference between the

forecasted quantities and the realised demand in a

particular week i, where the fi–j,iforecast for week i was

generated on week i-j, uiis the realised demand of week

i, aj is a discount factor and n is the length of the

stability horizon, the basis of the measurement In a

commonly used version of this formula the absolute

value of fi–j,i – ui is taken An important property of

this kind of error measurement is that it charges a

double penalty if some demand is shifted from a week

to another within the stability horizon

However, in cases when demand is fulfilled from

the inventory, such shifts are almost negligible

when-ever the total demand is not changed Therefore also a

different type of measurement is proposed that

regards the precision of a single forecast generated

on a particular week The form of the plan deviation

measurement is the following:

deviationj;n¼Xn

i¼1

1

nfj;jþi  ujþi

In this case, discounting is not desirable, because it

would differentiate between the forward and backward

direction of the demand shift

Figure 6 presents a snapshot of a particular channel

history Weeks are indexed backwards from the actual

week For instance, the column of manufacturingweek -5 contains the forecasted demand for this week,generated on weeks -8, -7 and -6, respectively, while therow of week -5 shows the realised demand (1,108,530).Having calculated the errors for week -5, the maximalnegative deviation (MND) was -19.5%, the maximalpositive deviation (MPD) was 28.6%, the discountedforecast error (DFE) was 5.4%, while the discountedabsolute forecast error (DAFE) was 23.2% Accord-ingly, the demand was overestimated, and the rela-tively large distance between the deviations shows thatthe stability of plans are low (which is the cause of thenervousness syndrome) The row of week -5 containsalso the forecast for the next three weeks, made onweek -5 Since the realised demands of the next threeweeks are already known, one can compute the value

of the plan deviation (PD) which is 74.8%, i.e., thisforecast underestimated the demand

When comparing the absolute errors with theabsolute plan deviations one can see that errors areusually larger, because they contain double penalty fordemand shifting, while plan deviation disregards them.The choice between these two types of evaluationshould be based on the production and purchasecharacteristics: if the demand shifts cannot causeshortage or necessary re-scheduling, then plan devia-tion is appropriate, otherwise the forecast error should

Figure 6 Measuring weekly forecast error and plan deviation on a planning and stability horizon of 3 weeks

should not remain without their required components.

The authors have implemented the measurement of the

supplier’s service level using the principle of forward

coverage: the customer’s on-hand inventory together

with the supplier’s scheduled delivery must always

cover demand of the next few days Any situation

where this coverage does not hold calls for immediate

actions: either from the supplier (urgent delivery) or

from the customer (re-scheduling), or from both This

service level is measured with the CLIPA and CQPA

(Component Line Item/Quantity Production

Attain-ment) values:

CLIPAðnÞ

¼ 1  tasks without components of the next n days

total number of tasks on the next n days

CQPA is similar, but regards the number of items

instead of tasks

7 Providing input for the LP

7.1 Planning at the customer

Static, master data about the channels are fed into the

LP from the transactional ERP system of the

consumer As for the dynamic data, the customer

periodically runs a master planner that determines the

output of end products for a longer horizon

Depen-dent component demand forecasts are generated from

this master plan by standard material requirements

planning (MRP) methods (Hopp and Spearman 1996)

Demand for the same components are summed up and

feed into the LP as medium-term demand forecasts

Purchase order channels are also filled in with orders

generated by the supply planners of the customer

In parallel to developing the LP, an automatic

production scheduler was developed and deployed at

the production facilities of the customer (Dro´tos et al

2009, Monostori et al 2009) This scheduler assigns in

time each task to appropriate production resources so

that it guarantees the satisfaction of hard

technologi-cal, temporal and resource capacity constraints, while

approaches optimisation objectives such as maximal

resource utilisation and minimal backlog The

sche-duler concerns also the availability of material: in a

pre-determined time window it takes material

avail-ability as hard constraint and postpones tasks that

have no guaranteed supply Component supply is

sufficient if the total of inventories together with the

scheduled delivery matches the actual demand on each

day of this time window Short-term scheduled

demand for components is derived from this detailed

production schedule As for an example, consider

Figure 5: in the next few days no daily shortage is

projected, hence the production schedule is feasible,even thought with a longer, 10 days look-aheadmaterial shortage can already be expected given theactual inventories, scheduled delivery and demand

7.2 Planning at the supplierFor supporting the cost-oriented decisions of thesupplier on the planning level, the authors havedeveloped a portfolio of novel coordinated supplyplanning methods that take into account all thelogistics costs and calculate also with the uncertainty

of demand that may stop whenever market demand forthe end-product(s) built of the component ceases forany reason In this situation called run-out thecomponent inventory becomes obsolete The methodsdecide the time periods and quantities of componentproduction The total cost to be minimised includes thesetup, inventory holding and expected obsolete in-ventory costs Hence, decisions that coordinate aspecific channel can be made on the basis of informa-tion coming partly from the customer (demand and itsuncertainty) and partly from the supplier (setup,production and inventory holding costs) Variousmethods have been developed for different situations:typically, when plans should be made for the wholehorizon (Egri and Va´ncza 2007), or when it is enough

to plan for the close future (Va´ncza and Egri 2006).The coordinated supply planning methods havebeen extensively tested on industrial datasets andsurpassed the performance of the actual methods(Egri 2008), although, it is the individual supplier’sdecision whether they apply them or stick to theirexisting practice This is the case also with generatingdelivery schedules: while some suppliers rely on thebuilt-in methods of the system that produce deliveryschedule as a response to scheduled demand auto-matically, some apply more advanced distribution andtransportation planning methods and optimise notonly for minimal material shortage but for minimaltransportation costs, too

8 Experiences and extensionsThe logistics platform has been deployed at the focalmanufacturer of a production network producingcustomised mass products and is now used on a dailybasis both by the planners of this manufacturer (in thecustomer’s role) and of its strategic suppliers As fornow, channels for almost 10000 components (includingpackaging materials) have been set up Since thecurrent business processes are based on orders, onlyPurchase Order type channels could be defined.The first main experience is that thanks to thisapplication, the overall supply planning process

became much more transparent The misfit of the

aggregate planned demand and detailed schedules,

which is rather the rule than the exception in practical

applications (Buxey 2003), became evident at first

sight In contrast to the actual practice, the LP made

explicit a number of conflict situations For instance,

purchase orders for components do not cover in every

case the actual scheduled demand However, it did not

take much to the planners to see this positively: while

having the LP do the routine tasks (including the

execution of various checks) they could focus on the

really challenging tasks whose resolution needs human

intervention, negotiation and creativity

Just before the deployment of the LP, the automatic

production scheduler discussed above was set into daily

operation at the focal manufacturer Given cc 100

production lines, alternative bills of materials (BOMs)

and alternative routings and some thousands tasks to

schedule at a time, the scheduler strived to minimise the

delays and maximise resource utilisation However,

after making material availability checks using the

on-hand inventory at the customer, the production

schedule had to be modified extensively because of

short-term material shortage Now the scheduler and

the LP are synchronised and the material check

considers also the delivery schedules of suppliers

Provided suppliers are really able to keep their

promises, this integration leads to far less re-scheduling

and a more stable operation of the overall network

The routine use of the system made also clear that

the network as a whole needs novel business models

and processes An improvement easy to accomplish will

be the application of Coordinated channels where

orders will be generated automatically whenever

demand and supply matches on the short term

However, in reality the assumption of truthfulness

that underlies the logistics platform (and most systems

in production informatics) is untenable Planning

functions, even within the walls of the same enterprise,

have distinct, occasionally conflicting objectives To

remain on the safe side, they have incentives to distort

the information they share with others (see Section 6)

Note that this does not imply deliberate change of some

data; it is just enough to tweak some control parameters

of a complex planning procedure

All in all, in order to align potentially conflicting

incentives, a further step has to be taken towards

cooperation Hence, based on recent experiences, the

authors have started to design mechanisms that drive

the partners towards disclosing and using unbiased

information when trying to coordinate a channel

Accordingly, the supplier provides a service to the

customer by committing itself to meet all short-term

demand In return, the customer pays for (1) the

components delivered, (2) the flexibility of the supplier,

and (3) the forecast deviation (Va´ncza et al 2008) Theperformance measurements and the so-called coordi-nated channels will have a central role in thesedevelopments when the LP, that controls now onlythe flow of information and components, will beaugmented with the flow of financial assets The basis

of financial calculations will be a fair share of the costsand profits of operating on a risky market

Another option for cooperation is to tackle issues

of product design, production and logistics on thenetwork level and respond to dynamic market demandwith a product mix that can be built up from modular,easy-to-configure elements According to the initialexperiences, proper product design would significantlydecrease the complexity of planning processes and thetotal logistics costs

9 Conclusions

In this paper a logistics platform has been describedthat provides a complex service for communicatingand evaluating all the relevant information in aproduction network that may have an effect on theflow of materials The LP provides specific media andrules for managing the interactions and dependencies

of the partners By means of the LP, the productionnetwork becomes integrated both within and betweenenterprises At the same time, the LP links basicplanning functions of various aggregation levels withinthe enterprises

The main benefit of using the application is that thesupply and demand planners can see, compare andanalyse dynamic information coming from differentand heterogeneous sources This way, they cananticipate future conflict situations The final goal is

to make efficient local planning and reliable real-timeplan execution each partner’s primary interest This isthe key for inspiring cooperation in a productionnetwork whose overall performance depends on thedecisions of its autonomous members

Acknowledgements

This work was supported by OTKA No T73376 and OMFB

No 01638/2009 grants, as well as by the FP6 project AC/DC

con-Dro´tos, M., Erd}os, G., and Kis, T., 2009 Computing lower

and upper bounds for a large-scale industrial job shop

scheduling problem European Journal of Operational

Research, 197 (1), 296–306

Egri, P., 2008 Coordination in production networks PhD

thesis Eo¨tvo¨s Lora´nd University, Budapest

Egri, P and Va´ncza, J., 2007 A logistics framework for

coordinating supply chains on unstable markets In: P.F

Cunha and P.G Maropoulos, eds Digital enterprises

technology – perspectives and future challenges Berlin:

Springer, 59–66

Fleischmann, B and Meyr, H., 2003 Planning hierarchy,

modelling and advanced planning systems In: A.G de

Kok and S.C Graves, eds Supply chain management:

Design, coordination, cooperation Handbooks in OR and

MS, Vol 11 New York: Elsevier, 457–523

Hon, K.K.B., 2005 Performance and evaluation of

manu-facturing systems CIRP Annals Manumanu-facturing

Technol-ogy, 54 (2), 675–690

Hopp, W.J and Spearman, M.L., 1996 Factory physics –

Foundations of manufacturing management New York:

McGraw Hill

Immelt, J and Stewart, T., 2006 Growth as a process: An

interview with Jeffrey Immelt Harvard Business Review,

84 (6), 1–11

Li, X and Wang, Q., 2007 Coordination mechanisms of

supply chain systems European Journal of Operational

Research, 179 (1), 1–16

Maropoulos, P.G., Kotsialos, A., and Bramall, D.G., 2006

A theoretical framework for the integration of resource

aware planning with logistics for the dynamic validation

of aggregate plans within a production network CIRP

Annals Manufacturing Technology, 55 (1), 483–486

decision support for planning, scheduling and ing tasks in a focused factory Computers in Industry, 50,5–14

dispatch-Monostori, L., Va´ncza, J., and Kumara, S., 2006 based systems for manufacturing CIRP Annals Manu-facturing Technology, 55 (2), 697–720

Agent-Monostori, L., Kis, T., Va´ncza, J., Ka´da´r, B., and Erd}os, G.,

2009 Real-time, cooperative enterprises for customizedmass production International Journal of ComputerIntegrated Manufacturing, 22 (1), 55–68

Mula, J., Poler, R., Garcia-Sabater, J.P., and Lario, F.C.,

2006 Models for production planning under uncertainty:

a review International Journal of Production Economics,

103, 271–285

Pochet, Y and Wolsey, L.A., 2006 Production planning bymixed integer programming Berlin: Springer

