Nghiên cứu sữa part 6

However, integration of expert systems with a process control system is complicated by the real-time interaction requirements.. To build intelligent applications of scheduling, simulatio

3 The expertise can be made more widely available.

4 The expert's time can be saved Routine decisions delegated to an expert systemallow the human expert to concentrate on the more abstract, currently developingproblems

5 It takes years for human experts to learn their specific skills Expert systems can

be copied on magnetic media in seconds or minutes

6 Humans become sick, retire, and die Expert systems continue to work sistently and predictably

con-7 Human expertise is expensive

8 Savings can be realized on maintaining and updating the knowledge base ware maintenance is a large cost of software systems over their lifetime Much

Soft-of the programming time is spent in finding and adjusting for modification sideeffects on programs The emphasis on structured analysis and programming tech-niques in software development because of deadline pressures detracts from thegoal of a more structured approach The result is that a less than ideal code isgenerated The clear distinction of facts, heuristics, and inferencing knowledge

in knowledge-based systems reduces maintenance and update costs becauseeffects of changes are restricted

Disadvantages of expert systems that can be considered are as follows:

1 When considering cognitive activities to other human tasks, expert systems aregood at extracted, cognitive, logical thinking They are not well suited for man-aging highly sophisticated sensory input or mechanical motor output

2 Expert systems exhibit a narrow band of simulated intelligence based on a narrowrange of codified, heuristic knowledge They do not respond well to situationsoutside their range of expertise

3 Expert systems are weak in common-sense knowledge Human reasoning usesassociations, which may not be appreciated or even realized when developingthe knowledge base These associations and thought processes are based on arange of contextual information including social surroundings, random memories,feelings, emotions, and other nonrational information Even more difficult tocapture is human intuition In this case, humans draw spontaneously from theirsubconscious of creativity and insight If a decision maker goes by hunch morethan by facts or logical arguments, the problem is not appropriate for an expertsystem

4 It is difficult for an expert system to learn unless through the human user orknowledge engineer

3.3*2 Knowledge Acquisition

Expert systems rely greatly on knowledge In order to obtain quality knowledge forimplementation of a successful expert system, the following elements should beconsidered: application selection, domain expert selection, knowledge engineer

selection, tool selection, knowledge acquisition, and system development anddeployment.

A useful technique in developing a list of potential expert system applications is

to observe any knowledge bottlenecks These can be recognized as people waitingfor others to m&ke decisions and people stopping to search for needed information.Typically, small expert systems are used as intelligent job aids Most expert systems

in use today are of this size They accomplish small procedural or diagnostic tions such as equipment maintenance or product defect analysis An expert systemcan be selected based on its ability to assist, accelerate, or improve the quality ofdecision-making in groups already using computers

func-Midsize expert systems are designed to be installed on mainframe computers Inaddition to the functions of the small systems, most midsize applications includediagnostics and large configuration and scheduling systems Large expert systemshave traditionally been developed in LISP and on LISP machines They capturelarge amounts of expertise and make it widely available through the organization.Valuable heuristic knowledge that might be lost through retirement is often anappropriate candidate for an expert system application On-line operators could takeadvantage of expert knowledge when the human expert is not available However,the task should be well defined and narrow

The domain is the area of knowledge or expertise being captured A knowledgedomain expert will typically have 10 or more years of experience If possible, anindividual should be sought who can analyze a problem and explain his or herthinking in specific terms He or she should have the ability to analyze problemssystematically A good working environment needs to be created In addition, thedomain expert needs to be both knowledgeable and cooperative

The role of the knowledge engineer is to research existing knowledge and to helpthe expert describe his or her own problem-solving procedures The knowledgeengineer needs to understand symbolic programming techniques Interpersonal skillsare also a necessary attribute

The expert system tool should be selected in terms of the type of problem to besolved Types of systems include diagnostic, training, and decision support Morespecific information of tool selection will be covered in Section 3.3

The knowledge acquisition process involves gathering recorded relevant mation and preparing it for entry into the computer Knowledge can be represented

infor-in several ways infor-includinfor-ing frames and rules

Initially a prototype is constructed to demonstrate the basic operations ments can then improve accuracy, completeness, and user-friendliness The proto-type is then field tested to verify its accuracy and usefulness Users can documentall of their decisions in the domain area for a specified time period The periodshould be long enough to include multiple occurrences of most events in the system.When the system is operating in an on-line situation, it should be taken off-line andpresented with the same problems The answers should be in agreement

Refine-An important factor in successful implementation of the system is user training.Users' concerns and needs should be anticipated Continued support should be avail-able as new knowledge accrues and new needs develop

3.3.3 Tool Selection

The term tools refers to expert system software Shells are developmental tools thatprovide users with a framework within which to build their knowledge base Thealternative is to use a symbolic language such as LISP First-generation tools werewritten in LISP, but eventually tools were designed to integrate with existing hard-ware and software Languages are more flexible than shells, but more difficult andtime consuming In order to match the tool to the problem, it is necessary to knowproblem requirements and the types of knowledge representation and inferencingstrategies a particular tool does well

In selecting an appropriate tool, both the shell and the interfaces need to beconsidered Although the shell may allow the desired knowledge development, if it

is unable to interface with the process and existing software, its practical applicationwill be limited Debugging aids and technical support are additional factors.Expert systems can be categorized into four classes: specialized tools, smallertools, middle-range tools, and sophisticated tools.5 An example of a specialized tool

is PICON [LISP Machines, Inc., (LMI), Andover, MA, U.S.A.] PICON is used inprocess control equipment and in complex applications such as weather forecastingand financial market monitoring Examples of smaller tools are VP Expert (Paper-back Software, Berkeley, CA, U.S.A.) and Magellen (Emerald Intelligence, AnnArbor, MI, U.S.A.) Generally, these tools use a single representation scheme andare designed for lower priced microcomputers Middle-range tools include productssuch as GURU (Micro Data Base Systems, Inc., Lafayette, IN, U.S.A.) and KEW

& KESII (Software Architecture and Engineering, Arlington, VA, U.S.A.) ticated tools such as ART (Inference Corporation, Los Angeles, CA, U.S.A.)and KEE (Intellicorp, Mountain View, CA, U.S.A.) employ multiple knowledge-representation schemes and advanced graphical display mechanisms

Sophis-The basic design of expert system shells centers around the representation structures and inference mechanisms The basic schemes are If/Thenproduction rules, frames and networks, objects, and access triggers or demons Mostproducts represent knowledge using If/Then production rules as the primary scheme.Goal-directed, backward chaining is the simplest form of If/Then production rulerepresentation and inference mechanism, followed by forward chaining combinedwith backward chaining Some systems support only one method at a time whereasothers support a mixed mode The number of premise conditions allowed for eachrule is sometimes limited To avoid complicated logic or decision trees, modulardevelopment capability is required This may be referred to as rule sets, knowledgemodules, sections, state objects, or frames Other factors to consider in tool selectionare support of uncertainty, rule sequencing procedure, user interface, help facilities,knowledge acquisition features, access to other programming facilities, capacity andresponse time, hardware requirements, pricing, and vendor support

knowledge-The demands on a well utilized expert system normally increase, so expandability

is important Also expert system tools consume large quantities of RAM and CPUMIPS This is particularly the case with the middle-range and higher systems

Larger systems are referred to as environments Large-scale environments provide

a wide range of development and execution utilities They have multiple ming and support capabilities Some of the features include utilities for design ofcustom windows and mouse-sensitive menus, debugging tools, multitasking services,libraries of routines, and mathematical and statistical functions

program-3.4 Expert Systems and Process Control

Before discussing the application of expert systems to process control, a description

of traditional computer process control is useful This type of control is in widespreaduse and expert systems generally build on or integrate with traditional systems.Athough promoted expert systems are being applied to an increasing number ofmanufacturing processes, reports of applications actually in use are few Problemsthat have hindered their progress will also be discussed in the following sections

3.4.1 Preexpert System Developments

Although the study of dairy processing often centers around specific classes of dairyproducts, there are often common operations involved such as cooling, heating, ormixing A systematic approach to the study of dairy processing is to examine thesecommon or unit operations A unit operation accomplishes some specified function

on a product such as heating, cooling, pumping, mixing, evaporating, dehydrating,separating, and cleaning

The control of process parameters such as temperature and pressure was originallymaintained normally by human operators as they observed gauges and sight-glasses

In the 1940s pneumatic instrumentation was developed, which was able to senseprocess parameters and provide feedback control Signal transmission to a remotecontrol room was possible The control equipment usually remained with the piece

of equipment or was routed to a control panel Pneumatic instrumentation is stillused in plant control equipment Its longevity is due in part to its high reliability.These systems are based on single-loop control throughout a plant Single loopsinvolve one measurement, one control algorithm, one actuator, and one processvariable For instance, the function may be to maintain a temperature or a particularflow rate Feedback control is observed based on the difference between the meas-urement and a specified setting or set-point

In the 1960s digital computer usage began in process control Analog tation became available, and electronic instrumentation became more stable andrepeatable than comparable pneumatic controllers Signals could be transmitted overlonger distances With traditional control, operating conditions are predetermined tomaintain certain levels or temperatures Programmable logic controllers (PLCs) re-placed many of the discrete relays Computer-based control systems were much moreflexible than traditional hard-wire, relay-logic control systems

instrumen-Widespread adoption of the technology was slow because of concern over sive problems due to a single computer failure By taking advantage of PLCs indistributing computing over a wider area, concerns over large-scale failures werealleviated.

mas-Digital-controlled systems are widely used control mechanisms in use today.Once computer-based control systems are interfaced with sensors, valves, andswitches, the measurement devices can provide input for the computer while signalsfrom the computer provide input for the control device A key feature in this system

is the ease with which process changes can be made by reprogramming the computer.Computer-based process control begins with the monitoring and control of unitoperations These include blending, pumping, heating, and storage operations as well

as clean-in-place (CIP) operations

Blending or formulation can be considered as a unit operation For instance,computers can be used in standardizing cheese milk to give the correct casein/fatratio Yield prediction formulas can be used in the calculations The 44ASEA MasterBatch" control system for process control in ice cream manufacturing uses programscomprised of modules from an integrated software library.6 Batch movement andformulation are fully automated Quantities of raw materials and finished productare tracked and reported

Blending operations using least cost formulation and computer-aided optimizationare well established in the food industry Blending operations must also considerlegal and sensory requirements

Metering of milk and other ingredients is often computer controlled The casein/fat ratio for cheesemilk can be determined on-line using an infrared multicomponentanalyzer The speed of the :ream meter is adjusted as needed Pumping systems havebeen designed to avoid problems associated with centrifugal pump cavitation Thesystems include a PLC, a sound sensor, and a variable speed drive at the pumpmotor The sensor can detect preliminary cavitation impulses, signaling the variablespeed drive to adjust the pump's revolutions.7

During pasteurization, a computer can monitor all of the parameters of the ess These include temperatures, valve positions, liquid levels, and feed rates Somecontrol systems are limited to data acquisition without real-time control ability Inthese cases, process control adjustments are still dependent on the operator A retortmanagement system has been developed (TechniCAL, Inc., Metairie, LA, U.S.A.)that provides temperature and pressure control The system monitors these signalsalong with all facets of retorting including cooking, venting, and cooling If a tem-perature deviation occurs, the system automatically recalculates a new process timeand makes all necessary adjustments USDA and FDA officials were involved indeveloping and implementing the system to ensure regulatory compliance

proc-Controllers and recording devices for dairy pasteurizers must also meet federaland state health codes Instrumentation for these applications includes differentialpressure controller and single- and dual-point diversion recorder/controllers Thepressure controllers measure and indicate pressures at the raw product inlet of a high-temperature-short time (HTST) pasteurizing regenerator The diversion recorder/controllers specify control temperature levels

Liquid levels in tanks can be monitored simultaneously The data can be used inproduction tracking and inventory control Enclosed cheese vats are well suited forautomated control Ingredient addition, temperature control, cutting, stirring, drain-ing, and washing are commonly computer controlled.