Schuh, G., Kampker, A., Narr, C., Potente, T., and Attig,

Ph, 2008 myOpenFactory International Journal ofComputer Integrated Manufacturing, 21 (2), 215–221.SupplyOn, 2007 Available from: http://www.supplyon.com/[Accessed 15 January 2010]

Stadtler, H., 2005 Supply chain management and advancedplanning – Basics, overview and challenges EuropeanJournal of Operational Research, 163, 575–588

Va´ncza, J and Egri, P., 2006 Coordinating supply networks

in customized mass production – A contract-basedapproach CIRP Annals Manufacturing Technology, 55(1), 489–492

Va´ncza, J., Egri, P., and Monostori, L., 2008 A coordinationmechanism for rolling horizon planning in supplynetworks CIRP Annals Manufacturing Technology, 57(1), 455–458

A new vision for the automation systems engineering for automotive powertrain assembly

I Haqa, R Monfaredb*, R Harrisona, L Leecand A Westa

a

Department of Mechanical and Manufacturing Engineering, University of Loughborough, UK;bWolfson School of Engineering,

University of Loughborough, UK;cFord Motor Company, Dunton, UK(Received 26 March 2009; final version received 6 January 2010)Pressure on the powertrain sector of the automotive industry is mounting as market demand for higher variety andlower-cost automation systems increases To maintain the market competitiveness, design-to-market time for newproducts should be significantly shorter and considerable cost saving needs to be made during the design andmanufacture of production facilities Virtual construction, test and validation of systems prior to build are nowidentified as crucial because engineering changes owing to untested designs cannot be afforded any longer, andapproved designs need to be reused more efficiently

In this article, the authors report research collaboration between Loughborough University and Ford MotorCompany, to improve the current business and engineering model used in the powertrain industry The currentproblems are highlighted and corresponding industrial engineering requirements are specified The existing end-userand supply-chain interaction models are captured and new business and engineering interaction models are proposed

to address the requirements A set of engineering services required for the new interaction models is described and anevaluation approach to identify the impact of the new model on the current enterprises is explained In addition, anoverview is given on the research findings on the predicted impacts on the current businesses based on a set ofevaluation criteria

Keywords: automation system; engineering services; powertrain; process and simulation model

1 Background

According to a survey carried out in 2008 (SMART

2008), the global automation market is worth around

£180 billion with an average estimated growth of 7.8%

annually (prior to the current economical downturn)

Factory automation takes 38% of this market, of

which £62 billion is the size of the European market in

automation systems for control and monitoring

sectors These typically include application design,

simulation and modelling, manufacturing, installation

and maintenance

However, the global automation industry is

chan-ging rapidly The product lifecycles shrink but

demands for product variety and complexity increase

and therefore profit margins decline (Molina et al

2005) This industry is also facing the advent of

globalisation businesses, manufacturing practices,

or-ganisational and information structures are changing

rapidly Companies are moving from traditional

methods, where in-house development teams typically

work at a single site, to completely outsourcing or

using specialised designed teams working from

multi-ple sites

For rapid response to such ever-changing marketdemands, the automotive industry is under pressure toshorten production lifecycle time, for example whenintroducing new engine models in a powertrain sector.The time taken by western automotive firms to design

a new engine model, build production lines andcommence mass production is typically about 42months while Japanese firms take around 36 months(Harrison et al 2001, Monfared et al 2002, Haq et al.2007) Also it has been recognised in the automotiveindustry that 6 months’ delay for the launch of a newproduct such as motor vehicle or large subassemblies,e.g transmission units, will cause a significant reduc-tion of its profit margin (Lee et al 2007) However, theexisting state-of-the-art approaches to manufacturingautomation systems are facing fundamental limitationsand complexity to reconfiguration, integration withsupplier chain systems and optimisation Because of atraditional hard-coded deterministic approach to thelogical control of most production automation sys-tems, it is too rigid and inflexible to enable efficientconfiguration and robust operations (Harrison andColombo 2005), which is of particular importance

*Corresponding author Email: r.p.monfared@lboro.ac.uk

Vol 23, No 4, April 2010, 308–324

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2010 Taylor & Francis

DOI: 10.1080/09511921003596780

when existing plant is being upgraded or a new

production system is being installed

Therefore, the migration from today’s control and

management strategies to more flexible, intelligent

manufacturing systems is one of the most difficult tasks

facing this industry today It is envisaged that a more

proactive engineering approach and lifecycle support

to automation systems is required to facilitate highly

flexible and agile manufacturing systems capable of

providing easier and configurable design, installation,

commission and maintenance (Harrison et al 2006)

This article summarises ongoing research efforts on

the development of a new approach to the powertrain

sector of western automotive industry in particular

Ford Motor Company The research team at

Lough-borough University, in close collaboration with Ford

Motor Company as a major European automotive

manufacturer and its leading automation machine/

component builders (e.g Krause, Schneider Electric,

Bosch Rexroth), is investigating a solution for

im-proving the current engineering approach to design

and development of powertrain programs

2 Product lifecycle management

Business changes in all manufacturing sectors,

parti-cularly in the automotive sector, can be more

effectively achieved if appropriate manufacturing

systems are able to support reconfiguration, faster

ramp-up and better lifecycle support (Haq 2009) Such

change is not only limited to the technical systems but

it is also essential to extend it to the organisation and

employees to achieve an adequate level of

change-ability This transformation process becomes an

important business process that must be pre-planned

and managed effectively (ElMaraghy 2005) To

stream-line product development and boost innovation in

manufacturing by managing all the information about

an enterprise throughout the product lifecycle

(Sudar-san et al 2005), the concept of product lifecycle

management (PLM) was introduced in the 1990s as a

business strategy to rapidly plan, organise, manage,

measure and deliver new product and services much

faster and cheaper in an integrated way (Farhad and

Deba 2005, Ming et al 2005) The importance of PLM

solutions has been realised by investment of over $2

billion by different manufacturing companies, mainly

automotive and aerospace companies (Sudarsan et al

2005) However, a big gap still exists between the

increasing demands from industrial companies and

available solutions from vendors, e.g using traditional

product data management systems Exchanging

en-gineering data with suppliers has proved difficult, slow

and has geographic limitations Flawed coordination

among teams, systems and data interoperability and

complex approval processes are common (Ming et al

2005, Ming et al 2008) Furthermore, serious datainteroperability issues exist because the PLM systemsthat a company employs to support its activities can bemade of many components and each of thosecomponents can be provided by different vendors(Shyam 2006) Current available engineering systemsare considered too complex and general and aretypically not focused on the user’s specific needs Forinstance providing a visualisation environment toassist the end-user for concurrent design and validation

of machine control and mechanical layout is identified

as an important engineering requirements needed byend-users (SOCRADES 2008) To facilitate suchrequirements, different commercially available engi-neering solutions offer end-to-end PLM solutions andprovide an environment to implement three-dimen-sional (3D) models of production systems with editing,testing and debugging of system control logic against3D model (SOCRADES 2007) This includes DelmiaAutomation developed by Dassault Systems (DEL-MIA 2009) and em-PLC developed by Tecnomatix andSiemens as leading vendors (Tecnomatix 2009) Inthese applications, the implementation of a 3D model

of the production system (which leads to virtualengineering) is only possible from their proprietarycomputer aided design (CAD) packages rather thanthrough standard 3D formats Also their editingenvironment and user interface are quite complexand unintuitive Several training courses and a largeamount of support is required to complete theevaluation process, which reduces the effectiveness ofthese tools in a mixed/non-specialised skill environ-ment and in the early design phases Furthermore,end-users cannot modularise and reuse the designindependently from the vendors (SOCRADES 2008).Within the automotive industry the design, develop-ment and implementation of a new production systemswith subsequent lifecycle support involves many glob-ally distributed supply chain partners This requiresspecific engineering tools to enable virtual engineeringand manufacturing activities to occur concurrentlybetween globally distributed supply chain partners Inrecent years a potential breakthrough approach for asustainable manufacturing industry was initiated by aproject named Component based Paradigm for Agileautomation (COMPAG 2004) The major goal of thisproject was to achieve a more efficient and robust design,build, implementation and reconfigurability of anautomation system via a functionally modular/compo-nent-based approach In response different key areaswere identified and investigated with their required newengineering services in depth to improve efficiency andmodularity of automation systems This includes recon-figurability, virtual engineering, concurrent support to

product, process and control engineering, lifecycle

support and vendor’s independent and open engineering

environment Existing research at Loughborough had

created basic technology for a component based

approach to automation with the provision of new

engineering services But no research has been

under-taken on the application of this approach in a user

engineering and business context This research paper is

to summarise this prototype method and associated

engineering tools and to devise novel business and

engineering processes to enable the component-based

approach to be applied in industry

The research in this paper is based on on-going

research projects at MSI Research Institute,

Loughbor-ough University The focus of these research projects is

to develop methodologies and tools to support globally

distributed engineering of powertrain assembly

ma-chines A major goal of this research is to achieve more

efficient machine design and reconfigurability using a

functionally modular, component-based approach to

the powertrain assembly systems Similar to this

research there are many research projects For instance,

NSF Engineering Research Centre for Reconfigurable

Manufacturing Systems (NSF 2009), Radically

Innova-tive Mechatronics and Advanced Control Systems

(RIMACS 2007), Model Driven Embedded Systems

Design Environment for the Industrial Automation

Sector (MEDEIA 2009), Distributed Intelligent Sensing

and Control (DISC) for Automotive Factory

Automa-tion (DISC 2009) and Distributed IEC 61499 Intelligent

Control of Reconfigurable Manufacturing Systems

(IEC61499 2008) The mentioned research projects

typically investigated general lifecycle management or

very low-level automation systems design, and do not

sufficiently address reconfigurability of manufacturing

assembly systems in terms of their hard/physical and

soft/logical aspects The focus of this research is to

address existing and future challenges faced by

power-train assembly systems within the automotive industry

This research aims to facilitate (a) new engineering

environment to build and configure machines from

reusable smart modules, (b) concurrent engineering

between product, process and control engineering to

achieve up to 100% virtual design and validation of

manufacturing systems prior to build, (c) offer lifecycle

support from a new set of engineering tools, (d) vendor’s

independent environment and (e) provide support for

globally distributed engineering teams within the supply

chain of powertrain assembly systems (i.e remote

monitoring and maintenance)

3 Problems facing the powertrain sector

According to the European Automobile

Manufac-turers Association (ACEA), Europe is the largest

vehicle producer in the world with over 13 majorautomobile manufacturers, contributing significantly

in the EU economy (ACEA 2008) Ford MotorCompany is one of the world’s largest manufacturers

of its kind, taking around 15% of the European carmarket (Bekker 2009) Ford is involved with globallydistributed suppliers for their automation systemsdesign and development, i.e Powertrain Systems.With ever-growing emphasis being placed on globalproduction systems, service and lifecycle support hasbecome an integral part of the manufacturing system.However, similar to the other leading players in theautomotive industry, to maintain competitiveness,Ford is also facing extreme pressure to provide moreagile engineering system to enable rapid response tomarket changes (Haq et al 2007) Despite the extremeexpertise available in this industry, the engineeringsystems are fragmented and typically result in delays inproduction launch and therefore extending the ramp-

up time (the ramp up is the time required to get fromthe first day of series production to the point where it isable consistently to run at the design speed, commonlyknown as maximum ‘jobs per hour – JPH’) (Haq2009) The notion of modular and reusability of designand manufacturing is not new in this industry, none-theless, the infrastructure required to enable reuse ofpast production knowledge is still not in place.Furthermore, it is a common understanding that theability to rapidly reconfigure previous designs (e.g.customisation) must be embedded into the engineeringlifecycle; however application of various engineeringtools used by hundreds of suppliers, make it almostimpossible for the end-user (i.e Ford in this research)

to provide a consistent control over the engineeringdevelopment lifecycle Throughout the lifecycle of anautomation system, there is no common representa-tion/visualisation of engineering activities, betweensupply chain partners (Ong et al 2006)

To demonstrate the magnitude of the mentioned problems, which typically lead to delays inlaunch of a new product or shutting down theproduction line, it suffices to mention that according

above-to statistics captured by this research work, 50%saving in the ramp-up time would typically save 20million Euros in a typical European production line,and every minute delay/malfunctioning in productionline cost up to 6000 Euros for the end-users (Harrisonand Colombo 2005)

4 Current approach to the engineering of powertrainautomation

In a typical powertrain programme, a new engineproject starts with strategic planning and marketstudy, which leads to the identification of the product

specification and the requirements, volumes, and fund

approval Following several simultaneous engineering

meetings with suppliers and machine builders, the

machine production lines are conceptually designed

and manufacturing of lines is started at the machine

builders’ sites

The machine builders carry out the detailed design,

test and installation of machines, with frequent

inspections by engineers from the engine

manufac-turers Machine builders also subcontract machine

components to specialist component builders and

concentrate on overall line design, line assembly and

commissioning of the mechanical, electrical, hydraulic,

and control systems Typically about 4 months prior to

the completion of the project, the machinery should be

dismantled and delivered to the manufacturer sites for

final tests and try-out machining Onsite engineers then

perform a detailed examination of all production and

assembly lines at the site At the end of this stage

(known as job 1) engine manufacturer is ready to

produce the first engine and commence mass

produc-tion At the same time, the assembly lines are ready to

assemble various engine components to the engine

block The lines will be under constant inspection for

several months to avoid problems related to the

training, machine adjustments and maintenance

Conventionally the design activities by machine

builders take place sequentially beginning with

me-chanical engineering followed by electrical, hydraulic

and control engineering activities, as illustrated in

Figure 1 In the existing approach, the product

specifications are typically interpreted by process

engineers to produce a suitable machine configurationwith process cycle charts written to specify thenecessary timing of machine movements, which arelater interpreted by programs to produce structuredcontrol software Associated operator interface screensand machine diagnostics and monitoring applicationsare finally added (Harrison et al 2001, Harrison et al.2006) As a result, the design activities of the hardwareand control system remain isolated from one anotherand their verification can only be carried out duringcommissioning after build, which ultimately causes alonger and more costly ramp-up period

Despite the significant developments in thedomain of assembly system design there is still alack of well-developed assembly system engineeringtechniques and methodologies, also highlighted bydifferent research works (Harrison et al 2001,Harrison et al 2004, Harrison et al 2006), forexample: the existing state-of-the-art automationsystems are relatively effective but the approach todesign and build process is almost entirely sequentialand heavily segmented organisationally into differentengineering disciplines

This approach also has cost/quality impact later inthe production phase of the machines lifecycle Forexample if a change is required after several months ofoperation the engineers involved will be required torevise a large/if not all of the process in order toidentify and limit the impact of the change on themachine

Furthermore, the end-user is involved with anumber of suppliers, and therefore deals with

Figure 1 A convectional sequential approach to the development of automation systems

inconsistent document formatting and structures It

has been observed that translations to the end-user

required format has been a root of many problems

since there are (a) no common system representation,

(b) ad-hoc integration of engineering partners using

fragmented design tools, (c) the machine control logic

is only understood by specialists and (d) there is no

modelling of machine operations More comprehensive

description of the existing system is documented in

Monfared et al (2002) and Haq (2009)

5 Next generation of automation design and build

The present global and competitive environment poses

formidable challenges to global manufacturers

includ-ing the automotive industry To facilitate and

accom-modate unforeseen business changes within the

automotive industry, a new proactive approach is

required to design, build, assemble and reconfigure

automation systems Such innovative approach would

require promoting new technologies and engineering

methods to: (a) enable engineering concurrency, (b)

investigate design alternatives prior to building and

testing physical systems, (c) provide predefined and

pretested design components (as well as physical

components), and (d) enable application of virtual

engineering at the early stage of program design

phases Furthermore, such technologies and methods

needs to be sufficiently end-user oriented to allow them

as major investor on the systems to own the

engineer-ing knowledge and be able to reuse the business and

engineering knowledge for the future programs

A component-based approach to the development

of an automation system is illustrated by Figure 2.Lifecycle phases of automation system design anddevelopment and primary role of each supply chainpartner are depicted on the left-hand side of the figure.During machine design, implementation, build andvalidation phases, existing approach followed bysupply chain partners is also shown in the upper rightside of the Figure 2 Based on ten years of experiencewith world-leading automotive manufacturers (i.e.Ford and its supply chain collaborators), the Lough-borough research group has recognised that thispresent methodology for design/built is causing funda-mental limitations and difficulties in the service,reconfiguration, integration and optimisation of ma-chines, particularly in the face of rapid and oftenunpredictable business changes The current engineer-ing approach may offer adequate operational perfor-mance owing to well-proven and established methods.However, it is not able to cope well with new customerrequirements and globally distributed manufacturingdemands The current approach is typically dominated

by the use of general-purpose engineering tools and thecontinual reinterpretation of paper-based specifica-tions Throughout the design process, few tools areavailable to integrate and verify new design beforeactual building Such sequential nature of the detailedengineering design of automation systems provideslittle chance of concurrent engineering processes inorder to shorten the lifecycle Performing test andverification processes at the end of the design phasepresents risk of very costly rework on design and build

Figure 2 Proposed component-based approach to the development of automation systems

Moreover, lack of a repository system to store and

reuse design mechanisms and manufacturing process

modules (known as bill of processes) causes inefficient

reuse of engineering knowledge collected from

pre-vious engineering programs

A new modular approach proposed for the design

and build of automation system is also illustrated in

Figure 2 The vision is to decompose automation

systems into standalone subsystems and components in

a generic manner, which are configurable with a set of

parameters and may vary based on specific

applica-tions The components include complete design for

mechanic, electric, hydraulic, and control aspects, and

are commissioned fully in respect with the

function-ality defined for the component Such predesigned and

pretested components (or a combination of some

components and subsystems) are to be stored in a

library of reusable mechanisms In this approach the

concurrency of design can be significantly improved,

leading to compression of the program lifecycle, with a

much reduced risk of design-related malfunctioning

owing to the use of pretested systems modules This

new approach facilitates early virtual integration and

commissioning of pre-defined and pre-commissioned

mechanical, electrical fluid/software components As a

result less business/engineering process management

efforts are required and better lifecycle support can be

provided Nevertheless, the proposed approach to the

design demands a set of advance engineering services

such as comprehensive virtual engineering tools (to

develop and then deploy the library of mechanisms), a

consistent approach to the system design format acrossall supply chains, and a new business and engineeringinteraction model In order to support a new engineer-ing environment between globally distributed supplychain partners, a new vision of a Collaborative WorkCentre (CWC) is proposed The aim of CWC is toestablish and maintain a vendor- independent environ-ment in the form of generic and configurable buildingblocks of machine families (i.e library of modules ashighlighted in the Figure 2) prior to ‘productengineering’ Therefore CWC has significant potential

to bring agility within the manufacturing systems withpotentially reducing the time, cost and resources Thenew approach can also enhance the robustness of thesystem design, improving the responsiveness andcompetitiveness of automotive industry

5.1 Next-generation collaborative and configurableautomation systems (NGCCAS)

To deliver agility through modularity and ability within the future of automation systems, thisresearch work has proposed and developed a newrealisation approach called next-generation collabora-tive and configurable automation systems (NGCCAS).Conceptually the application of NGCCAS bringsagility and reconfigurability via new business andengineering process interactions for the powertrainautomation systems Figure 3 illustrate a realisationmodel based on the NGCCAS approach The principlefocus behind this new realisation approach is to

reconfigur-Figure 3 Proposed next-generation collaborative and configurable automation

primarily provide (a) reconfiguration, (b)

collabora-tion, (c) visualisation and (d) lifecycle support for

future automation systems The application domain of

NGCCAS includes integration of a generic library of

modules with the use of new engineering services

Initially with the application domain of NGCCAS up

to 100% virtual design, build and its validation and

verification could be achievable during study and

planning phases called ‘simultaneous engineering’, as

shown in the Figure 3 This figure describes a

systematic way to design and construct a new

automation system based on a more integrated,

concurrent and vendor-independent engineering

envir-onment However, complete business and engineering

process models required for the NGCCAS approach

are described later in this paper

Based on the future needs of the automotive

industry, CWC is being developed for multiple

facilities (the proposed idea is to hold and support a

number of different generic solutions) with minimum

complexity, risk, lead time and minimum skill level

using advanced communication technologies To meet

such requirements the CWC consists of (a) product

engineering (i.e engine), (b) required generic solutions,

(c) new engineering method (NGCCAS), (d) new

engineering services and (e) business and engineering

process work flows required to design and develop new

automation systems The existing ad-hoc integration

mechanisms are replaced by CWC to offer more

service-oriented support and collaboration within the

supply chain for future businesses Therefore such an

environment can bring engineering concurrency and

can investigate new design alternatives prior to

physical build and test of production/assembly

machines

As illustrated in Figure 3 the proposed CWC can

facilitate the study and planning phase to manage

virtual design, build and validation of new automation

systems for a new set of business requirements (e.g a

new engine) in a virtual environment (prior to the

physical build) The initial product of the study and

planning phase will be a validated virtual design for the

complete assembly or manufacturing lines related to

the new engine program The virtual design will then

be sent to the machine builders for final detailed

design, manufacture and installations within the

factory site To support migration from existing to

NGCCAS practice different steps prior to the actual

building and implementation of the automation system

are introduced during planning phases These steps are

called in-process steps (i.e step 1 to 8) as highlighted in

Figure 3, originating from the end-user requirements

and performed by domain experts Initially, it contains

a standard library of reusable, predefined and

pre-validated mechanisms, i.e predefined system

components and bill of processes (BOP) that isrequired to produce the components Such a library

is expected to be developed and completed gradually asknowledge of more engine program is captured Based

on preliminary reconfiguration of generic mechanisms,end-user planning teams can identify commonality ofthe new project in comparison with the past programsand develop new process plan for the reuse of theexisting system components and also develop newcomponents and subsystems level requirements Thisallows the program manager to make adequateplanning at early phases of the program Afterplanning, new mechanisms will be virtually designedand built by the supply chain experts and integratedinto the existing reused components Following thevirtual engineering, the mechanisms will be assembledwith existing mechanisms available in the library toachieve component, subsystem or systems Further-more, as part of the detailed design phase, analysis andoptimisation services will enable domain experts tooptimise components, subsystems and systems levelrequirements in terms of their cycle time, kinematics,and their control behaviours of the kinematics This isproposed to be provided through a simulation serviceswithin the machine design build phase (e.g checkingthe components/system design integrity, conditionsand interlocks, and cycle time) The validation process(i.e pre-commissioning) starts at the fifth ‘in-processstep’ as illustrated in Figure 3 It is proposed thatvirtual engineering services are required to enableverification of assembled components (and theirassociated subsystems) to verify fully the new devel-oped system prior to the real implementation of thephysical system At this stage, a set of engineeringapplication capabilities are prescribed to provide thevirtual engineering services Having completed thesubsystems verification, the complete manufacturingand assembly lines should be virtually tested andcommissioned By completing this process a new

‘Virtual Design J1 (VD1)’ milestone would be met.This new milestone is proposed to provide an approval

by the end-user to authorise development of thephysical systems

The proposed CWC also highlights the need forengineering capabilities to develop remote mainte-nance infrastructure to allow machine and componentbuilders to provide diagnostics, repair, and monitoringservices for the end-user during the installation andafter production launch In addition, the CWCproposes a model in parallel with the business modelconstantly to analyse (and predict) the programresources costs and time as it progresses