CIP systems have long been established as major computer process control tems in the dairy industry Recent developments include computer-controlled deter-gent dosing systems, environmentally friendly CIP systems, automatic analyzers,and CIP data logging systems

sys-Some difficulties encountered in automation and computer utilization in the foodindustry are a lack of suitable sensors, low profit margins, use of batch/continuousoperations, and the installation of equipment that is not integrated into the wholeprocess

Beyond unit operations, these systems currently can supply data to a higher levelhost computer for further data manipulation Devices can be mixed and matched andintegrated into plantwide control schemes The use of single loop controllers involvesonly configuration without dedicated software

Configurable software is used to sequence control systems so that the processoperates as a sequential series of linked operations that can operate independently

of each other (fill tank, empty tank, sterilize, etc.) but are still related in terms oforder and integrity to process operation To design suitable control systems aroundthese concepts, all tasks must be uniquely defined and self contained and the plantmust be divided into areas of unit operations The distributed operator interface canthen use a single coaxial cable to connect modules rather than thousands of strands

of wire

For example, consider diverse processes such as manufacturing regular, flavored,and evaporated milk; storing products; and cleaning The status of the process isshown by means of matrices at the operator station Some of the linked computerscontrol manufacturing while another acts as a management computer With a com-puterized control system, the desired functions can be monitored and the systemprogrammed for the next product Computers can chart equipment conditions andlocate developing problems at any point in a process Products such as the System

30 (APV Crepaco, Inc., Chicago, IL, U.S.A.) are useful for small plants yet allowfor expansion to scan and control thousands of sensors and actuators and combinewith many different users Features of the System 30 include the ability to networkwith other systems; to communicate with a wide range of protocols including intel-ligent sensors, bar code readers, PLCs, and PC software packages; and to providefault-tolerant operation

3.4.2 Expert System Applications

Application of software-based automation to a process plant results in much lessrepetitive manual adjustment Automation requires clearly defined algorithms andthe appropriate design, installation, and maintenance of sensors, controllers, andactuators Included in the form of software is much of the operator's knowledge andexpertise in management of the process This almost suggests a sort of preexpert

system Process control software is able to monitor many operating, trend, and alarmparameters Networked with a host computer to make process information available

to appropriate personnel, the necessary information system is in place to permit line, expert-system applications

on-Placing microcomputers in plants for local solutions is useful However, tion operations require changes in programs and few qualified technicians are avail-able to make changes when problems occur Documentation is often incomplete.Expert systems can be used to take the place of experts or qualified technicians.Three types of expert systems used in process control are self-tuning controllers,control system configurations, and fault analyzers

produc-Self-tuning controllers automatically change controller settings based on controlloop performance Proportional-integral-derivative (PID) refers to a three-mode al-gorithm for this type of controller PID controllers use either pattern-recognition orfrequency filters With pattern-recognition, responses are characterized for over-shoot, damping, and period, and corrective adjustments are made Frequency filteringachieves control by forcing the outputs of two filters to conform to a given ratio.Using pattern recognition, a controller can identify a disturbance response in theloop-error signal When the loop error exceeds a specified threshold level, it is testedagainst a series of rules The information obtained is used for a tuning calculation.Another type of self-tuning controller is model based These controllers are adjustedaccording to the difference between the process response and a model used forcomparison

Control system configurators are used to connect components together to form acontrol system With many controlling devices interacting, the optimal configurationbecomes important Expert systems are able to select optimum configurations forcontrol systems The user can enter known operating conditions and process param-eters and the program will select the configuration that best satisfies the requirements.Expert systems can provide steps on configuration of the system Connectingmany I/O points with accompanying information can be accomplished more easily.The system can be regularly checked for accuracy with on-line help provided asrequired Expert system configurators can simulate data in order to test a systemprior to actual operation Therefore, errors in control or other functions can be ob-served and corrected

Fault analyzers diagnose problems as they arise and provide corrective tions to operators In serious failure conditions an expert operator may not be avail-able or may have insufficient time to respond An expert system can be used tointerpret a series or pattern of alarms quickly, resulting in a description of the faultand a recommendation for corrective action

instruc-On observing an out-of-control situation, an operator must determine the cause

of the situation and the best solution The knowledge of how one or more expertswould respond could be obtained from a properly designed expert system Theknowledge base can be continually updated by human experience and by informationthe computer gains as failures occur Several advantages the expert system has arethe depth of experience coming from several human experts and the much larger

amount of data and factors an expert system can assimilate that is available to theuser.

Personal computer (PC)-based systems can be networked to provide performancelevels equivalent to much larger machines Coprocessors are additional processorsused to speed up operations by handling some of the duties of the main processor

or CPU Coprocessor technology in the PC allows the accessing of real-time mation PLCs can transfer data directly into a monitoring PC, which can run anexpert system based on those data

infor-A user can define conditions or rules that are continuously checked If conditionsare found to be true, certain functions are performed For example, a cluster of valvesmay be checked to see if all are closed If all valves are closed, the status of thecalculated point is false If a valve is open, the status is true Another rule sends astart command to the pump if a valve is open or a stop command if all valves areclosed This rule based arrangement, sometimes referred to as an event processor,allows maximum flexibility in developing control strategies

The processing capability of computer-based controllers can be increased withthe integration of expert systems Expert systems can provide an intelligent interface

to the controlling device or sensor Many process variables can be examined andassimilated However, integration of expert systems with a process control system

is complicated by the real-time interaction requirements Not all tools are useful forthis purpose, and the language constructs become critical For these reasons incor-poration of expert systems into on-line process control has been slow to develop.Some of the problems associated with expert system integration have been addressed

by workers at Honeywell, Inc resulting in their expert systems development anddelivery environment called TDC 3000 Expert

The development of many successful expert systems has centered mainly aroundsmall-scale prototypes Some of the requirements may differ with medium and large-scale systems In the larger systems, integration with the existing data structures isrequired To build intelligent applications of scheduling, simulation, supervision,and statistical process control, integration of management information systems anddatabase technology with expert systems methodology is critical

To successfully integrate expert systems into a process control environment, theexpert system must access the real-time manufacturing data and communicate withthe human operator The inference engine instructs the data accessor Because alarge number of variables are probably being referenced, and data points are chang-ing often, access needs to be timely Otherwise the data collection can interfere withthe reasoning process and reduce the pertinence of the system's advice The inferenceengine keeps track of what data are needed and then instructs its acquisition whilethe knowledge base is inactive In providing information to the operator, the expertsystem must prioritize results as to their importance Specific declarative and pro-cedural mechanisms allow the expert system to reason about how intelligent it canafford to be and still provide a timely response Language constructs need to supportchange and trends Procedural representation should be avoided

For a knowledge base, input is needed from expert and process engineers aboutnormal plant behavior as well as problem situations What constitutes a problem

depends on the state of the process For example, what constitutes high pressureduring the cheesemaking set is different than high pressure during stirout of the curd.

A slight problem with temperature control adjustment on an enclosed cheese vatmay be less critical than an out-of-position discharge valve

3.4.3 Knowledge Representation in Process Control

Classes of objects and their attributes can be used in a knowledge base to tracknormal behavior of a plant Various classes of objects are defined by name Members

of a class have various attributes For example, a class named vats could have theattributes pressure, diameter, or phase-of-process Classes may inherit attributes orthey may be defined for them Context-sensitive rules change their own values andmake concomitant assertions about values of other variables in the system Someclasses and attributes can be used in a knowledge base to represent and track thenormal behavior of a plant For example, filling a vat covers different phases Theseall belong to the phase attribute Each context of the phase attribute provides rulesfor recognizing when a process for the vat has moved to the next phase

To track abnormal or undesirable behaviors, the expert system needs (1) tions from which problems conceivably can arise; (2) evidence that can confirm orrule out the actual existence of a problem; (3) descriptions of other problem situa-tions, if there are any that can possibly be causes of the problem; and (4) actions ofthe operator to remedy the problem If a knowledge base contains information toidentify a problem, there must be a structure to the knowledge so that the firing of

condi-a rule or frcondi-ame lecondi-ads to the condi-appliccondi-ation of other relcondi-ated knowledge Knowledge could

be represented using rules or general purpose frames, but those techniques havedisadvantages An inherent structure is needed in the knowledge base to identify thereasons for a process upset Using rules, this structure is not evident At run timethe inference engine would not know the boundaries of its own knowledge In adynamic, real-time application, a simple forward or backward chaining inferenceengine cannot distinguish between a lack of rules resulting from the temporary ab-sence of pertinent data and lack of rules because of an inadequate rule base

A second and related disadvantage to the use of general purpose knowledge resentation techniques is that a lack of knowledge structure makes selective activa-tion and deactivation of knowledge at run time very difficult

rep-With generic rule or frame representations, the different criteria for ordering vice-giving knowledge are difficult to distinguish A knowledge base written asgeneric situation-action rules requires more development and maintenance effort bythe knowledge engineer Frames can be organized into specialized structures calledsituation frames Slots are available for holding knowledge Slots connect with otherframes in knowledge base links An inference engine can traverse cause-effect struc-ture from top down and can issue operator advisories

ad-The human interface aspects of integration need to be considered Collaborationsshould take place with the operator An exhaustive search for all possible undesirableprocess situations must be performed As not all evidence can be obtained frominstrumentation, what is practical needs to be decided Redundant and trivial nuis-

ance messages, which the operator already knows, must be eliminated The operatorneeds time to observe, comprehend, and act The operator needs to access processdata in the knowledge base using terminology with which process engineers arefamiliar.