Furthermore, the proposed business and ing model described as NGCCAS approach potentiallyoffers significant improvement to the management of

engineer-the powertrain programs when a new variation of

en-gine is introduced to an existing line (known as second

cycle) This is mainly due to the reusability of the

system components and re-configurability of systems

based on the proposed engineering services

The new engineering model described above is being

developed in conjunction with real industrial case at

Ford Motor Company, in the UK In the remainder of

this article a prediction on the implementation impact

of the new engineering model is discussed

6 Evaluation of impacts on engineering processes

In a close collaboration between Loughborough

University and Ford UK, following many industrial

visits and several brainstorming sessions, some of the

more urgent user needs were realised as follows The

current ramp-up period and reconfiguration of

power-train assembly lines are too costly and too long The

scope of virtual engineering during different phases of

automation system lifecycle is limited due to the

application of general-purpose engineering tools

There is a great difficulty in reuse of the knowledge

from the past powertrain programs There is no

efficient way to predict the cost and effort required

for engineering changes (both product and processes).Currently a manual estimating process is used, which isslow, labour-intensive and at times does not generateaccurate study results Verification of design can only

be completed after build and installation, and thereforethey are very costly to change There is no uniformengineering application available to the end-user tomonitor, control and in later projects reuse theknowledge generated by globally distributed supplychain partners involved with a powertrain program

To develop the application of NGCCAS within areal industrial environment, Ford Engineering centre

at Dunton Technical Centre, and Dagenham EnginePlant, in Essex, UK were targeted It was envisagedthat prior to any recommendations for change, it isnecessary to understand the existing business andengineering processes and be able to propose newNGCCAS approach in a form compatible with theend-user business processes

Figure 4 illustrates the three step approach taken tocapture the existing processes for a typical powertrainprogram for the end-user and evaluation criteria/metrics defined to evaluate the required processeschanges when proposed migration is deployed asillustrated previously by Figure 3

Figure 4 Enterprise modelling approach deployed in this research

As shown in the Figure 4, the definition of the

end-users’ requirements constitutes the modelling

objec-tives in this research The aim is to understand and

subsequently compare the two engineering models

based on a set of criteria to enable prediction of the

change impact on the existing engineering model In

general, the Enterprise Modelling (EM) approach has

been adopted in this research to capture and formalise

system interactions The aim of EM is not only to

represent complex process structure of the

correspond-ing organisation, but also to identify and propose

possible improvements within systems EM provides a

solid foundation for the capturing, modelling and

analyses of the business and engineering system In this

research, EM has been used in compliance with

specifications defined by two international standards,

namely ISO-19439 and ISO-15704 (ISO 19439 –

Enterprise Integration, Framework for Enterprise

Modelling; ISO 15704 – Industrial Automation

Sys-tems, Requirements for Enterprise – Reference

Archi-tectures and Methodologies) The existing processes

were captured from the end-user sites and formalised

in the form of a set of static diagrams developed at

Loughborough University (Monfared et al 2002) in

compliance with the CIMOSA modelling architecture

(ESPRIT 1993, Berio and Vernadat 1999, Mertins and

Jochem 2005) The static models were used to develop

process simulation models, as both models share

similar modelling constructs (e.g process

decomposi-tions, information and resource objects) The process

simulation models facilitate customisation of models

based on different variables captured from the physical

environment Both process models and process

simula-tions have undergone a vigorous validation process

leading to the approvals from end-users and supply

chain engineers In addition, other standard validation

approaches were also deployed, such as those

sug-gested by (Robinson and Bhatia 1995, Robinson 1997,

Robinson 2006, Monfared et al 2007) The NGCCAS

approach continues by designing a new business and

engineering processes and supply chain interactions

The proposed model is designed based on the end-user

requirements and engineering services required to meet

those requirements The new engineering model is also

subjected to the process modelling, process simulation

and validation steps Completing the development of

both business process and interaction models, the two

models are compared on the basis of end-user most

important business performance metrics, e.g cost, time

and reliability of the design as engine program

progresses Finally, the process simulation models are

customised and the modelling results are analysed

based on the evaluation criteria, to provide predictions

on the impact of introducing the new business model

to the existing engineering systems

7 Case studyFord’s DVM4 (Dagenham Plant, UK) engine assem-bly line, known as ‘Tiger assembly line’ was considered

as the case study in this research, as shown in Figure 5.Ford’s engine production lines at this site are state-of-the-art industrial application to complex engineassembly operations The lines typically includevarious combinations of production resources such asmachines, conveyors, human operators The introduc-tion of a new engine project requires significantengineering competencies, time and budget Multipleend-user program teams are involved to coordinatethousands of parallel engineering activities with supplychain partners This assembly line was installed andcommissioned in 2007, with a substantial investmentfor a capacity to produce over 500,000 mixed productsper year (combination of various size diesel engines).KrauseTM(a global automation machine builder) wasmainly involved as a supply chain partner in the designand development of the Tiger assembly line includingconveyors and work stations The fully automatedstations with robotic arms were supplied by theABBTM suppliers (another global automation robotvendor) Assembly line illustrated by this figureconsists of work stations and transport system, i.e.conveyors that link together with various assemblystations A conveyor carries pallets with loaded engineblocks which are then moved onto different work-stations distributed along the transport system Atdifferent workstations various engine parts are as-sembled e.g pistons, connecting rods, cylinder headetc Sensors and mechanical stops are used throughoutthe transport system to track the pallets and directthem down to different conveyors according toinformation stored in each respective pallet

The expected working life of this assembly line isabout 7 to 10 years However, in today’s verycompetitive and turbulent automotive industry, anyassembly line with such long life is required to produce

Figure 5 Tiger assembly line, Dagenham Plant, Ford, UKand a typical schematic for similar engine assembly line

many different engine variations Therefore robust and

less costly re-configurable automation systems are

prerequisite

7.1 Process modelling

As discussed earlier, the process model is represented

in the form of a set of diagrams as enhanced CIMOSA

representation view (Monfared et al 2002) According

to this approach, system processes are decomposed

into Domain Processes (DP), Business Processes (BP),

and Enterprise Activities Depending on the

granular-ity of a model, a system can be broken down into

various combinations of these modelling constructs

The models are represented in a form of five different

interlinking diagrams formalising context, structure,

interaction, activity, and process definitions

The current business and engineering processes are

captured and formalised in the manner described

above Figure 6 illustrates a sample of the extensive

engine program model developed in this research

The process model is developed from the end-user

perspective in which ‘planning and business office

(DP1)’ and ‘manufacturing engineering (DP5)’

do-mains are responsible to design and build new

automation systems from program start to its

comple-tion The ‘DP5’ is further decomposed systematically

into seven subdomains as highlighted in Figure 6 Inorder to capture and model rigorous business andengineering processes within all eight different do-mains, a complete understanding of the ‘V’ systemengineering was developed as adopted by the FordMotor Company as shown in Figure 6 Differentmilestones from end-user perspective are introduced

on this ‘V’ model to assure successful completion ofnew automation from concept to launch Thesesmilestones are further mapped into business andengineering process domains Therefore all the eightdomains are firmly linked with ‘V’ model as shown inthe Figure 6 Further these eight domains are system-atically decomposed into activities and mapped withdifferent milestones on the time scale as highlighted inthe figure As an example Figure 6 describes activitiesperformed by DP1 and DP5.1 domains with theirrequired inputs and outputs Initially a new businesscase is developed by DP1 to meet future businesstrends, which ultimately become input to sub domains

of ‘DP5’ This includes ‘program management(DP5.1)’ and ‘program planning and feasibility(DP5.2)’ as shown Based on a new business case

‘DP5.2’ develops a comprehensive document on ‘newprogram planning and feasibility’ and delivers to the

‘DP5.1’ In parallel ‘DP5.2’ start communication withdifferent suppliers In response machine builders

Figure 6 Part of the process and interaction models developed for the current powertrain engineering system

proposed their new ideas and cost models Finally a

letter of intent is issued to the selected supplier

Program Management (DP5.1) starts when new

business case is initiated by DP1 and finishes after

completion of job 1 and the launch period Initially an

important document called ‘new program work plan’

and ‘long lead funding’ is developed and delivered by

‘DP5.1’ to business process ‘Simultaneous Engineering

(SE) BP525’ as highlighted in Figure 6 Based on these

documents end-user negotiates on time and cost for

new program with machine builders During SE

process detail level of understanding is developed to

identify system and subsystem level requirements

After detailed negotiations between end-user, machine

builder and control vendors program is approved by

board of directors and ‘First Order’ is placed by the

end-user to machine builder

Once first order is placed, machine builders start to

design and build new assembly machines At the same

time domain process ‘program engineering (DP5.3)’

start preliminary activities for mass production

Typically new machines are delivered to the end-user

over a 12 to 18 months’ time period During this time

these new machines are partially commissioned and

verified by end-user witness teams in a process called

‘first run-off and tryout phase (BP542)’ at the vendor’s

site Finally all newly built machines are shipped to the

end-user to start the domain process ‘installation and

commissioning DP5.5’ In this domain the machine

builders contribute with highly skilled commissioning

teams to prepare new machines for a vital domain

called ‘Running Rate and Quality Test (DP5.6)’ The

last domain of new program is known as ‘job 1 and

launch (DP5.7)’ The main focus of this domain is to

achieve rate of climb (ROC) i.e to run the line at its

designed capability rate Once ROC is confirmed

program management (DP5.1) publishes a completion

report in the business process ‘lesson learned (BP572)’

and confirm launch readiness for the vehicle plant

7.2 Process simulation

Process simulation modelling is used to measure and

analyse performance of static process models The

process simulation also allows customisation of the

process model based on the operational parameters

The simulation models analyse the process model over

a period of typical powertrain program and measure

key performance factors and allow execution of

‘what-if’ scenarios It was envisaged that application of

simulation model provides sufficient capability to

extend the static model to a dynamic environment

and enable the end-users to customise the model for

individual cases that ultimately assisted the validation

of the models However, other approaches such as

optimisation methods were also studied to providesupport for analysing the key performances, inparticular, calculating the robustness factor For thisparticular factor mathematical approaches were em-bedded into the simulation model to provide bothflexibility that simulation provides and accuracy in theresults

Figure 7 illustrates part of a developed processsimulation within this research In developing theprocess simulation the prime focus was to maintaincomplete consistency between static process modellingand simulation model in order seamlessly to integratethe key modelling constructs between the two models(i.e process and simulation models) One of theessential features within the process models used inthis research is a hierarchical support structure andreusability of modelling constructs used to developprocess models Similarly, the process simulationmodels (developed using ArenaTM commercial soft-ware (Seppanen and Kumar 2002, Bapat and Sturrock2003)) were designed in several levels (i.e submodels)

as highlighted in Figure 7, which correspond to thehierarchical structure and allow re-usability of model-ling modules These submodels represent eight differ-ent business and engineering process domains asillustrated by Figure 6 Enterprise knowledge capturedfor process modelling has different views (i.e func-tional, information, resource and organisation view).Such views are either defined as an input or outputrequired for each business and engineering process Todesign, build and execute the simulation model allthese views are categorised into: (a) functional objects,and (b) behavioural objects Functional objects areeither inputs or outputs for each process, which may ormay not dependent on other process functional objects(e.g flow of information, physical resources etc.) Onthe other hand behavioural objects describe the logic

or sequence of processes, i.e to define process logicallyeither to make sequential or concurrent flows Toutilise these objects within the process simulationenvironment, simulation parameters (i.e executionvariables modules) were used to facilitate populatingand configuring the process simulation models forspecific powertrain program In order to facilitatetracing and validation of simulation variable in suchlarge and complex models, variables (e.g information,event, time or human resource) are defined in asymbolic way to represent their concerned domainprocesses For instance ‘INFO_BP435’ or ‘PR_EA21’represent information or a physical resource object for

a certain business process or enterprise activity Thisapproach also enables triggering simulation processbased on preconditions, which correspond to thebusiness and engineering processes of the powertrainprogram Furthermore, such a structured approach to

the simulation design allows integration of the

simula-tion models to (and from) other engineering

applica-tions, such as existing project management tools, and a

central data repository system

As illustrated by Figure 7, the simulation process

runs based on a set of default parameters suggested by

the end-user for typical (semi-generic) programs The

users vary parameters and execute the model In

addition the Arena tool supports the quantitative

analysis (e.g probability function and random

genera-tion of entities) as required in this research to measure

and compare future approaches robustness In this

research an innovative method is developed to enable

domain experts to quantify and predict robustness of a

given system for selected issues before implementation

For instance Figure 7 illustrates comparison between

existing and new NGCCAS approach against those

specific issues which were identified and quantified

Following execution of a number of simulation

replications, the modelling results are exported to

external analysing tools to be compiled in a suitable

format (e.g reports, comparison graphs, etc.)