Reference needs to be made to trends in process data over various time intervals.Trend intervals and statistics can be predetermined and precomputed into the system.Historical buffers of raw data can be maintained and made available for the engi-neer's questions

Ways to interact with an operator in real-time application need to be provided

Occasionally, the system cannot get information from the data and needs to query

the operator However, an operator may not be there, and the system cannot stop Aspecial class of query objects can be treated for which instances are declared Aninstance can be when an answer attribute is needed to evaluate the expression, thelatest answer is unknown, and the corresponding question is not displayed Theevaluation then causes the text of the question to get delivered to the operator'sconsole

The main decision is to determine what kind of explanation capability should beprovided and what information operators will find useful The expert system couldexplain why the problem exists, how to perform the recommended action or how toobtain the information needed, why the action should be performed and how it willhelp, and how the conclusion was determined or the line of reasoning

3.4.4 Commercial Examples

In developing industrial touchscreen workstations, a graphic display is configuredusing "If/Then/Else" statements and English language commands (Nematron, AnnArbor, MI, U.S.A.) The G2 real-time expert system has been implemented in anumber of process control situations including chemical process control, flight mon-itoring, network management, manufacturing, simulation, training, energy manage-ment, robotics, and water treatment G2 uses object-oriented representation of plantequipment and models of process behavior Heuristic and analytical knowledge isused Heuristics are a simplification tool to reduce the search in a large problemspace To avoid time constraints, the system uses metaknowledge to focus the in-ferencing resources Metaknowledge involves rules acting on rules to reduce thesearch space The application developer can create classes of objects or import preex-isting object classes from G2 An integrated simulator allows the developer to test

an application prior to its deployment After the application is developed, tested, anddeployed on-line, G2 can communicate with control systems, PLCs, databases, orother sources of real-time data

INFI 90 (Baily Controls, Wickliffe, OH, U.S.A.) is entitled a strategic processmanagement system This process control system is able to access embedded expertsystems Referred to as EXPERT 90, the expert system is represented as a series of

"If/Then" rules that may involve time relationships as well as uncertainty data Theexpert system offers advanced advisory, analytical, and control functions such asadaptive control, alarm interpretation and management, and cause-and-effect advi-

sories These can act as guides before operating boundaries are breached Problemswith using expert systems for real-time control include separation from the processcontrol system, extensive interface hardware, and complex program creations EX-PERT 90 has addressed these by embedding the expert system within the ProcessManagement System This allows easier exchange of data between the expert systemand the control system EXPERT 90 also provides a distributed architecture facilityand a modular system building approach.

Although not an expert system itself, the Lynx-OS (Lynx Real-time Systems,Inc., Los Gatos, CA, U.S.A.) is a real-time operating system specifically designedfor process control, data acquisition, and communication Lynx-OS is multitasking,allowing users to run several control applications, expert systems, program devel-opment jobs, and maintenance tasks concurrently It accommodates an expert system

in a real-time environment by preempting it with a higher priority control task ever necessary In the food industry it has been successfully used to standardize andoptimize beverage extract quality

when-A process control system with expert system features has been incorporated into

a sugar refining operation (System 3, Rosemount, Inc., Eden Prairie, MN, U.S.A.).Referred to as a distributed process control system System 3 offers diagnostic fea-tures and allows in-house installation and configuration The step of sugar boilingrequires highly skilled and experienced operators The program for the operationalsequence of the vacuum pans is based on a flow chart constructed from the knowl-edge and experience of the expert operators One of the results was a reduction inprocessing time An expert system is being developed to measure mixture consist-ency (OpCon, Eaton Corp., Milwaukee, WI, U.S.A.) Torques on mixing shafts aremeasured These values can be combined with other process information such astemperature, pH, moisture, and color

3.5 Business and Manufacturing Operations

This section is an expansion of Section 3.4 In that section the elements of based process control were mainly applied to unit operations This section will de-scribe additional computer-based activities in the plant including information sys-tems As monitoring and control devices for unit operations have developed andhave been applied in food processing plants, information systems have also becomewell established Information from many areas of the company including receiving,inventory, scheduling, quality systems, distribution, and marketing is being managed

computer-on computer databases Part of this informaticomputer-on is being generated automatically bydata collection devices Expert systems can help manage and analyze the increasingamount of data This section will discuss efforts to integrate the manufacturing sys-tems with the information systems

3.5.1 Physical Goods Management

Physical goods management refers to inventory and distribution control One of theearly applications of computers in manufacturing plants was the tracking and dis-

Next Page

sories These can act as guides before operating boundaries are breached Problemswith using expert systems for real-time control include separation from the processcontrol system, extensive interface hardware, and complex program creations EX-PERT 90 has addressed these by embedding the expert system within the ProcessManagement System This allows easier exchange of data between the expert systemand the control system EXPERT 90 also provides a distributed architecture facilityand a modular system building approach.

Although not an expert system itself, the Lynx-OS (Lynx Real-time Systems,Inc., Los Gatos, CA, U.S.A.) is a real-time operating system specifically designedfor process control, data acquisition, and communication Lynx-OS is multitasking,allowing users to run several control applications, expert systems, program devel-opment jobs, and maintenance tasks concurrently It accommodates an expert system

in a real-time environment by preempting it with a higher priority control task ever necessary In the food industry it has been successfully used to standardize andoptimize beverage extract quality

when-A process control system with expert system features has been incorporated into

a sugar refining operation (System 3, Rosemount, Inc., Eden Prairie, MN, U.S.A.).Referred to as a distributed process control system System 3 offers diagnostic fea-tures and allows in-house installation and configuration The step of sugar boilingrequires highly skilled and experienced operators The program for the operationalsequence of the vacuum pans is based on a flow chart constructed from the knowl-edge and experience of the expert operators One of the results was a reduction inprocessing time An expert system is being developed to measure mixture consist-ency (OpCon, Eaton Corp., Milwaukee, WI, U.S.A.) Torques on mixing shafts aremeasured These values can be combined with other process information such astemperature, pH, moisture, and color

3.5 Business and Manufacturing Operations

This section is an expansion of Section 3.4 In that section the elements of based process control were mainly applied to unit operations This section will de-scribe additional computer-based activities in the plant including information sys-tems As monitoring and control devices for unit operations have developed andhave been applied in food processing plants, information systems have also becomewell established Information from many areas of the company including receiving,inventory, scheduling, quality systems, distribution, and marketing is being managed

computer-on computer databases Part of this informaticomputer-on is being generated automatically bydata collection devices Expert systems can help manage and analyze the increasingamount of data This section will discuss efforts to integrate the manufacturing sys-tems with the information systems

3.5.1 Physical Goods Management

Physical goods management refers to inventory and distribution control One of theearly applications of computers in manufacturing plants was the tracking and dis-

Previous Page

position of goods using computerized databases Whether dealing with raw materials

or finished products many of the techniques are the same Both deal with handlingand delivering items to a customer, one being internal and the other external Ma-terials handling systems control raw material storage, conveyance, and batch blend-ing of dry and liquid products Material consumption and inventory is automaticallymonitored and logged Integration of this information with other areas of the man-ufacturing operation is a basic part of computer-integrated manufacturing (CIM).Automatic storage and retrieval systems (AS/RS) have enabled companies to reducelabor costs and product damage More efficient structures and energy saving designscan be achieved using AS/RS Once an automated system is in place, computerintegration and intelligent control can be incorporated

Computerized delivery and distribution systems are widely used Sales and orderinformation from remote locations can be logged via a modem to the central office

on a daily basis This information can be useful in timely production planning andscheduling Raw material demands can be more accurately predicted The sys-tems can coordinate the information for delivery documents and vehicle loadinginformation

Computer Aided Production Management (CAPM) is a subset of CIM dealingmainly with information management The software for CAPM is modular with eachmodule representing a different area of production The central module is a singledatabase of information on materials, production machines, operations, and routings.Another database module manages the inventory This information includes rawmaterials, partially completed products, and finished goods Current information onlabor is also included Another module deals with sales order processing This systemprovides information for documents such as order acknowledgements, picking lists,and invoices Current information of sales order, stock levels, and discounts is readilyavailable CAPM systems can also access external accounting software

An integrated software system enables data from one module, such as sales orderentry, to be automatically shared throughout the system Errors due to reentry ofdata can be eliminated A customer order can be used to develop production orders,and production supplies can be subtracted from inventories Invoices can be sent outmore rapidly Customer payment information can be used for sales analysis Inven-tory and sales data can be updated rapidly

Several commercial applications are described A cane sugar computer programwas developed to calculate the balance of materials at each stage of the sugar pro-duction process.8 Many developments in warehousing and material handling havebeen observed A system of driverless forklifts has been installed in a bottled waterfacility in France

Computer systems designed mainly for the dairy industry are available fromAlbasoft Systems International Ltd (Glasgow, U.K.) The systems are fully inte-grated and provide control of distribution, doorstep delivery, stock, processing, sales,purchasing, payrolls, and transportation

A software program called SLAMSYSTEM (Pritsker Corporation, Indianapolis,

IN, U.S.A.) can evaluate high-volume packaging lines The system can simulate apackaging line design It can examine interactions of line components and evaluate

operating efficiencies The user observes problem areas and modifies the modelaccordingly The effects of the changes can then be measured in the form of diag-nostic reports and graphs.

An expert system called Packaging Advisor (E.I du Pont de Nemours & Co.,Wilmington, DE, U.S.A.) was reported for rigid food container design.9 PackagingAdvisor was implemented to help induce customers to examine du Pont's barrierresins It automates the design process, allowing the customers to design their ownpackages The system provides information on alternative materials, quantities tomeet performance specifications, and estimates of costs The packaging advisor wasused to inform customers and field office staff of new and existing products

An expert system for truck routing and certain warehouse management functionshas been developed by Performance Analysis Corporation (Research TrianglePark, NC, U.S.A.) Performance Truck Routing was specifically designed for fooddistribution

An example of a highly automated warehouse is provided by C&H Sugar ett, CA, U.S.A.) Flow of sugar into packages is regulated by PLCs Bags are bundledand bar coded After they are palletized, the AS/RS takes control A scanner readsthe bar code and palletizer logic configures the load pattern After the load is stretchwrapped and weighed, sensors check the profile of the load misshapen stacks Atthe warehouse the computer checks available space among racks and assigns a rackspace Items with a higher turnover are placed closer to the front Product types arebalanced among six aisles for optimum availability Shipping schedules and ordersare generated The trucker gets a printout of the proposed load pattern along withthe manifest

(Crock-Performance Analysis Corporation (Research Triangle Park, NC, U.S.A.) buildsexpert systems and near optimal systems for truck routing and certain warehousemanagement functions Performance Truck Routing forecasts future orders based onpast orders Driving distances are minimized as routes are constructed and deliverytimes established Because the optimal route is known, the cost of any extra customerservice can be quantified The system was designed specifically for food distribution

to a fixed customer base

Performance Analysis has developed a warehouse efficiency application that termines the best location for item storage Balancing capabilities are provided rela-tive to aisle, loop, and level for optimal retrieval An expert system is used formaintenance of product layout and family grouping Family grouping saves time bygrouping products that can be selected and palletized together The expert systemrecommends an initial pick slot assignment based on predefined parameters such asthe family group or case height Family groups can be defined based on store mer-chandizing, thus reducing the time required at the store to sort the products Pallets

de-of product arriving at the store can be taken directly from the truck into the aisleand stocked

3.5.2 Time Management: Planning and Scheduling

Common goals of a computer-based planning system would be to schedule ment and processes, maximize effective use of equipment, eliminate idle time, and

equip-provide intelligent advice to the user Various expert systems for scheduling havebeen developed and implemented in industrial settings Some of the systems arecoded in LISP and require sophisticated hardware Smaller packages, usually coded

in PROLOG, PASCAL, or C, have also been successful in some applications.The number of precise computer-based planning and scheduling systems has beengrowing in recent years Advances in database technology have helped make datamore accessible for planning and scheduling Expert systems are qualitative and lessstructured and use inferential reasoning Operations research designs are well struc-tured, quantitative, and numerical in their optimization procedures The integration

of operations research, expert systems, and database technologies can more tively solve planning and scheduling problems Although a number of expert systemtools are available on the market, to deal with large scheduling problems an expertsystem needs to be embedded within the database technology

effec-The three-way integration of operations research, expert systems, and databasetechnology is the basis for the software products MIMI/MJ and MIMI/E (ChesapeakeDecision Sciences, Inc., New Providence, NJ, U.S.A.) MIMI/MJ is designed forproduction planning and scheduling It includes long- and short-term planning, in-teractive reporting, and database integration functions MIMI/MJ considers inven-tory, customer orders, raw material lead times, production capacity, setup cost, andforecast demands It can than deliver daily production schedules, material and re-source requirements, equipment utilization, projected inventories, potential prob-lems, and expected production costs MIMI/E captures the knowledge of expertmodelers and the intuition of experienced production schedulers MIMI/E analyzesand improves the schedule based on the scheduler's knowledge of the plant Pro-duction schedules can be developed using the knowledge and intuition of an expe-rienced scheduler MIMI/E uses both backward and forward chaining on a flexibleknowledge base It can be linked to other databases and can be completely integratedwith operations research techniques

Japanese workers have reported using an expert system to solve planning lems in a goods distribution situation.10 They examined work scheduling, packinglayout planning, and perishable food processing planning

prob-Schedulex (Numetrix, Toronto, Ontario, Canada) is a production scheduling tem that combines the computational capabilities of the computer with the intuitiveabilities of the human scheduler The goal of the scheduler is to find the lowest sum

sys-of all inventory and manufacturing costs Once the costs are defined in a schedulingmodel, Schedulex can evaluate various "What/If" scenarios Optimization algo-rithms can search for an optimum solution The optimization algorithms are a com-bination of nonlinear integer math programs and heuristics The simulator in Sched-ulex uses data in the model to evaluate schedule alternatives Advantages of thissystem include inventory reduction, less overtime, and fewer rush shipments.The development and implementation of an expert system for scheduling in aliquid packaging plant have been reported.11 The packaging plant produces cartonsfor milk, juice, and other beverages A critical point during the process is the printingand cutting operation Two to four cartons are produced simultaneously at the press.When changing orders, the entire press is shut down and the time required depends

on both the previous order and the new order Common colors or common plates

reduce the preparation time The amount of preparation time required and the cacy of the package design affect the quantity of waste generated before the processsettles It is desirable to be able to change more than one carton while the press isstopped However, sufficient labor must be available at that time Also, the rate atwhich one side of the press can run due to the nature of the package controls therate at which the other side of the press runs.

intri-First a model was formulated based on the quantity of cartons, the degree ofdifficulty, the color codes, and the plate codes Initially, the makespan, or timerequired to finish all jobs, was minimized After interviewing the human expertscheduler and plant management a multiple objective scheme was determined to bemore useful The program was written in PASCAL and was incorporated in anintegrated software package, which includes (1) a database management program inwhich the orders are filled, (2) the daily and weekly scheduling program, and (3) a

4 'manipulator" that allows the user to make last-minute changes The manipulatorwas found to be critical to the successful adoption of the system The user wouldoften have last-minute information not available from the database It also accom-modated disruptions and unforeseen events

3.5.3 Computer-Integrated Manufacturing

Computer-integrated manufacturing (CIM) brings together manufacturing systems,information systems, and human systems Manufacturing systems include productdesign, production process, material flow, machine performance, and plant layout.Information systems include system architecture, databases, communication net-works, fault tolerance, and man-machine interfaces Any successful implementation

of a CIM requires close work and communication with the human element Thefinancial risks associated with automation need to be delineated The goals andobjectives need to be clearly defined and the technical and managerial skills neededfor operations identified Appropriate training programs can then be developed.Currently, it is more common for factory automation implementation to occur insmall increments Also, implementation of CIM techniques is usually a gradual proc-ess involving detailed planning, equipment acquisition, training, and implementa-tion A common practice is to begin with islands of automation that are well suitedfor eventual computer integration Stages for a large-scale implementation of CIMmight include defining and developing the system concept; functional requirements;functional design; detailed design, coding, and testing at the unit, module, and systemlevel; then installation, start-up, and audit

The objective of integrating automated factory systems with office informationsystems is to improve performance Specific goals include improved productivity,reduced inventory costs, improved quality, and more flexibility As computer-basedtools are becoming more available, the competitive advantages of using CIM tech-niques are increasing Product analysis from suppliers can be received and processedquicker, reducing inventory holding time prior to shipping With the ability to delivertimely information, a company can respond to customer demands more readily

A major advantage of CIM lies in automatic data acquisition If handled properly,there is less duplication of data and effort There are fewer mistakes, and the dataare more accessible This results in more consistent decision-making Centralizedprocess control increases the value of individual PC-based stations throughout aplant By integrating these stations, the data acquisition rate can increase Improvedmonitoring of product quality is beneficial for regulatory compliance Before begin-ning a CIM project the application needs to be adequately defined and all of the risksand benefits considered Mathematical models of the system can be used to simulateperformance.

Differences exist between the food manufacturing industry and other industries

in which CIM has been applied Operations in which low volume, high return, food products are made will benefit more from a more flexible system The addedexpense may not be justifiable in a high-volume, commodity-oriented business.However, many parts of the food industry have been difficult to automate because

non-of the high degree non-of process flexibility required These same companies stand tobenefit greatly if they can adopt automation technology For example, process controlparameters are more easily modified and the accompanying quality and inventorydata made available sooner Along with the benefits of integrating various computersand control stations throughout a plant, meters, sensors, and other data acquisitionand control devices are becoming more sensitive and accurate than previously usedequipment

Current software programs allow engineers to configure cell level monitoring,diagnostic, and supervisory control systems The software can then gather, analyze,and present data to floor operators It can respond to detected conditions and takecontrol actions automatically Specific applications include data acquisition On-lineconfiguration enables definition and modification of system objects without shuttingdown the system Various alarms are provided Most cell alarms are handled at thecell level Certain alarms can be made visible for hierarchical alarm strategy Con-figured cell applications can initiate transactions with other applications Typicallythe cell serves as a bridge between the shop floor and the rest of the factory Thesystem implements requests from other applications and generates transactions tonotify the factory of material movements, production results, and quality data Re-cord management and historical data management functions are provided Reportsand plots can be generated from logged data Statistical process control techniquescan be implemented and control charts read automatically An alarm is generatedwhen the process is out of control or violates trend rules

With CIM the business person needs a general knowledge in many areas includingfinance, computing, manufacturing, and marketing

In an effort to determine the extent of CIM implementation in the food industry,

a recent survey of western United States food companies was taken.12 The surveyfocused on eight CIM functions including computer-aided design, computer-aidedmanufacturing, computer-assisted quality control, automated materials handling,production planning and control, maintenance scheduling and control, distributionmanagement, and finance and accounting Companies were specifically queried as

to what level computer integration was used in their business The most widespread

use of computers was in the area of accounting, followed by production planning,and distribution management Database systems have become well established in theoffice environment.

The next most common application was in production planning and control Allfirms used computers in inventory control, whereas fewer used them in materialrequirement planning and manufacturing resource planning Less than half of thecompanies used computers in distribution management, which includes order proc-essing, sales inventory, shipping, and invoicing Also, less than half of the companiesused computer-aided manufacturing techniques (CAM) CAM refers to automatedmanufacturing techniques including weight monitoring, numerical control, robots,and flexible manufacturing systems (FMS) About one-third of the companies usedcomputers in quality-control activities Statistical quality control is causing this per-centage to increase Materials handling, maintenance scheduling, and computer-aided design are used in a relatively few number of companies

Dairy processing plants face various challenges as they adopt CIM systems Manyolder and smaller plants simply have a control panel with on-off switches for thedifferent pumps and valves Automated process control may not extend beyond theflow-diversion valve on the HTST pasteurizer and the CIP system Other plants mayhave several PC units without a central computer One approach taken for a majorupgrade in the process control system is to install the new system in parallel withthe old The new system can be tested during nonworking hours Changeover to thenew system will require only a matter of minutes Another approach that is usefulwhen several isolated control units are in place is to integrate them one at a time ACanadian company with a large number of stand-alone PC units in operation changedgradually by first selecting a central control room and an appropriate computer tocommunicate with the existing PCs.11 Next, they converted the manometric gaugesystems showing milk volumes with electronic pressure sensors Then, various me-ters were interfaced with the control center This method was chosen to minimizeplant disruption

In reporting the development of CIM in one food company, a number of vations were made.14 Initially the company wanted a network of integrated dataprocessing systems, but rapid development of process control technologies forcedtheir inclusion This required significant coordination between electronic engineersand data processing specialists

obser-Relative to product development, prior to CIM implementation there was moreflexibility on issuing new product formulations Although the reduced formulae flex-ibility was good for the quality, production, and planning staff, it was viewed ascounterproductive for marketing and product development

Work load was increased in certain areas, such as receiving, to identify incomingmaterial and match it with purchase orders However, this was compensated for bywork savings elsewhere Manufacturing information was more effectively sharedbetween different areas With management of data as a corporate whole, new appli-cations were designed to take advantage of already available data

Orlac Dairy (Vienne, France) uses a three-level hierarchy of computers to controland monitor all plant operations One computer handles raw milk reception, milk

storage, and ultrahigh temperature (UHT) processing; another deals with packagingand palletizing; and the third controls overwrapping, pallet labeling, and finishedproduct storage.