7.3 New engineering process and interaction models

Innovative engineering and interaction models are

proposed to address a number of user requirements

identified at the earlier stage of this research These

include: development of design mechanisms libraries tofacilitate reusability, introducing new engineeringservices to enable consistent virtual design across theend-user (and supply chain) program lifecycle, descrip-tion for new supply chain interaction models tointegrate the machine/component builders engineeringefforts in line with the end-user activities, and providemore concurrent design processes to shorten theoverall program time

Figure 8 illustrates part of the new engineeringprocess model developed in this research As partiallyappears in the figure, libraries of pretested compo-nents, bill of processes, and design mechanisms areavailable to the end-user engineers at various phases ofpowertrain program The required engineering servicesare interpreted into engineering application toolsrequired at each stage of the engineering model Theproposed models identify in great details whatapplication functionality (e.g component builder orsystem viewer – see Figure 8) is required for eachbusiness process (BPs) and what engineering expertise(with what skill level) should use the new engineeringapplications The new model also specifies changes onthe current process flow, information and resourcerequirements for each process Furthermore, it intro-duces a set of interaction mechanisms with the supplychain (e.g exchange of information, documents andthe timing within the program lifecycle) to outsource

Figure 7 Snap shots of the developed simulation and analysis models

certain part of the design process without losing

control over the program management or the

knowl-edge ownership Comparing the models illustrated by

figures 6 and 8, it is clear that due to the application of

virtual engineering, more engineering activities can be

completed concurrently, which will result in

compres-sion of the program overall time In addition, the new

‘Virtual Design 1’ (VD1) program management

mile-stone introduced in this research will represent the

phase that theoretically the design of all processes and

facilities are completed virtually and tested fully (up to

100% as libraries are gradually populated) For further

detail on the proposed engineering model refer to (Haq

2009)

7.4 Evaluation criteria

After deploying the new business and engineering

approach, to predict potential improvement on

power-train automation system, a number of evaluation

criteria were identified based on the end-user

require-ments These criteria naturally include cost and time

In this research domain, the cost and time of the design

and manufacturing processes and resources, and the

cost of changes (due to errors, or design changes) are

from outmost important factors However, there are

unquantifiable factors that also have significant impact

on the program performance and are key indicators for

the end-users to evaluate an engine program design

and build For instance, the correctness of the design at

each stage of the design of manufacturing andassembly lines has direct impact on the cost and time

of engine program A slight design misalignment at theearly stages of the program lifecycle, may lead to amajor costly and timely re-design or re-build duringthe test and installation In this research, an innovativemethod was developed to enable measuring andprediction of improvement on design correctness ateach phase of the lifecycle The ‘robustness’ factor wasintroduced as new evaluation criterion to compute (viasimulation model) the potential improvement on thedesign correctness due to the application of pre-defined/pre-validated system component approach.The robustness is defined as a risk factor in achievingplanned automation system design due to certainproblems associated with design processes A robust-ness ratio is calculated based on multiplication of(a) severity of impact on the production, (b) frequency

of occurrence, and (c) ability to detect and eradicate

a problem at a certain phase of the lifecycle Thedesign problems during the production launch arecategorised into six different groups These problemsinitiate due to inaccuracy in: (a) product conceptdesign, (b) time to achieve production volume, (c)machine design, (d) tooling process, (e) predictedbreakdowns, and (f) productivity assumptions Differ-ent robustness ratios are calculated for eight differentdomain processes modelled (see Figure 6) for this casestudy Based on the data captured and analysed fromthe end-users, quantified values were associated to each

Figure 8 Part of the new business and engineering models proposed in this research

elements of the robustness (i.e severity, occurrence,

and detection)

A robustness simulation model was developed to

compute the robustness of each domain process based

on calculating the probability of each three elements of

robustness ratio against the six groups of design

problems

The occurrence of the design problem in this

domain was envisaged to have a uniform distribution

owing to the nature of the domain processes In

addition consecutive design problems would

accumu-lative impact on the overall robustness Therefore a

cumulative discrete probability function was selected

for calculating the robustness

The robustness ratio (Rr) for each design problem

group of one domain process is calculated by Rr ¼

Sv*Oc*Dt (i.e Severity, Occurrence, and Detection)

The probability of occurrences (POc) are calculated by

the simulation model based on distribution function of

DISC(CumOc1, XRep1, , CumOcn, XRepN), when X is

the possible discrete value and Rep is the range of the

function The range in this simulation model defined

by the number of replications of the simulation model

and was set to 25 (larger number of replications has

insignificant impact on the simulation results)

There-fore overall robustness (R) for one domain process is

calculated byRrP6

0POc, which is cumulative ability of six different problem groups for one domain

prob-process Similar approach was taken to calculate the

overall robustness related to one problem group

against the eight domain processes

For instance, it is unlikely to reach the nominal

volume of engine production as originally scheduled

This is attributable to many reasons at different phases

of the design and build In this example, according to

the data captured from domain experts, problems

occurred in the ‘Program Planning and Feasibility’

domain (DP5.2) have severity (Sv) impact of 8 (out of

10) on achieving nominal volume However at this

phase of the engine program it typically occurs on 60%

of the cases (Oc) and can be detected (Dt) at this phase

in 6 out of 10 cases The simulation model shows that

robustness (R) of the system design at this domain is

52% that means there is a 48% chance that nominal

volume will not be met as scheduled owing to problems

in DP5.2

8 Predicted results

The developed process models of the powertrain

program highlighted problems areas in the current

business and engineering processes The findings

include lack of infrastructure and application tools to

enable reusability of knowledge (e.g design and

processes) and lack of ability for rapid reconfiguration

of design after process/product changes or following asecond cycle production plan (i.e introducing newproduct to the production line) These are represented

in the process models as lengthy and expensivebusiness processes, and also as major delays onproduction launch It was also realised that many ofthe process bottlenecks identified by the developedmodels correspond with the initial business require-ments set stated by the end-users This fact not onlyvalidates the reliability of the modelling approach, butalso highlights that end-users understand their currentengineering problems as a whole, however they havedifficulties in pinpointing the problems within thecontext of their engineering lifecycle and thereforeunable to rectify them

The new business and engineering process modelprescribes an enhanced approach for managing power-train programs, which is predicted to improve some ofthe current problems This should be possible throughintegration of proposed new engineering services (andtheir application tools – see Figure 8), and revisedsupply chain interaction models suggested by thisresearch In addition, development of pre-validateddesign modules and the library of reusable moduleshave key importance on the business processimprovement

The process simulation models led to an extremelydetailed calculation and comparison of system speci-fications before and after implementing the newbusiness model The modelling outcomes indicatevery promising results in terms of saving in engineeringcosts, shortening the processes, and improving thereliability of the design of production lines (viacalculation of the design robustness) The modellingoutcomes are summarised in Table 1

For instance, some of the areas that is predicted to

be influenced heavily by the proposed approach are

‘Program Planning and Feasibility- DP5.2’, ‘ProgramEngineering – DP5.3’, and ‘Job1 and Launch – DP5.7’.(Further information on the modelling approach andcomplete modelling and predicted results are docu-mented by Haq (Haq 2009).) The modelling resultspredict 38% reductions on the length of DP5.2, and63% on DP5.3 Similarly, the process DP5.7 isexpected to initiate 5 months earlier than the currentapproach (in a 42-month program) As a result overall24% less time and 27% less resources can save 30%cost for future new powertrain programs In particular,resource group called ‘process and automation en-gineering’ requires 37.5% less resources However, anincrease in ‘Virtual Engineering’ efforts by 3 times ispredicted Furthermore, the mathematical calculationssuggest a significant improvement in overall designrobustness For instance, the robustness ratio in DP5.2and DP5.7 is expected to increase from 52% and 60%

to 92% and 96% This has been achieved with theproposed new business and engineering processes, newmilestones and usage of new engineering tools ashighlighted in Figure 8 This is a significant improve-ment on the design process, which has direct impact ontime and cost of the programs by avoiding reworksand changes.

However, to understand fully the impact of theproposed next generation collaborative and configur-able automation systems (NGCCAS), the new ap-proach should be implemented within an industrialenvironment Preparation is being made to implementthe NGCCAS approach within the Ford engineeringcentre during its next major engine program in the

UK Initially, the new engineering model will be used

in parallel with the program management system toshadow the processes This will enable a directcomparison of predicted results with actual benefits

of utilisation of engineering tools in real engineeringprocesses

9 ConclusionsLack of agility and responsiveness to the marketchanges were identified as some of the existingproblems with automation industry An approach tothe next-generation collaborative and configurableautomation systems (NGCCAS) was proposed toimprove the current problems in the design and build

of powertrain automation systems The developedapproach proposes establishing a link between busi-ness requirements and engineering applications Itprovides changes in the business and engineeringprocesses within this sector of industry, and describes

a new supply chain interaction mechanism Theexisting enterprise processes were captured and com-pared with a new model of the engineering paradigm.The comparison indicates considerable improvement

in the way current automation programs work A set

of engineering services combined as a collaborativework centre (CWC) was defined to be integrated in thecurrent business and engineering model Descriptions

of the corresponding engineering applications andtheir potential implementation phase within theengineering lifecycle were briefly discussed The use

of enterprise modelling and process simulations wasdiscussed to visualise the current and the futureenterprise processes, and enable detail analyses ofvarious production scenarios

On the basis of the modelling approach taken, itwas predicted that the application of the NGCCASapproach should enhance significantly the agility andresponsiveness of automation system development.The new engineering model is planned to be tested in areal industrial environment as part of collaboration

between the research group in Loughborough

Uni-versity and Ford Motor Company, UK This research

to date has been principally aimed at the end-user

However there is strong desire to expand this core

concept within the business context of other supply

chain partners This will identify their detailed business

needs and to understand their current approach to the

design and build of powertrain automation systems A

large body of further research is needed to extend the

proposed idea of a CWC and to study new role of

supply chain partners and their more service-oriented

relationships Particular attention is required to

con-sider the costs and deployment efforts needed in

developing a generic bill of process i.e libraries of

mechanisms within the powertrain sector of the

automotive industry

Acknowledgement

The authors gratefully acknowledge the support from

Innovative Manufacturing and Construction Research

Cen-tre (IMCRC) as part of the Business Driven Automation

Project

References

ACEA, 2008 The engine of Europe European Automobile

Manufacturers’ Association [online] Available from

http://www.acea.be/index.php/news/news_detail/the_engine_

of_europe/ [Accessed March 2009]

Bapat, V and Sturrock, D.T., 2003 The Arena Product

Family: Enterprise modelling solutions Proceedings of

the 35th conference on Winter simulation: driving

innova-tion, New Orleans, Louisiana, 210–217

Bekker, H., 2009 Best-selling car manufacturers in Europe in

2008 International trade [online] Available from http://

internationaltrade.suite101 com/article.cfm/bestselling_

car_manufacturers_in_europe_in_2008 [Accessed March

2009]

Berio, G and Vernadat, F.B., 1999 New developments in

enterprise modelling using CIMOSA Computers in

Industry, 40, 99–114

COMPAG, 2004 Component based Paradigm for Agile

Automation [online] Available from http://www.lboro

ac.uk/departments/mm/research/manufacturing-systems/

dsg/doc/compag.htm [Accessed February 2009]

DELMIA, 2009 DELMIA digital manufacturing and

pro-duction [online] Available from http://www.3ds.com/

products/delmia/welcome/ [Accessed March 2009]

DISC, 2009 Distributed intelligent sensing and co ntrol

(DISC) for Automotive factory automation [online]