A vendor-independent systems integrator, ITP/Boston, Inc (Cambridge, MA,U.S.A.) describes systems integration as the combining of the three components thatmake up a factory: manufacturing systems, information systems, and human systems.The emphasis on human systems addresses concerns expressed by others.15 Some

of these include a need for more user training, more end-user involvement in thedecision-making process, organization of a formal project team headed by users, andinstallation of the required manpower prior to implementation

Relative to the food industry, ITP developed a large, real-time hierarchical controlsystem for material handling at the Kellogg Salada cereal plant in London, Ontario,Canada The equipment that this system controls includes a high-rise AS/RS, whichhas 6000 storage locations in six aisles, a monorail system, three roller conveyornetworks, and 61 special automatic and manual material handling operation stationsserving the various plant production areas The control system is based on a hier-archical architecture of computers and PCs The PCs are coordinated by real-timeminicomputers The entire operation is directed by material handling managementsoftware on a central plant computer The system is integrated with other productioncontrol, scheduling, and plant management systems

CIM systems have been developed with expert system capability To be practical,

an expert system must integrate itself in MIS and CIM environments Access toexternal databases is almost always necessary in order to perform the desired infer-encing procedure A difficulty lies in the lack of industry standards for intercon-necting knowledge bases A successful architecture is to place the expert system asthe center of a star with the other systems each connected Wide, local, and bus areanetworks are non-AI techniques for integrating data

MAC-PAC (Andersen Consulting, Chicago, IL, U.S.A.) is an operational systemdesigned to integrate manufacturing, distribution, and finance Applications provided

by MAC-PAC include production scheduling, order processing, inventory control,purchasing, accounting functions, capacity planning, product costing, and materialrequirements An expert system technique used by MAC-PAC is Expert Configu-rator With this application the expert knowledge of the employees is transferred tothe system to control all processing from sales order entry through manufacturing.Its comprehensive design can be applied in production and process definition, cus-tomer service, inventory management, and production planning and control ExpertConfigurator lets the user design order entry screens and help text Valid values may

be accessed from each screen for rapid order entry The Expert Configurator adapts

to a particular pricing structure with procedures such as adding a fixed price when

a particular option is selected, calculating a percent discount for a customer type orclass, or calling external programs for more complex calculations Bills of materialand routings can be generated dynamically for selected options

The I/A Series Systems (The Foxboro Company, Foxboro, MA, U.S.A.) are other class of industrial automation that is up a level from distributed control TheIntelligent Automation Series integrates the entire production process It offers an

an-open architecture in which specifications are made public A dual operating systemruns in every processor module in the system A Real-Time Executive in each pro-cessor oversees the real-time tasks, while an Application Executive runs under theReal-Time Executive to simultaneously manage the more resource-oriented, high-level applications tasks such as process optimization and report writing.

The control strategies include EXACT, a knowledge-based system able to tunedifficult control loops EXACT stands for Expert Adaptive Controller Tuning and

is reported by Foxboro as the first successful application of expert system technology

in process control For example, in filling a container or vat it is able to respond tofouling, varying flow rate, nonlinear behavior, vessel modifications, and other proc-ess dynamics Other features referred to as high-level application tools combine withthe relational information management structure to provide high-quality, real-timeinformation Intelligent measurement products such as flow, pressure, and leveltransmitters have been developed to help integrate process information Sophisticatedself-diagnostics isolate problems and report them to the system The I/A Series Sys-tems support high-level application tools for reports, cost analyses, product qualitytracking, and process optimization Foxboro has developed several modular packagesfor specific industry applications, including a multiple effect evaporator control forthe food industry

Gensym (Cambridge, MA, U.S.A.) has made the observation that up until the lastfew years a commonly held philosophy was that plant automation was simply anaid or substitution for manual operator control The introduction of regulators fornumeric control equipment with centralized monitoring was often the extent of au-tomation However, by using advanced programming and integration techniques,much better results could be achieved In applying their INTEGRAL 2000, Gensymoutlines four levels in an integrated factory automation scheme: Level 1, processcontrol; Level 2, process monitoring; Level 3, process management; and Level 4,management information systems In the third level of Process Management, Gen-sym utilizes their expert system G2 In order to maximize economic results, imme-diate access to both technological (first and second levels) and strategic-economic(fourth level) data is necessary The management staff must be able to modify andimplement it with ease and have an on-line simulation facility to allow the conse-quences of management decisions to be examined G2 provided the required com-putation and simulation functions The application of this system in the food industryhas been observed in a sugar factory On-line simulations are used to calculate theeffects of the modifications and the most appropriate strategy to follow For example,

if a centrifuge fails, the processing capacity is reduced Once repair time is mined and entered, the system evaluates the best operating strategy on the basis ofthe forecast duration of the fault, costs incurred, the space available to stock thickjuice, the possibility of recycling the juice, and changes in power requirements.Process Operations Management System or POMS (Industrial Computing De-signs Corporation, Reston, VA, U.S.A.) is a PS/2-based software system developedexpressly for the food and pharmaceutical industries to accomplish three objectives:(1) link planning, MRP, quality analysis, and engineering with process control andoperations; (2) create a complete electronic batch or run record; and (3) monitor and

deter-enforce good manufacturing practices In daily use POMS sends either the operator

or the controller instructions, interfaces with on-line sensors, and passes processinformation along the network POMS is designed to incorporate the common in-formation needs of users within a specific vertical market The POMS softwareconsists of seven main nodules including manufacturing procedures, host interface,orders server, operations supervisor, and operator's module In addition to the POMSmodules, third-party software packages can be incorporated into POMS These in-clude expert system based process modeling and schedule optimization programs.Gensym has developed an architecture for distributed real-time expert systems

and communications products known as Gl Network G2 can integrate a wide range

of computer products and operating systems from major vendors Each G2 edge base may be accessed from all other G2s in the network This allows users totransfer knowledge freely around the organizations Local expert systems solutionscan be combined with decision support expert systems covering an entire organi-zation The GSI data server can connect to multiple data sources such as data ac-quisition systems, mainframe databases, and user programs in other computers Thecommunication between computers in the G2 Network is generally over Ethernet.With the G2 Network users can build networks quickly and easily Each G2 expertsystem can oversee a multitude of tasks occurring within a complex local application

knowl-By connecting several G2s, a distributed intelligent network can be assembled windows enables users in G2 to get real-time advice from a remote G2 knowledgebase Engineers, operators, and maintenance personnel can get current expert infor-mation about the system being monitored

Tele-Databases are a major feature of CIM Closely related to the application of expertsystems in CIM is the use of intelligent databases Mercury KBE (Artificial Intelli-gence Technologies, Inc., Hawtorne, NY, U.S.A.) is a knowledge base environmentgeared for the construction of intelligent database applications It is able to integratesmoothly with SQL compliant relational databases such as Oracle and Db2 SQLstands for structured query language, which is a language designed to interrogateand process data in a relational database The production engine provides forward,backward, and integrated chaining An SQL object-oriented compiler generates code

at the lowest level of the supported database for increased speed The developer addsdeclarative statements to class definitions in the object system They are then auto-matically transformed into the proper SQL accessor functions Included is a pres-entation management facility for generation of end user interfaces including forms,menus, charts, and icons An application can be delivered on a single work station

or a network of computers

The Ministry of the Environment in Germany contracted with Digital EquipmentCorporation GmbH to develop an intelligent, on-line, environmental detection andcorrective action system The system detects potential environmental pollution, sug-gests corrective actions, and performs on-line decision support Various databasesare supplied by on-line data acquisition Also, laboratory tests of air, soil, and waterquality are entered into databases The user interface includes a map of Germanyand presents any state of contamination, the likely cause, and a recommended cor-rective action

3.6 Quality Management Applications

3.6.1 Quality Control Programs

The nutritional properties of dairy products make them useful foods for the growth

of microorganisms as well as humans Rigid sanitation and quality standards havebeen imposed both by governments and companies to provide a safe product with amaximum shelf life Various programs and tools have been developed to help ac-complish this The American Butter Institute/National Cheese Institute (Chicago, IL,

U.S.A.) has prepared a Total Quality Systems Handbook—HAACP along with a

video presentation HACCP (Hazard Analysis Critical Control Points) is a system

of quality assurance that provides monitoring of critical points during food ing in order to prevent defects before they occur.16 General steps include construction

process-of product and production flow charts, identification process-of control points, monitoringprocedures, and a record-keeping system Although the program is not computerbased, its efficiency can be greatly enhanced with the use of currently availablesoftware In addition to computer programs for flow charts, inspection, and reportgeneration, expert system advisors could be useful for correction of problems when

they develop For instance, an example given in the Total Quality Systems Handbook

suggests some actions for starter culture problems: (1) carefully review all dures used in the starter room; (2) test the pasteurized milk for inhibitors; (3) reviewthe sanitation procedures at the cheese vat; (4) check for phage buildup; (5) rotatecultures; (6) use a direct vat set backup; (7) use frozen or dried backups; and (8) tryanother supplier's culture Statements such as these could be prepared as goals in abackward chaining expert system Information given by the user in response to thesystem's inquiries about the problem can lead to the most likely successful solution.The application of statistics in controlling a process is referred to as statistical qualitycontrol (SQC) or statistical process control (SPC) The use of SPC assists supervisorsand other managers to know if a process is operating within predetermined limits.Inspection at a 100% level is unnecessary because properly designed statistical sam-pling can accurately estimate the population Computer programs for SPC havebecome widely available These systems are often able to interface with externaldata collection devices in real-time and other software programs (SPC Express, Ma-jor Microsystems, Huntington Woods, MI, U.S.A.) Quality-control procedures arecombined with data analysis and graphics Control charts, diagnostic tools, and sta-tistical reports can be provided to help measure performance and performance po-tential (STATGRAPHICS, STSG, Inc., Rockville, MD, U.S.A.) Various manipu-lations can be performed on the data Programs such as Minitab (Minitab, Inc., StateCollege, PA, U.S.A.) are capable of integrating SPC procedures with statistical func-tions for monitoring and troubleshooting

proce-In response to foodbome disease outbreaks in the 1980s, the Milk proce-Industry dation and the International Ice Cream Association prepared a computer programknown as the Product Assurance Safety System (PASS, Diagonal Data Corporation,Lakeland, FL, U.S.A.) to assist in improving quality-control programs PASS helps

Foun-to identify risk items, such as the cleaning procedures in the receiving area Specificduties are entered into the computer and a work order is generated For example, alist of duties could include the following: (1) clean and sanitize using CIP systems;(2) clean receiving hose by hand (cream loadout hose stored in separate location);and (3) clean and inspect manhole vent at the end of the day and replace filter Alsolisted are costs and hours required for the activities Duties for management can also

be specified These might include checking records from the receiving room, such

as the tank CIP charts Another management duty could be an inspection of thereceiving area for irregularities

A major objective behind the development of SPC for unit operations has beenimproved performance Depending on the needs of the application, this performancecould be defined as improved precision Losses due to overfilling containers could

be reduced as the standard deviation of the fill weights is reduced Another definitioncould be reduced human operator or worker hours through automation The devel-opment of microprocessors has allowed process control devices to greatly improve

in reliability and usefulness Coupled with data acquisition capability, computercontrol can provide operators, quality support personnel, and general managementlarge amounts of information on which intelligent judgments can be made

Although increased automation decreases operator time, it does create new mands The remaining operators will be required to have a higher level of expertise

de-in order to supervise the system de-in an out-of-control situation The judgment required

to correct a process failure within a highly automated system may reside with only

a few individuals They may be unavailable at the time of the crisis In these tions an expert system can function as an on-line advisor

situa-Reports of expert systems in food quality control programs are emerging Ananalysis and selection computer program was reported for malting barley quality.17Quality traits are stored on a database Data can then be queried to select the breedinglines for any set of criteria contained in the database The program offers a summary

of quality traits and an objective means of selecting for overall malt quality It isalso applicable to other crops The use of expert systems in quality control using ayogurt manufacturing process as an example has been reported.18

Although the applications of expert systems in dairy processing are only nowemerging, their use in various agricultural and dairy production activities has beenwidely investigated.19'20 Assistance to patrons regarding dairy production operationscan help ensure high-quality milk for processing Automatic registering and record-ing of milk production data can be achieved with sensors, terminals in the barn andmilking parlor for direct access by the herdsman, and a master computer to enteradditional data and perform various calculations Information about milk yield andcomponents is useful not only for quality monitoring but also for controlling diseaseand adjusting rations Expert systems for monitoring dairy production can compareincoming data to preestablished standards, interpreting deviations to provide infor-mation on corrective action In addition to this type of monitoring activity, specificexpert systems can be used as diagnostic devices and planning systems

to justify the expense of laboratory automation However, with the introduction oflaboratory robotics providing flexible automation for various laboratory unit oper-ations (LUOs), new applications are developing Unit operations include repetitiveactivities such as weighing, pipetting, separation, and pH determination These LUOscan be combined and rearranged for various applications.

Software systems are available to integrate or network all lab data for archivalstorage, planning, plotting, and processing The platform is able to access dataquickly, and from a single location Data transfer from lab instrument to PCs may

be required Data or format conversion may be necessary Typical software canprocess spectra for word processing or slide-making procedures Systems can au-tomatically control and acquire data from instruments, analyze and process data, andproduce customized reports and real-time charts

Digital computers are components of many laboratory instruments Electronictransducers convert physical or chemical information into electronic form The de-velopment of solid-state electronics and microprocessors has had a significant impact

on analytical instruments Microprocessors are used in analytical instruments as trol elements and as data recorders

con-Improved methods of data acquisition, sensor development, increased regulatoryactivity, improved instrumentation, and inexpensive electronic memory have all con-tributed to the accumulation of large amounts of quality-control data Knowledgebased systems can assist in the screening and analysis of these data The application

of computers to several quality-control areas is reviewed, along with techniques toassimilate that data into useful information

In addition to the general LIMS, other computer programs and systems are able that will accomplish a variety of specific tasks A computer-based system forcereal dough quality data acquisition and analysis has been developed using threemixographs and one farinograph interfaced to an IBM PC.22 The application softwareused to program ASYST (ASYST Software Technologies, Inc., Rochester, NY,U.S.A.) PSYCHR is a program designed to calculate psychometric properties in-cluding enthalpies, entropies, and exergies.23 At a given pressure, the program re-quires input of two properties and then outputs 23 corresponding properties Salt

avail-penetration into hams can be observed using computer X-ray tomography.24 Creepbehavior in viscoelastic foods can be analyzed much more rapidly.25 The needs formore rapid and more reliable methods of microbial food analysis are being met withautomated techniques such as impedance measurement, image analysis, and com-puter-assisted identification of bacteria.26 Robots have been used for automatic ti-trations, total solids sample preparation, gas chromatography, differential scanningcalorimetry, thin-layer chromatography, nitrogen analysis, texture measurement, mi-crobial plate counts, and fat analysis An expert system entitled SEXIA was devel-oped to characterize certain foods, particularly olive oils.27 Up to 50 analytical var-iables such as acidity, color, fatty acids, sterols, and triglycerides can be examined.Characterization of foods uses numerous tools including discriminant analysis andcluster analysis These tests are based on chemical data A problem with this andany type of analysis is that dispersion or variation in data can occur over time due

to environmental factors such as temperature and humidity This type of dispersioncan be reduced using expert systems Each rule is associated with a chemical param-eter Because expert system rules are constructed independently, the displacement

of a chemical parameter over time is minimized with the data of another chemicalparameter

With the expert system SEXIA, samples of olive oil can be characterized andidentified according to the following parameters: (1) if oil is genuine olive oil; (2) theolive zones within Spain where oil was obtained; and (3) the majority variety amongthe most representative varieties of those zones With SEXIA interviews take placewith an analyst and the database Inference rules then relate the findings The expertsystem handles values of 50 analytical variables The taxonomic organization of thedata is a tree graph The nodes of the tree contain information on the chemicalparameters This information is stored in frames Ranges of chemical parameters areobtained from results of the statistical tests that analyzed the data distribution Theywere specified as low, medium, and high

The system uses three kinds of rule sets: the identification rule set, the interrelationrule set, and demons The demon rules are interrelated with the user/computer in-terface The identification rules deal with varieties, olive zones, and denominations

of origin

There are three groups of rules The first determines confidence factors associatedwith varieties, zones, and origins The second rule group works with the resultsobtained with the different kinds of varieties, olive zones, and origins It determinesthe evidence of all categories The third group combines results obtained by thesecond kind of rules to get the final evidence

An improvement using SEXIA over stepwise discriminant analysis is that themany rules based on different chemical parameters can be used to get optimal results.The statistically based classification models of discriminant analysis programs aremore limited and produce less confident results

Overall, SEXIA tries to eliminate data dispersion in the identification of foods.More specifically, SEXIA presents the following features: (1) it allows the use ofheuristic rules combined with others deduced by statistical programs; (2) rules andpropositions are built independently so the effect of random errors is minimized;

(3) it facilitates aggregation of evidence gathered at varying levels of detail: (4) acombination of certainty factors and the Dempster-Shaffer theory gives more vig-orous results than classical statistical methods; and (5) the system provides moreinformation than statistical programs Considerations for the next generation SEXIAinclude the use of fuzzy sets, which will eliminate the restrictions of numeric rangesspecified for the chemical parameters.

3.6.3 Quality Defect Analysis

Expert systems have found perhaps their most popular application in fault diagnosis

or defect analysis Because this is often an activity of the quality systems department,such technology should be useful An example of an expert system designed todiagnose faults on the production line of a chocolate biscuit (cookie) factory wasdescribed.28 The knowledge representation uses five different types of objects in itsknowledge base A trigger notifies the operator that something is wrong Hypothesesare formulated that linked faults to triggers Each fault is checked by a test and action

is taken The system can use a combination of backward and forward chainingstrategies

Several expert systems for the quality control of cheese have been reported TheSEAF (Sistema Esperto per Analisi dei Formaggi) computerized system is designed

to permit rapid analysis of Sicilian cheese compared with a standard reference.29 Ituses and MS/DOS operating system and is written in Turbo Prolog

GRUYEX30 is an expert system for assistance in improving Gruyere technology.Although the Institut Technique du Gruyere (ITG) provides quality control and tech-nical assistance to the French Gruyere cheese industry, the large number of smallproducers have difficulty justifying the cost GRUYEX was developed to be madeavailable through the VIDEOTEX system VIDEOTEX is the widely used Frenchelectronic information access system made available for professionals and families.GRUYEX is designed to help factory personnel correct a cheese fault or determinethe risk of a technology modification

Five experts, the ITG manager, and a knowledge engineer were involved in theknowledge acquisition phase Because technicians were analyzing between 100 and

200 values during their visits, the possibility of eliminating irrelevant factors wasconsidered Only those parameters that were most easily explained and justified wereincluded By this means the pertinent parameters were reduced to 20 or 30 Theknowledge was represented by constructing trees showing links between the faults.All of the entities of the knowledge base are as follows: (1) description of the faults;(2) description of the hypotheses; (3) description of the parameters; (4) explanationrules; (5) verification rules; and (6) action rules

The GRUYEX system was developed using GOLDEN COMMON LISP in theexpert system tool GOLDWORKS (Gold Hill Computers, Cambridge, MA, U.S.A.).Fifteen typical faults were selected These included too many eyes, wrong overallshape, two-tone paste, cracks in the paste, and colored spots on the crust Some ofthe faults required more description and were divided into subfaults For example,

"the texture of the holes is not smooth" was further described as being "orange

rind-like or nut shell-like." Faults were classified into categories such as paste faults,crust faults, and taste faults to facilitate their selection Faults were represented asobjects in GOLDWORKS to assist knowledge-base modifications A fault was de-fined by (1) name, (2) definition, (3) documentation text, and (4) name of its group.The faults were considered as facts in the explanation rules.

The hypotheses were the causes of the faults Although some were very specificand others more general, no priority was attributed to a specific hypothesis Hy-pothesis were either established or not Probabilities or fuzzy logic were not used.The hypotheses were verified by parameters For example, a parameter to verify thehypothesis "soft cheese" could be "extrusion force." More specific hypotheses torefine the initial hypothesis such as "too fat," "too humid," or "too proteolyzed"could be verified with their own appropriate parameters Parameters are also repre-sented as objects and are characterized by (1) name; (2) explanation text; (3) type:numeric or not; (4) domain: list of alphanumeric values or numeric interval; and (5)group: used in prevision approach to help in parameter selection

The explanation rules link hypotheses to faults or other hypotheses The rules areused to generate all possible causes of a fault They can support both backwardchaining and forward chaining Verification rules link parameters to hypotheses.Action rules link verified hypotheses to corrective actions

The system can function in three ways In the fault approach, the user tries toestablish the fault, generate hypotheses and parameters, and provide corrective actionfor the cheese manufacturer In the report approach, the user presents a fault to thecomputer and receives all of the actions that apply to the specified fault The pre-vision approach determines the production risks incurred following some modifi-cation to the system The effects of changing a parameter, such as heat treatment,can be measured

An expert system was developed to trace Cheddar cheese defects to their source.31

In this program the user initially provides information about the sensory attributes

of the cheese such as appearance, body and texture, and flavor The system thenfollows its own lines of reasoning toward the eventual cause of the defect, askingfor additional information about the raw materials, the process, and analytical data

as it proceeds Several conclusions are normally given, with a confidence rating foreach one

3.7 Strategic Operations

3.7.1 Simulation

In order to meet changing consumer demands, flexible manufacturing strategies arerequired Much time and energy can be spent or wasted if the wrong decision ismade Simulation modeling is a tool that can help predict the performance of aproduct or process before it is made or built Simulation is the construction of acomputer-based model of an operation that predicts how the operation will perform.Conditions can be varied and their effects observed using the model system Simu-

lation modeling can be applied to many areas Most of the food industry applicationsinclude material flow and packing Expert systems are able to incorporate resultsfrom these models and make them more generally useful Expert systems can fuseknowledge from different sources and fields of knowledge, and can help in distrib-uting scarce human expertise to many locations.