Available from http://www.oooneida.org/RD_projects_

DISC.html [Accessed March 2009]

ElMaraghy, H., 2005 ‘Flexible and reconfigurable

manu-facturing systems paradigms International Journal of

Flexible Manufacturing Systems, 17 (4), 261–276

Esprit, C.A., 1993 CIMOSA: open system architecture for

CIM Berlin: Springer-Verlag

Farhad, A and Deba, D., 2005 Product lifecycle

manage-ment: closing the knowledge loops Computer Aided

Design & Applications, 2 (5), 577–590

Haq, I.U., 2009 Innovative configurable and collaborativeapproach to automation systems engineering for automo-tive powertrain assembly Thesis (PhD) Wolfson School

of Mechanical and Manufacturing Engineering borough, Loughborough University

Lough-Haq, I.U., et al., 2007 Lifecycle framework for modularconfigurable automation systems Fourth IFAC Confer-ence on Management & Control of Production andLogistics, Romania, 465–470

Harrison, R and Colombo, A.W., 2005 Collaborativeautomation from rigid coupling towards dynamic recon-figurable production systems 16th IFAC World ControlCongress, Prague

Harrison, R., et al., 2004 Lifecycle engineering of modularautomated machines 2nd IEEE International Conference

on Industrial Informatics, 2004 INDIN ’04, 501–506.Harrison, R., et al., 2001 Distributed engineering ofmanufacturing machines Proceedings of the Institution

of Mechanical Engineers, Part B: Journal of EngineeringManufacture, 215 (2), 217–231

Harrison, R., et al., 2006 Lifecycle engineering of futureautomation systems in the automotive powertrain sector.In: IEEE International Conference on Industrial Infor-matics, 305–310

IEC61499, 2008 Distributed IEC 61499 Intelligent Control ofReconfigurable Manufacturing Systems [online] Avail-able from: http://www.ele.auckland.ac.nz/*vyatkin/publ/MITRA.pdf [Accessed September 2009]

Lee, S., et al., 2007 A component-based approach to thedesign and implementation of assembly automationsystem Proceedings of the Institution of MechanicalEngineers, Part B: Journal of Engineering Manufacture,

221 (5), 763–773

MEDEIA, 2009 Model-driven embedded system design

(MEDEIA) [online] Available from: world org/ProjectDetails.aspx?ProjectId¼5ab450360ec246aeb244128a1356d12f [Accessed September 2009].Mertins, K and Jochem, R., 2005 Architectures, methodsand tools for enterprise engineering International Journal

http://www.ist-of Production Economics, 98 (2), 179–188

Ming, X.G., et al., 2005 Technology solutions for borative product lifecycle management – status reviewand future trend Concurrent Engineering, 13 (4), 311–319

colla-Ming, X.G., et al., 2008 Collaborative process planning andmanufacturing in product lifecycle management Com-puters in Industry, 59 (2-3), 154–166

Molina, A., et al., 2005 Next-generation manufacturingsystems: key research issues in developing and integratingreconfigurable and intelligent machines InternationalJournal of Computer Integrated Manufacturing, 18, 525–536

Monfared, R., et al., 2002 An implementation of thebusiness process modelling approach in the automotiveindustry Proceedings of the Institution of MechanicalEngineers, Part B: Journal of Engineering Manufacture,

216 (11), 1413–1427

Monfared, R.P., et al., 2007 Evaluating a new FlexibleSoldering System for Electronics SMEs Proceedings ofthe Institution of Mechanical Engineers, Part B: Journal

of Engineering Manufacture, Part B, 222, 273–283.NSF, 2009 NSF Engineering Research Center for Reconfigur-able Manufacturing Systems [online] Available from:http: //erc.engin.umich.edu/research/thrustareas/ TA1.html

Ong, M.H., et al., 2006 A structured approach to evaluating

the impact of implementing a component-based system in

the automotive engine manufacturing domain

Interna-tional Journal of Production Research, 44 (13), 2645–

2670

RIMACS, 2007 Radically Innovative mechatronics and

advanced control systems [online] Available from:

http://www.ist-world.org/ProjectDetails.aspx?ProjectId¼

058a4a31cf604869ba0825e07417a065 [Accessed

Septem-ber 2009]

Robinson, S., 1997 Simulation model verification and

validation: increasing the users’ confidence Proceeding

of the 1997 winter simulation conference, 53–59

Robinson, S., 2006 Conceptual modeling for simulation:

issues and research requirements Proceedings of the 2006

Winter simulation conference, 792–800

Robinson, S and Bhatia, V., 1995 Secrets of successful

simulation projects Proceeding of the 1995 winter

simulation conference, 61–67

Seppanen, M.S and Kumar, S., 2002 Using simulation to

teach business processes design and improvement

Proceedings of the 34th conference on Winter simulation:

exploring new frontiers, San Diego, California, 1809–

1814

Shyam, D., 2006 Industry requirements and the benefits ofproduct lifecycle management Thesis (MSc) School ofIndustrial and Manufacturing Science Cranfield, Cran-field University

SMART, 2008 Monitoring and control: today’s market, itsevolution till 2020 and the impact on ICT on these,European Commission, DG Information Society andMedia [online] Available from: http://www.decision.eu/smart2007.htm [Accessed September 2009]

SOCRADES, 2007 Investigation of enabling technologies,methods and tools for application engineering [online].Available from: http://www.socrades.eu/Home/default.html [Accessed September 2009]

SOCRADES, 2008 Early prototyping of application tion and support methodology Available from: http://www.socrades.eu/Home/default.html [Accessed Septem-ber 2009]

descrip-Sudarsan, R., et al., 2005 A product information modelingframework for product lifecycle management Computer-Aided Design, 37 (13), 1399–1411

Tecnomatix, 2009 Siemens PLM software transforming theprocess of innovation[online] Available from http://www.plm.automation.siemens.com/en_us/products/tecnomatix/[Accessed March 2009]

A computational simulation approach for optimising process parameters in cutting operations

A Jeanga*, Huan-Chung Liband Yi-Chi Wanga

a

Department of Industrial Engineering and Systems Management, Feng Chia University, Taichung, Taiwan, ROC;bDepartment of

Industrial Engineering and Management, Chin Min Institute Technology, Miao Li, Taiwan, ROC

(Received 17 March 2008; final version received 13 August 2009)The aim of this study was to determine the optimal parameters of cutting operations in order to obtain a minimisedcost per unit time, with an acceptable quality level The cost per unit for cutting operations is partly related tocutting time and tool life The acceptable quality level is measured mainly in regard to surface roughness The cuttingtime, tool life and surface roughness are all functions of cutting parameters For cost and quality optimisation, incutting parameter determination, an integration of three elements in problem formulation is necessary However,with recent developments in computer hardware and software, the computer experimental approach consisting ofboth a computer simulation and a statistical method has become possible With removal geometry configurationbeing specified, the computer simulation is performed by existing computer-aided engineering software, such asCATIA and Computer Cutting Data Service The simulated outputs are analysed by statistical methods, such asresponse surface methodology, to obtain the predicted cutting time and tool life functions These two functions will

be further plugged into a unit cost model as the objective function, along with quality level as constraints Then, theproblem is formulated with mathematical programming to determine optimal cutting parameters for both qualityand cost

Keywords: computer experiment; cost; cutting parameter design; cutting time; optimisation; quality; RSM; tool life

1 Introduction

In today’s market, quality and productivity are

essential factors for achieving success In most areas

of industrial production, cutting tools play a major

role in the production of goods, with regard to

economy and quality A properly designed machining

process can significantly affect overall production costs

and quality levels To minimise the cost of work-piece

machining, cutting parameters must permit the

reduc-tion of producreduc-tion time and cost to the lowest levels

(Dewes and Aspinwall 1997, Choudhury and Appa

Rao 1999, Lee and Tarng 2000, Juan et al 2003,

Bouzid 2005) In the literature review, several cost and

optimisation studies on determining cutting

para-meters have been studied Several previous modelling

approaches include: 1) Taguchi method (Youssef et al

1994, Lin 2002, Singh et al 2002, Shaji and

Radhakrisnan 2003, Manna and Bhattacharyya 2004,

Oktem et al 2006); 2) response surface methodology

(RSM) (Lee et al 1996, Fuh and Chang 1997, El-Axir

2002); 3) mathematical programming, dynamic

pro-gramming or usual optimisation techniques (Gupta

et al 1995, Tan and Creese 1995, Lee and Tarng 2000,

Hui et al 2001, Liang et al 2001, Bouzid 2005); 4)

heuristic search (Lee and Shin 2000); 5) genetic

algorithm (Chen and Tsai 1996, Liu and Wang 1999,Onwubolu and Kumalo 2001, Krimpenis andVosniakos 2002, Wang et al 2002, Cus and Balic

2003, Savas and OZay 2008); 6) simulated annealing(Chen and Tsai 1996, Chen and Su 1998, Youssef et al

2001, Juan et al 2003) However, deficiencies exist inthese previous works; the analytical functions orprediction functions describing the relationships be-tween the response values of interest and the cuttingparameters of processes are usually unknown Thesefunctions can be estimated via statistical regressioneither from historical or experimental data Never-theless, the historical data may not be available and theexperimental data may not be cost effective forperforming a set of experimentation The developedrules derived from specialists’ knowledge and experi-ences do not easily lend themselves to mitigating theuncertainty of the cutting process The rules in seekingacceptable solutions differ for various cases Because theconvergence speed may be slow, the approaches mayrequire considerable execution time to reach near-optimal solutions The near-optimal solution found isnot guaranteed to be identical for each attemptedsolution search The previous works considered costand quality aspects as independent issues for cutting

*Corresponding author Email: akjeang@fcu.edu.tw

Vol 23, No 4, April 2010, 325–340

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2010 Taylor & Francis

DOI: 10.1080/09511921003667680

parameter optimisation In other words, none of the

previous works considered the concurrent optimisation

among cutting time, tool life and surface roughness for

economic and quality objectives

To determine the optimal cutting parameters,

prediction functions have to be formulated to correlate

the cutting parameters with cutting time and tool life

However, it is acknowledged that the prediction

functions are not easy or economical to obtain either

from known analytical equations or from physical

experimental trials preceding the cutting operation

being actualised Because there are no universal

predic-tion funcpredic-tions in representing cutting time and tool life

phenomena, the level of difficulty in obtaining

predic-tion funcpredic-tions becomes even higher when considering

the various and complex geometrical configurations of

work-pieces that need to be processed in a shop If the

difficulty in deriving the prediction functions can be

overcome, formulating and optimising the cost model

with acceptable quality levels for cutting parameters

determination would become possible In this regard, an

alternative approach, such as robust design via an

integration of computer experiment, statistical method

and optimisation technique, is introduced in this paper

One of the robust designs is the experimental design

approach Experimental design methods play a major

role in engineering design activities, during which new

products or processes are developed and existing ones

improved These methods have broad applications in

many industries The purpose of this experimental

approach is to determine which factors significantly

affect the quality, functionality and cost of a product

or manufacturing process This experimental approach

may lead to the development of designs with enhanced

quality, low cost and shorter design and development

cycles With this method, an exact functional

relation-ship between inputs of cutting parameters and outputs

of cutting time and tool life can be obtained from

statistical analysis This function will be formulated as

an objective function in mathematical programming

for further optimisation

Conceptually, physical experiments may become

impractical tasks in certain situations, particularly at

the beginning of the design stage With the recent

developments in computer-aided design (CAD)

soft-ware, design engineers can study design problems

without needing to know the functional relationships

between input and output in advance That is the

reason why many designs now routinely proceed with

the aid of computer experiments The other reasons for

replacing physical experiments with computer

experi-ments are to reduce the cost of experimentation and,

perhaps more importantly, speed up design activity

(Welch et al 1990, Jeang 2008, Jeang and Liu 1999,

Jeang et al 2002, 2008)