Computer models have long been used for thermal processing.32"35 Real timecalculations of time and temperature relationships can be made with models Cook-ing effects such as bacterial destruction, nutrient loss, and sensory deterioration can

be predicted As an alternative to placing a thermocouple in a test can with everyretort batch, the temperature of the geometrical center of a can can be predicted.Using a numerical computer model to simulate thermal processing, rapid evaluation

of process deviations and on-line corrections can be made (TechniCAL, Inc.,Metairie, LA, U.S.A.) During simulation, measurable changes in temperature areprojected for the entire process cycle Results include reduced downtime and im-proved productivity

The Pathogen Modeling Program (PMP) (USDA/ARS Eastern Regional ResearchCenter, Philadelphia, PA, U.S.A.) explores how combinations of various factorsaffect the growth of pathogenic organisms Version 3.0 considers five organisms

including Salmonella spp., Listeria monocy to genes, Shigella flexneri, cus aureus, and Aeromonas hydrophila Factors considered are temperature, pH,

Staphylococ-sodium chloride content, Staphylococ-sodium nitrite content, aerobic conditions, and anaerobicconditions Equations for the models were derived by response surface analysis PMPwas designed to be used with Lotus 1-2-3 (Cambridge, MA, U.S.A.) The "pro-gram" was written using a series of Lotus macros and menu-driven queries A series

of queries is presented, and with the last query, the program will display a secondmenu through which results can be viewed A kinetics options displays the calculatedvalues for exponential growth rate, generation time, lag phase duration, and maxi-mum population density The time option provides an estimate of the time requiredfor the microorganism to grow to the predefined level A growth curve calculatedfrom the combination of the variables selected can be displayed Any of the factorscan be changed, and new results presented

Initial efforts in microbial modeling have been directed at pathogens However,

as the databases and models develop, information on spoilage organisms will cumulate Forecasting during each stage of food production will then be possible.This will allow a more integrated approach from raw materials to finished product.Computer programs have been useful in the control of energy consumption.36Controlling energy consumption is important economically due to the high cost ofenergy, but keeping track of consumption can help lead to other more specific prob-lems For example, high energy may be traced to poor heat exchange The poor heatexchange may represent food or cleaner residues coating the equipment surfaces.Once the problem is recognized, corrective actions can be taken

ac-Consumption of fuel or electricity can be compared to predicted values, and can

be expressed as fuel or electricity per unit of milk processed When prepared ingraph form, deviations are easily removed Various sums and ratios of raw materialsused, energy consumed, products consumed, and costs incurred can be calculated

An expert system may be useful here in providing specific recommendations tocorrect out-of-control situations.

Fouling of food processing equipment involves a two-stage process during whichthe surface is first conditioned by protein/surface or protein/salt interactions to allownucleation This-is followed by heavy deposits based on protein/protein interactions.Chemical processes involved in fouling have been examined in developing foulingmodels for a heat exchanger.37 The models have been used to assess fouling problemsduring operation and cleaning and to develop procedures and designs to minimizefouling Each model is dependent on the type of food material involved

The structural properties of aluminum cans can be tested using a computer ulation program.38 To traditionally test a can design, various cans would be manu-factured, filled, and then physically abused by dropping or hitting In a simulateddrop test, the computer knows the shape of the can, its impact speed, the mass andpressure of the contents, and the characteristics of the can material Calculationsrelate factors such as mass, velocity, acceleration, and pressure Using this procedure,testing costs are greatly reduced

sim-Mathematical models have been developed for many processes ranging frompotato storage systems39 to sun drying of tropical fish.40 A computer model devel-oped by an ingredient supplier determines the most effective chelent formula tocontrol metal ions in food processing operations Selecting the right chelant to tie

up unwanted metal ions such as iron, copper, and zinc is complicated and normally

a trial-and-error procedure The computer program, using a database of more than

2700 equilibrium reactions, will predict the most effective chelant or combinations

of chelants.41

Simulation modeling has been applied to the development of a new cheese coolingand brining process.42 A mathematical model was developed based on the transient,three-dimensional heat transfer of a cheese block The shape of the block was con-sidered, as well as the rate of temperature change in the block and in the brine.Information from the model was used in the decision of whether or not to purchase

a new brining system

Expert system technology has been combined with modeling and simulation nology and referred to as intelligent simulation Traditional simulations have littleflexibility and require much customizing to have practical applications A high level

tech-of computer and simulation expertise is required to design and implement simulationtechniques These limitations can be overcome with the development of an expertsimulation tool With the development of larger expert system shells, process spe-cialists have been able to capture the knowledge of simulation experts and creategeneric equipment models (Mercury ISIM, Artificial Intelligence Technologies, Inc.,Hawthorne, NY, U.S.A.) The mathematical techniques are selected and monitored

by the expert system The user describes the processing plant by specifying ment, and the system responds by asking questions about flow rates, pressures, andother process parameters

equip-Traditionally, simulation designs consist of blocks linked together The blocksare subroutines with predetermined solutions based on mathematical equations A

simulation usually consists of at least several hundred blocks Their complexity leads

to difficulty when changes are made

A particular configuration represents the final equations after the interactions tween equipment and materials have been considered This makes it difficult toisolate and see the change of any one part or material The ISIM (Intelligent Sim-ulation) system was created using an object-oriented methodology Each component

be-of the simulation is implemented separately, and the implementer can specify howthe components interact ISIM simulation blocks represent physical devices thatmanipulate objects Traditional simulation blocks manipulate numbers The manip-ulation of an object or material is called a simulation variable Codes associated withthe simulation variable can determine the effects of various actions taken For ex-ample, a boiler will add BTUs to material in an evaporator, without knowing thephysical properties of the material The simulation variable modeling whey proteinconcentrate will then change the temperature appropriately The simulation variableacts to integrate separate parts of a process acting on material

Each unit operation can be modeled independently by a domain expert withoutmodifying other modules Another advantage is that the knowledge base can useheuristic simulation methods Traditional simulation environments require much cus-tomizing in practical applications With ISIM the user can much more easily build

a manufacturing unit from unit operations The use of object-oriented programminghas been an important factor in the development of intelligent simulation

3.7.2 Research and Development

Many of the activities of a research and development (R&D) department can beassisted by the use of expert system programs Expert systems can provide advice

on functions such as database manipulation, ingredient interactions, and strategicplanning Although expert systems have been developed for many food applications,there are certain tasks to avoid Reasoning involving volatile expertise, which isconstantly changing, is difficult to capture Also, knowledge that is disputed betweendifferent experts should be avoided R&D departments often deal with new andchanging information Disagreements over new knowledge domains are common

A major advantage of expert system programs is to free the human technician fromroutine or tedious tasks and allow more time for the type of reasoning for whichhumans are better suited

An R&D operation is very dependent on the information it can acquire and nipulate Intelligent databases are a new technology for information management.The amount of information is constantly growing and the integration of an expertsystem with the database makes the right information more accessible In addition

ma-to in-house databases, large external databases are available through a number ofsearch services Intelligent information retrieval can assist in finding specific refer-ences more easily Expert systems provide not only a method of organizing expertise,but also higher level language constructs that can be used in database programming

Several key databases include product line information, raw material tions, and finished product shelf life progression These functions overlap with qual-ity control responsibilities, but provide useful information in product development.

specifica-A knowledge of the functional properties of food ingredients is critical for R&D.The wide range of ingredients available and the many interactions with other ingre-dients in the food product justify the time and expense involved in capturing as muchrelevant information as possible Information is available in the scientific and tradeliterature Much information is resident with the human experts Because expertknowledge tends to be quite specialized, several experts may be needed during theproduct development process Capturing this domain specific knowledge is the object

of much expert system development work

Quality of products can often be traced to poor experimental design An plete understanding of cause-and-effect relationships during the design phase of aproduct can lead to high costs of scrap, rework, and inspection Statistically basedexperimental designs allow scientists to get the largest amount of information fromthe smallest number of experimental trials The development of computer softwarepackages with statistical techniques has greatly simplified the use of experimentdesigns Dziezak43 suggests the following steps to implementing designedexperiments:

incom-1 Define the purpose of the study and identify the factors and the responses Thefactors can be ingredients or process conditions The responses are the dependentvariables that are measured

2 Develop a model for each response that will predict response values for differentfactors

3 Select an optimization design to test the factors in a minimum number of trials

4 Conduct the experiments in a random order if possible

5 Fit the model to the data using regression analysis generating a predictionequation

6 Examine the data in a graphical form to discern relationships and regions forfurther investigation

The beginnings of computer applications in sensory analysis were the use ofstatistical software packages to analyze the panel data Devices designed to automatedata collection are a more recent development A system may operate as a networkwith a server computer and several terminals Typical features of a sensory programinclude panelist registration, score card preparation and presentation, and resultsmanagement

Touch screens have been developed to automate data collection One systemcomprised of touch screen hardware, software, and user interface has been designated' 'computer aided sensory testing."44 Besides response input, the system can performother sensory tasks such as project setup, rating form construction, and data analysis.King and Morzenti45 compared a computerized mode of quantitative descriptiveanalysis (QDA) scoring with a manual mode Various samples such as turkey patties,doughnuts, and citrus fruit were judged by four to seven panelists Results indicated

that computerized QDA was better than, or as good as, manual QDA Other factorssuch as cost would need to be evaluated.