This study intends to determine the optimalparameters of cutting operations to minimise the costper unit time Cutting time and tool life functions arethe two functional relationships of greatest interesthere However, most of the time, these cost-relatedfunctions are unknown during the early stages ofcutting operations With the assistance of CAD soft-ware, these functions can be determined by regressingthe output responses, cutting time and tool life,generated from CATIA and Computer Cutting DataService (CCS) vs inputs, cutting parameters, respec-tively, with assistance from the statistical method One

of the statistical methods in design of experiment,RSM, is employed to study the response value forcomputer experimental models in order to determinethe cutting time and tool life functions These twofunctions will be further plugged into a unit cost model

as objective functions, along with required quality level

as a constraint The quality level in this study refers tosurface roughness Then, the problem is formulated inthe form of mathematical programming to determinethe optimal cutting parameters Additionally, a sensi-tivity analysis is performed under various surfaceroughness conditions for quality consideration As aresult, a robust and optimal parameter determinationfor cost and quality can be achieved for any specifiedgeometric configuration before cutting operations arerealised

This paper is divided into the following sections:section 1 provides an introduction; section 2 containsthe background related to the present research; section

3 presents unit cost estimation for cutting parameterdetermination; section 4 provides an application todemonstrate the proposed approach; section 5 is adiscussion; and finally a summary is given in section 6

2 BackgroundBefore developing the model, it is necessary tointroduce relevant background information regardingcomputer experimentation for cutting time and toollife prediction functions, as well as problem optimisa-tion in using mathematical programming

2.1 Computer experiment for prediction functionsUnder certain situations, historical data may beunavailable and physical experiments become imprac-tical or expensive tasks, particularly at the early stage

of product design and process planning With therecent advances in CAD, which involves analysis, that

is, the developed computer-aided engineering software,

it has become possible for computer experiments toreplace physical experiments Therefore, the reduction

in the cost of experimentation and, perhaps more

importantly, the speeding up of product development

or processing of design are realised Thus, a computer

experiment is used through computer software,

CA-TIA and CCS, to generate experimental data, thereby

replacing physical or historical data in finding

predic-tion funcpredic-tions for problem formulapredic-tion

To properly analyse a cutting problem, functional

relationships between the dependent variables (cutting

time and tool life) and independent variables

(cutting parameters) should be identified beforehand

Cutting time and tool life functions may appear in any

form However, the related functions are usually

unknown, difficult to obtain or exist in very complex

forms Fortunately, with the recent developments and

expanded power in computers and software, most of

the tasks are workable In this study, the CATIA V5R8

and CCS are employed to predict the presented cutting

time and tool life functions, respectively The CATIA

software has the capability to deliver the unified

product lifecycle management (PLM) solutions that

enable manufacturing industries to plan, detail,

simu-late and optimise their machining activities in order to

build better products The CCS software is an

information system related to tapping and thread

forming with the standard tools of Prototype-Werke

GmbH made from Headspace Sorptive Extraction

(HSSE) and solid carbide The system can provide tool

recommendation and the corresponding cutting and

power data after having entered the data of the cutting

parameters

A set of experimental data, cutting time and tool

life, are generated via CATIA and CCS software

without knowing the functions related to the cutting

operation The generated data are then analysed via

statistical methods, such as RSM analysis, to obtain

cutting time and tool life functions These functions are

considered as prediction functions and will be

em-ployed for optimising further analysis The ANOVA

tables will also be obtained after RSM analysis These

tables are used for ranking the importance of cutting

parameters for possible design improvements

2.2 Cutting time and tool life

In past years, the metal cutting of different materials

has caused problems in regard to manufacturing

technology The cutting process involves removing

the redundant metal to create the geometrical solid

model of the part (Paul et al 1988) One key property

of interest related to the cost is cutting time The

machining technological parameters that affect the

cutting time include feed, cutting speed, depth of cut

and percentage of tool diameter Previous works in

predicting the cutting time are mainly based on the

known analytical functions or through physical

experiments However, there is difficulty in ing these functions in most practical applications,particularly considering the various geometrical solidmodels needed in machining In this regard, theCATIA V5R8 (CATIA 1999) is employed to predictthe presented cutting time

determin-Another property related to cost is tool life In mostareas of industrial production, machine tools play amajor role in the production of quality goods Inrelation to the cost, tools might also make up asignificant portion of the expense of producing parts.Furthermore, increased worldwide competition, con-tinuous advances in automation and computer-aidedmanufacturing all highlight the importance of toolmanagement Therefore, a proper tool managementpolicy is necessary to improve part quality and reduceoverall production costs One of the most importantaspects of tool management is dealing with possibletool failure F.W Taylor (Dos Santos et al 1999) pro-vided his famous ‘Taylor tool life equation’, whereintool life was related to the feed rate, the cutting speedand the depth of cut In the present study, the toolinformation system, CCS software, will be used to esti-mate the tool life The hardness of the work material,the parameters of the feed rate, the cutting speed andthe depth of cut are input into the CCS software

2.3 An integration of computer experiment and RSMBefore performing computer experiments, an appro-priate experimental design matrix of input factors(controllable variables) must be chosen The example isshown in Appendix A and Appendix B RSM involvesthe application of a statistically designed ‘experimen-tal’ matrix in N-dimensional space, to approximate theresponse of output variables as a function of N inputfactors The matrix consists of ‘E’ trials (or runs),where E depends on the number of factors and thechoice of the experimental design This study uses aBox-Behnken design because it allows efficient estima-tion of the first-order and second-order coefficients(Montgomery 1991, Myers and Montgomery 1995).For cutting analysis, the input factors consist ofparameters X1 as the feed rate (mm/rev), X2 as thecutting speed (m/min), X3as the depth of cut (mm) and

X4as the percentage of tool diameter For example, thetotal number of trials (or runs), E, are 27 and 15 forAppendix A and B, respectively The associated low,middle and high levels of input factors Xi should bedecided before performing computer experiments.Various level combinations are considered as inputsfor computer experiments through the CATIA or CCSsoftware Afterward, the simulated outputs Y, whichare experimental data, are obtained The set ofoutputs, Y, will be immediately considered as response

Figure 1 Computer experiment via CATIA and response

surface methodology (RSM)

Data Service (CCS) and response surface methodology

(RSM)

values for the next statistical analysis in finding: a)

cutting time and tool life prediction functions; b)

optimal cutting parameter values; c) order the F-ratio

of cutting parameters based on ANOVA examination

in RSM analysis The order indicates the priority for

improvement The flow for integration of computer

experiments and RSM is shown in Figures 1 and 2

2.4 Problem formulation for optimisation via

mathematical programming

Mathematical programming can represent one

pro-blem formulation that generalises all deterministic

operations research techniques (Arora 1989) The

problem formulation is presented as:

is an objective function of minor concern, which issubject to restriction between L and U Thus, one hasthe most important objective function as the objectivefunction, while the remaining objective functions serve

as constraints Then, with the limits of the constraintsbeing varied, the consequent optimum decisions can beobtained In this way, the visible objective values areessential to make a clear choice among the alternatives(Arora 1989) The presented formulation also includesconstraints; Xj should fall between the specified lowlimit aj and upper limit bj Normally, the aboveproblem formulation can be solved by GAMS softwarewithout any difficulty (Brooke et al 1998)

3 Unit cost estimation for determining cuttingparameters

Liu et al (2007) proposed a framework for Product cycle Cost Estimation System (PLES), for Product Life-cycle Cost (LCC) estimation, by combining and applyingABC technique and machine learning (ML) techniquebased on information available The system consists ofthe following components: product life cycle database;LCC template manager; ABC module; ML module; syn-thesiser Liu et al.’s system framework of PLES for LCCdevelopment is followed to form the unit cost functions.The twin objectives of management in the cuttingoperation are to minimise cutting time and maximisetool life Due to the dependency between cutting timeand tool life, it is essential to incorporate both of theminto one objective function to ensure that a trulyoptimal solution will be obtained The unit cost of acutting operation consists of four main categories: rawmaterial cost; set-up cost; cutting cost; tool manage-ment cost The set-up cost contains the labour cost andidle cost in set-up duration The cutting cost includesthe labour cost and machining cost for the cutting timespan The tool management cost covers the toolchange cost, tool cost and idle cost This study assumesthat the idle cost in unit time can be estimated asmachining cost per min Thus, the exact function formfor estimating the unit cost is:

where Cu¼ unit cost ($); Cmat¼ cost of raw material

per part; C1¼ labour cost ($/min); C0¼ unit cost for

manufacturing expense ($/min); Ct¼ cost of a cutting

tool ($);ts¼ set-up time; ttc¼ tool changing time;

CT¼ cutting time equation; TL ¼ tool life equation

The ABC module makes it possible to clearly

identify the main resources or lifecycle activities of

interest In this study, resources include Cmat, C1, C0,

Ct, ts, ttc, which are available for ABC cost estimation

The cutting time and tool life can be obtained by input

controllable variables, which are cutting parameters

and geometric parameters However, cutting time and

tool life functions are required for formulating the

objective function for cost minimisation, in order to

determine optimal cutting parameters However, this

function is not available at present in ML modules

Thus, it becomes necessary to construct the functional

link by using ML techniques based on historical data

Due to the lack of historical data for cutting time and

tool life, computer experimental data are generated to

replace historical data via computer software, CATIA

and CCS With computer experimental data provided,

the cutting time prediction function and tool life

prediction function can be estimated by statistical

method; in this case, RSM

Cost of manufacture has rarely been used in

optimisation since it is difficult to model cost in terms

of controllable variables An optimiser in the DATUM

project was used to search for the least while still

satisfying the constraints (Scanlan et al 2006) The

problem formulation presented in Equation (4) is

minimised with quality levels and cutting parameters

that fall within a feasible range Instead of repeating

for a certain number of iterations in seeking optimum

solutions in the DATUM project, the optimum values

are reached for one shot by GAMS in this study The

efficiency is due to two reasons: one is problems

formulated with mathematical programming and

solved by optimum technique GAMS; the other is to

integrate both cutting time and tool life into one

objective function to ensure that a truly optimal

solution will be obtained The flow of statistical

analysis and cost optimisation model for design

improvement is given in Figure 3 With the computer

software CATIA and CCS, problem formulation with

optimum technique GAMS can be built into the cost

optimisation process model by following the flow

introduced in the DATUM project

4 Application

The application is concerned with the minimisation of

production cost of end-milling operations for the

geometric solid part shown in Figure 4 The

geome-trical configurations of the part body are

200 6 200 6 160 (mm3) The tool geometry includes

a normal diameter of 32 mm, an overall length of

112 mm, a measuring cutting length of 32 mm and abody diameter of 32 mm The technology of the tool

Service; RSM¼ response surface methodology

removed volume¼ 999999 mm3

includes six flutes, right hand rotation is used, the

quality is rough machining and tooth material is a

high speed steel coated with Titanium CarboNitride

(TiCN) The tool path style is inward helical The cost

rate for cutting operations is given as: the unit cost

Cmat¼ $0.5; the material cost C1¼ $0.45; the unit

cost for manufacturing expense C0¼ $1.45; the labour

cost for per min Ct¼ $180 The set-up time is ts ¼ 2.0

min and the tool change time ttc¼ 0.5 min The

CATIA and CCS was used to obtain simulated cutting

time and tool life, respectively They will be discussed

in the order of cutting time first, followed by tool life

The parameters influencing the cutting time include

the feed rate, the depth of cut and the percentage of

tool diameter for a given part In order to make these

parameters readable, the geometrical solid model of a

part will be defined in part design module before

simulating the cutting time via the CATIA software

For the sake of convenience, these parameters are

called controllable factors in the following analysis

The simulated values associated with each controllable

factor will be input in the prismatic machining module

by following the experimental design matrix in order to

have the cutting time correspond

The feasible process capability limit of cutting

parameters, X1, X2, X3and X4, is given in Table 1 X1

is the feed rate (mm/rev), X2 is the cutting speed (m/

min), X3 is the depth of cut (mm) and X4 is the

percentage of tool diameter

The three levels of controllable factors are shown in

Table 2 The RSM experimental design matrix is

provided in Appendix A The item Xi is one of three

controllable factor levels that are selected for various

combinations according to the RSM experimental

design given in Table 3 Table 3 shows the result of

cutting time for 27 experimental runs in

correspon-dence to the matrix from Appendix A The result of

response surface analysis is given in Table 4 To

determine the relative magnitude of the effect of each

cutting parameter on the response value, cutting time,

the F values are used to rank cutting parameters in

order of importance (Myers and Montgomery 1995)

The larger the value of F, the more important the

cutting parameters are in influencing the response

value, cutting time, in a cutting operation The effect

on cutting parameters X1, X2, X3, X4is significant at a

5% confidence level, referring to Table 4 Hence, thesefour factors should be carefully monitored in regard toreducing the cutting time, with special attention tocutting parameter X3, as the depth of cut, for possibledesign improvement The optimal values of control-lable factors for the cutting time are summarised inTable 5 By referring to Table 4, the approximatingfunction of the cutting time (CT) predicted function is:

Table 2 Three levels of cutting operation for each factor

failure; these costs are related to tool life As is known,

the length of tool life is dependent on the cutting

parameters when X1is the feed rate (mm/rev), X2is the

cutting speed (m/min) and X3is the depth of cut (mm)

In this study, the tool life is obtained by CCS software

To obtain the tool life, the values associated for each

controllable cutting parameter will be input by

follow-ing the experimental design matrix provided in

Appen-dix B, to have the tool life correspond Table 6 is RSM

experimental combinations and experimental outputs,

which signify the tool life in this study The result of

Table 4 Response surface analysis for the cutting time of cutting operation

Response surface for variable time

Table 5 The optimal cutting time for cutting operation

(response surface methodology)

in order of importance From Table 7, the effect oncontrollable factors X1, X2, X3 is significant at 5%confidence level Hence, these controllable factorsshould be carefully monitored to increase the tool lifewith special attention to cutting parameter X2, thecutting speed, for possible design improvement Theoptimal values of controllable factors for the tool lifefrom RSM are summarised in Table 8 By referring toTable 7, the approximating function of the tool life (TL)predicted equation is:

As in the preceding discussion, prediction functions

(5) and (6) will be plugged into Equation (4) for problem

optimisation analysis, which is the objective function in

this example The objective function is given in

Equation (7) In addition to the objective function, the

controllable factors Xi are constrained by the limit,

which is the process capability limit falling within an

acceptable SUiand SLi SUiand SLias upper limits andlower limits, defined in Table 1 These constraints arerepresented by Equations (8)–(11) Mathematical pro-gramming can contain the above considerations in oneformula The problem is formulated as follows:

Table 6 Response surface methodology experiment design

for tool life and response data

Table 7 Response surface analysis for the tool life of cutting operation

Response surface for variable time

Result 1: It represents the optimal solutions for

cutting time except tool life and unit cost; however,

they can be computed by plugging the optimal solution

X1, X2, X3 and X4 into Equations (6) and (7)

Result 2: It represents the optimal solutions for tool

life except X4, cutting time and unit cost Assumed

X4, 27.28503925 from result 1 Similarly, both cutting

time and unit cost can be found by plugging the

optimal solution, X1, X2, X3, and X4, into

Equations (5) and (7)

Result 3: It represents the optimal solutions for thepresented approach, which is composed of Equations(7), (8), (9), (10) and (11)

Clearly, the presented approach (result 3) has theminimum cost among the three approaches Thereason is that results 1 and 2 only provide the localoptimal solutions for cutting time and tool life,respectively However, the presented approach (result3) has the cutting time and tool life combined in oneobjective function

F-ratios and important rankings for cutting time andtool life are summarised in Table 11 The combinedimportant rankings in Table 11 clearly reveal that theorder of importance is parameters X1, X3, X2and X4.Apparently, the critical parameters in optimisation ofcutting time and tool life are X1 and X3 Unlike theconventional approach, the presented critical path isjustified statistically with important rankings and jointlywith multiple objectives in the presented approach Ifimprovement for cutting operations is needed, theseparameters should be closely monitored and consideredhigh priority for enhancement Consequently, a globaland robust optimal solution can be reached in thepresented approach

The above approaches mainly focus on the ical cost consideration However, it is possible thatquality consideration is also an important aspect inprocess planning In this regard, a constraint to ensure aminimum quality level being satisfied has to be added tothe above formulation One quality measurement incutting process is the magnitude of surface roughness.The following equation represents the acceptablequality level that should be no more than the maximumsurface roughness (Rmax) for the quality requirement

Table 8 The optimal tool life for cutting operation

CT ¼ cutting time; TL ¼ tool life.

Table 10 Comparison among three results

Result 1

(CT-min)

Result 2(TL-max)

Result 3(cost-min)

CT ¼ cutting time; TL ¼ tool life.

Table 11 The F-ratio and important ranking for cutting time (CT) and tool life (TL)

Factor

importantranking (3)(3)¼ (1) þ (2)F-ratio

Important

Importantranking (2)

an additional cost to achieve better quality Hence, a

balance between cost and quality consideration is

necessitated for cutting parameter determination

Furthermore, tool life increases with the decrease

of Rmax A similar conclusion is still held for cutting

time estimation These facts are explicable because the

decrease of Rmax will lead to a small feed rate X1,

which eventually results in a longer tool life and

cutting time, Although, in theory, the depth X3

increases with the decrease of Rmax and should result

in a shorter tool life and cutting time, opposite

conclusions are obtained due to the fact that feed

rate X1plays a dominate role in tool life extension in

comparison to depth X3 However, when Rmax is

greater than 0.00001083, the optimal solutions become

insensitive to the varying surface roughness The

insensitivity is attributed to the fact that the associated

X1 for Rmax in exceeding 0.00001083 is always

0.073574 mm under the presented problem

formula-tion from Equaformula-tions (7)–(12)

Thus, with the preceding discussion and combined

importance ranking, the parameter X1 is the most

important concern in the cutting operation for an

economical and quality process parameter design

5 Discussion

Conventionally, the process design for the cutting

operation is based on engineering experience, standard

data or trial and error runs to find the cutting

parameters These approaches cannot guarantee that

the found cutting parameter will actually be the best one

among all the possible choices The other approaches for

finding the optimal cutting parameters are based on the

existing analytical function, which is derived from the

known geometrical model However, the former have

their defects because it is difficult or impossible to obtain

a solution when the analytical function is in a

complicated form or when the analytical function does

not exist Normally, there are quite a few alternatives

during the design stage; that is, a set of analytical

functions must be determined before optimisation

tech-nique can be applied In addition, optimisation

techni-ques cannot show the priority of each input statistically

With their ever-expanding power, personal ters have become useful tools for design activities Inthis regard, the recent development in CAD softwareintegrates the engineering and computer functions sothat design activities can be accomplished efficiently.CATIA and CCS CAD software for cutting designwere used in this study During the process designphase, there are many prototypes to be considered andthe designer is faced with the task of choosing fromamong these alternatives Today’s CAD softwarerarely provides the quantitative measurement ofquality and cost for each candidate selection Inaddition, the information drawn from CAD is usuallyconducted under the assumption that each input isthoroughly studied individually, when remaininginputs are fixed This approach is ineffective because

compu-a pcompu-airwise compcompu-arison is needed for ecompu-ach input untileach input has been evaluated Furthermore, thisapproach does not tell the designer how the effect of

an input changes when other inputs change due tointeractions Therefore, the present study integratesboth the available CAD software and statisticalmethod of RSM to extend CAD capability by takinginto account the quality and cost aspects The presentapproach can investigate all of the possible inputcombinations based on RSM analysis to maximise themost favourable result, particularly when interactionexists among design inputs

Decision making during process planning is adynamic and evolutionary process, which involvesadding or deleting design criteria, changing theassociated criteria value and adjusting preferencepriority for relevant criteria To maximise the effec-tiveness of the design process, it is necessary to provide

a method that enables designers to obtain feedbackand direction for design improvements Statisticalanalysis from RSM used in ranking importance order

is adopted for this purpose These features are veryimportant during design activities

In employing the present approach, two functionshave to be known in advance: cutting time and tool lifeprediction functions The cutting time predictionfunction has the function of parameters, such as feedrate, spindle speed, depth of cut and percentage of tool

Table 12 Comparison among three results

diameter The tool life prediction function involves the

first three parameters shown above However, the

function describing the relationship between the input

parameters of interest and the output responses may or

may not be available during the early stage of process

design Hence, an effective method to provide these

functions becomes critically important for a successful

cutting process design, particularly since geometrical

configurations needing to be cut are quite different in

each task One possible way is to collect a set of

experimental data generated from computer software

CATIA and CCS and run the statistical analysis,

RSM, to create these functions

In preparing to use the present approach, the

following important steps should be followed:

Step 1: Provide removal geometrical configuration for

the part design module of CATIA software

Step 2: Choose an appropriate design experimental

matrix over the various levels of inputs for

performing the computer simulation (see

example in Appendix A) These levels are

fractions of the inputs range Inputs for

cutting time estimation are feed rate, spindle

speed, depth of cut and percentage of tool

diameter Inputs for tool life estimation

involve the first three parameters shown above

(see example in Table 2)

Step 3: Under various levels of inputs, according to the

arrangement of the design experimental matrix,

run the machining module with the given

geometrical configuration to obtain the cutting

time via CATIA software Similarly, tool life is

estimated via CCS software (see example in

Tables 3 and 6)

Step 4: Use RSM to find the cutting time and tool life

prediction functions by regression analysis (see

examples in Equations (5) and (6)) Then,

employ ANOVA to determine the important

parameters (see example in Tables 4, 7 and 11)

Step 5: Obtain the best values of cutting parameters

for minimisation of total cost for unit time via

mathematical programming (see examples in

Equations (7)–(11)) Additionally, a sensitivity

analysis is performed under various surface

roughness conditions for quality

considera-tion (see example in Table 12)

Step 6: If improvement is needed, then the above steps

can be repeated based on the suggestions at

step 4

Consequently, a robust and optimal parameter

determination by statistical method and optimisation

techniques can be achieved, thereby reducing

opera-tion problems

For the purpose of comparison, Table 13 provides

a list of approaches representing various modelling andoptimisation techniques Although each approach hasits usefulness, there is no factor that is decisive on itsown, in regard to selecting the best one exclusivelyamong alternative approaches in metal cutting pro-blems The imperative criteria, which must be con-sidered for developmental approaches in futureapplications, include:

(a) A robust design via statistical optimisation isessential for an uncertain design environment.Thus, the incorporation of RSM and optimisa-tion techniques as introduced in this study wasconstructed The optimal cutting parametersassociated with the ranking of importance willassist the designers regarding decision making

in process planning

(b) Physical experimentation should be replacedwith computer experiment for economic andefficiency considerations In other words, theprocess design can be completed at the blue-print stage with shortened duration and les-sened cost

(c) The approach should be effective in ing the exact optimal cutting parameters Forthese achievements, the fidelity integration ofCAD, PLM and analysis techniques reachcrucial importance Thus, this study usesCATIA and CCS to generate experimentaldata, which are analysed by RSM in discover-ing prediction functions The objective function

determin-is formulated precdetermin-isely with prediction tions and ABC cost estimation Then, theproblem is optimised by GAMS software.(d) The optimal cutting parameters should bedetermined in the simultaneous consideration

func-of cost and quality Thus, the required qualitylevel, surface roughness, is set as a constraint inthe presented problem formulation

Based on the optimal cutting parameters, theestimated cutting time and tool life functions wereobtained in the early stage of production planning Theproduction scheduling decisions in allocation andmaintenance of the available capacity or resourcesover time become possible This possibility results in ahigh utilisation of labour, equipment and space withlow inventories and good customer service for aproduction system However, there is a limitation forapplicant usage during industrial exercises The limita-tion mainly comes from the assumption of Equation(12) The equation represents the acceptable surfaceroughness, Rmax, for the quality requirement with theassumption that the effects of cutting speed and depth