Even more extensive automation is reported in an Apparatus for Automated sory Testing (ASST).46 The system prepares samples for sensory evaluation, in-structs the subject during the test, records and processes the results, and providesthe next series of samples according to the previous performance of the subject.Expert systems can be applied to statistical modeling Object-oriented expert sys-tem tools provide a system for capturing statistical expertise A library of statisticaltechniques can be related to one another Objects, which pass data between them-selves, can be supervised by other objects or rules to direct problem-solving skillsand automated statistical reasoning Intelligent Statistical Process Analysis (ISPA,Artificial Intelligence Technologies, Inc., Hawthorne, NY, U.S.A.) contains threemajor cooperating expert subsystems The first subsystem is the Statistical Master

Sen-It is an object-oriented model of the expert procedure to develop statistical models.The second subsystem is the Process Expert This system asks the user about theprocess and then self-generates an expert system for use in model validations andvariable selection The last subsystem is the Object-Oriented Database It allowsspecific data to be manipulated and passed from module to module The total systembuilds a tree of potential model variables and the tree is then pruned to reduce thestatistical search space

Planning and scheduling procedures in R&D are similar to those previously cussed Corporate success will be achieved as the right plans are constructed andperformed In some respects, planning techniques are similar to design methods,only with the element of a structured time frame added It is desirable to have all ofthe available information possible R&D planners must also see how information isinterrelated Intelligent decision systems using expert system technologies can helpdeal with complex decision processes such as these The knowledge of leading R&Ddecision consultants has been captured in a computer-based decision workstationcalled R&D Analyst (Strategic Decisions Group, Menlo Park, CA, U.S.A.) TheR&D Analyst expert system constructs an influence diagram The influence diagramrepresents decision problems in a graphical format that shows relationships of de-cisions and uncertainties The Analyst constructs mathematical models that representthis information Critical factors such as number of competitors, foreign demand, orinitial price are then identified, quantified, and ranked

dis-The R&D Analyst also provides an analysis of the current projects showing aplot of the probability of technical success versus the commercial value given tech-nical success The result is an overall status of all the R&D projects showing themost profitable projects and the areas in which the R&D management should focustheir efforts The third analytical tool develops and runs decision analysis models.Commercial search services are an important resource for current information oncompetitive market activity and R&D activities (DIALOG, Palo Alto, CA, U.S.A.;Orbit Search Service, McClean, VA, U.S.A.) Information commonly accessed in-cludes financial condition of competitors, new product development, and marketingstrategies Expert systems can be useful in the development of searching strategies

3.7.3 Training

People are the most important assets of food and dairy processing plants The peopleoperate the plants They make decisions concerning operations at all levels Whilethe top management is charting the course with company goals and objectives, lineworkers and supervisors are making decisions and contributing to the company on

a daily basis It is critical to remember that people are more important than nology Unless projects have supportive people who are well informed, involved,well trained, and well led, difficulty and failure are likely results Unfortunately,much of the best training and development is provided for upper management, whilethe individuals closest to the product receive either less or poorer quality training

tech-A common complaint among employers is the need for extensive training ofrecent college or university graduates This is understandable because most foodscience and dairy manufacturing programs cover a wide variety of subject areas Athorough education in the basic sciences and communications skills should be ex-pected However, training specific enough to meet the needs of each processing entity

is unrealistic Some type of structured training for new employees is almost alwaysrequired

One area of employee training that has become mandatory is hazardous chemicalexposure The Occupational Safety and Health Administration (OSHA) has a goal

of ensuring that employers and employees know about chemical work hazards andknow how to protect themselves To accomplish this they have implemented a rulecalled "Hazard Communications" (HAZCOM) Under HAZCOM, chemical sup-pliers are required to communicate hazard information determined by the chemicalmanufacturers for each of their products Employers are required to communicatethe hazards of chemicals they use to workers They are also to provide training inchemical safety Hazard information and use of labels and material safety data sheetsmust be communicated to employees through a formal training program

One technique that can help reduce the manager and supervisor time devoted toemployee training is the use of intelligent tutoring computer programs These sys-tems not only allow self-paced, unsupervised learning, but they are able to make themost efficient use of the employee's time The time spent relearning what is alreadyknown and understood is eliminated

Individualized self-paced learning materials have long been available in variousforms These include frame-based written materials, slide presentations, movies andvideo presentations, various computer programs, and even interactive video disks.Intelligent tutoring, which utilizes expert system procedures, is one area within thelarger field of computer-assisted instruction The major advantage of intelligent tu-toring is the ability to focus on the specific areas that are lacking in the employee'straining or background For example, an individual may be very familiar with heattransfer and flow rates through a heat exchanger, yet have little understanding of theeffects of heat, acid, and mineral balance interaction on protein destabilization Theintelligent tutoring system is designed to detect those areas, instruct, and evaluatethe information transfer

Although little activity in the use of intelligent tutors has been reported in thefood and dairy industry, a prototype industrial-training system at Bell Communi-cations Research (Livingston, NJ, U.S.A.) called Word of Intelligent Tutoring Sys-tem (WITS) captures the knowledge of the company's best workers For feedbackpurposes, it compares student responses with that knowledge The system chartsstudents' progress and selects supplementary material Licensing of WITS is ex-pected in 1991.

For references to manuals expert system based software can be used on-line Thisallows the manual to be customized for a particular application or personnel Withexpert system techniques the system can determine the skill level of the operatorand provide the appropriate information Also, information can be made available

in spreadsheets and database software

The incorporation of hypertext into an expert system as a means of providing theuser with additional information during a problem-solving session has been reportedfor a program dealing with decreased milkfat yield.47 The form of the informationcan be textual, graphical, or procedural Several benefits include clarification ofambiguous and technical terms and description of underlying principles involved inthe problem-solving strategy The flexible hypertext interface allows users to exploreinformation in a nonlinear method according to their abilities and interests

3.8 Future Trends

The next likely period following the current information age could be referred to asthe knowledge age Expert systems extend the capabilities of traditional program-ming techniques to handle knowledge-oriented tasks Many people have becomeacquainted with knowledge-based systems as they have been developed thus far.However, the technology is changing and with it future trends change For example,the number of knowledge engineers bringing expert systems and domain expertstogether has increased greatly from its once shortage level In the meantime theexpert system tools, at least for small and midsized computers, are developing to thepoint where intermediaries may become less in demand

Standardization of knowledge bases from different suppliers is likely This willbecome increasingly important as more domain-specific systems are marketed Im-proved standardization will allow an expert system shell to support various knowl-edge bases and interface with other software with greater ease Any progress instandardization will require an increase in cooperation among suppliers

Expert system advisors will become more active than passive as they are applied

to real-time process control systems Improvements in software are already ing the use of on-line expert system strategies This trend will help ease the shortage

increas-of experienced operators The improved control and monitoring functions will alsohelp compensate for a decreasing regulatory presence

Consumer and industrial products will increase in intelligence using expert systemprinciples Expert systems are already being used in traditional software This trend

is likely to increase In this respect it is interesting that some vendors are careful to

make no reference to the use of expert systems or artificial intelligence Apparently,this is to avoid a possible stigma associated with this area of computer sciencebecause of its early promotion in the late 1970s and early 1980s, followed by slowimplementation.

An enormous amount of information is becoming available to business managers.Much data is generated around a company's products, including production records,analytical results, and distribution records Marketing data are available from sourcessuch as universal pricing code (UPC) records taken at retail stores, surveys, businessreports, and commercial databases In an effort to organize their data into usefulinformation, companies have started Decision Support System (DSS) groups Expertsystems are being investigated as tools to analyze large amounts of information.48

An expert system can take an overwhelming amount of data, interpret it, and return

a short and pertinent summary, complete with brief graphs

A trend that will occur among successful companies is increased integration ofmarketing and research and development departments Expert analysis of data willassist in this process by reducing miscommunications

Developments in areas such as physics will assist in the creation of new appliedtechnologies Computer science technologies are largely applications of the devel-opments in physics A breakthrough in parallel processing could supply unprece-dented quantities of processing power to tackle artificial intelligence research issues

3 Parsaye, K., and M Chignell 1988 Expert Systems for Experts John Wiley & Sons, New York.

4 Fox, J 1986 Knowledge, decision making, and uncertainty In W A Gale (ed.), Artificial

Intelli-gence and Statistics Addison-Wesley, Reading, MA.

5 Bowerman, G., and D E Glover 1988 Putting Expert Systems into Practice Van Nostrand

Rein-hold, New York.

6 Sletmo, K 1988 Something about how experience from ice cream factories has influenced the

development of an automatic control system N Eur Food Dairy J 54:22-29.

7 Honer, C 1989 Dairy plants get smart Dairy Foods 89:87-93.

8 Castillo, C 1988 A cane sugar production program Material balance calculations in white or raw

sugar production process Sugar Azucar 83:33.

9 Topolski, A S., and D K Reece 1989 Packaging advisor: an expert system for rigid food package

design In Proceedings of the First Annual Conference on Innovative Applications of Artificial

Intelligence, Stanford, CA, March 1989, pp 28-30.

10 Tsushima, L, and T Tenma 1990 Planning expert system in distribution industry Syst Control

Inf 34:332-339.

11 Adler, L B., N M Fraiman, and M L Pinedo 1989 An expert system for scheduling in a liquid

packaging plant In Proceedings of the Third International Conference, Expert Systems and the

Leading Edge in Production and Operations Management Hilton Head Island, SC, May 1989, pp.

505-518.

12 AIy, N A 1989 A survey on the use of computer-integrated manufacturing in food processing

companies Food Technol 43:82-87.

13 Honer, C 1989 'Smart Pump' Avoids Product Damage Dairy Foods 90:74.

14 Farrant, G T J 1989 Learning from experience Computer integrated manufacturing [of foods] at

batchelors Food Sci Technol Today 3:113-117.

15 Weiter, T R 1990 Move the wrench over and pass me the computer Industry Week 239:52-54.

16 Ellinger, R H 1990 Total Quality Systems Handbook—HACCP American Butter Institute/National

Cheese Institute, Washington, D.C.

17 Clancy, J A., and S E Ullrich 1988 Analysis and selection program for a malt quality in barley

by microcomputer Cereal Chem 65:428-430.

18 Vasquez, H J 1987 An integration of system analysis and knowledge base expert system

ap-proaches to the management of quality in food manufacturing systems In J Hollo and D Torley

(eds.), Biotechnology and Food Industry: Proceedings of the International Symposium Budapest,

Hungary, October 1987, pp 4 3 - 5 1

19 Kalter, R J., A L Skidmore, J D Ferguson, and C J Sniffen 1990 Development of an expert

system for management of dairy farms J Dairy Sci 73 (Suppl 1): 162.

20 Doluschitz, R 1990 Expert systems for management in dairy operations Comput Electron Agric.

5:17-30.

21 Spies, R D 1989 Use of a centralized computer system in a cereal laboratory Cereal Foods World

34:214.

22 Pon, C R., O M Lukow, and D J Buckley 1989 A multichannel, computer-based system for

analyzing dough rheology J Texture Stud 19:343-360.

23 Ratti, C , G H Crapiste, and E Rotstein 1989 PSYCHR: a computer program to calculate

psy-chrometric properties Drying Technol 7:575-580.

24 Froystein, T., O Sorheim, S A Berg, and K Dalen 1989 Salt distribution in cured hams, studied

by computer X-ray tomography Fleischwirtschaft 69:220-222.

25 Balaban, M., A R Carrillo, and J L Kokini 1988 A computerized method to analyze the creep

behavior of viscoelastic foods J Texture Stud 19:171-183.

26 Doring, B., S Ehrhardt, F K Lucke, and U Schillinger 1988 Computer-assisted identification of

lactic acid bacteria from meats Systemat Appl Microbiol 11:67-74.

27 Aparicio, R 1988 Characterization of food by inexact rules: the SEXIA expert system /

Chemo-metr 3 (Suppl A): 175-192.

28 Efstathiou, J 1986 Expert system case study: the chocolate biscuit factory Journal A 27:62-68.

29 Russo, C , C M Lanza, and F Tomaselli 1989 Use of expert systems in the quality control of

typical Sicilian cheeses Industrie Allmentari 28:119-130.

30 Malaureille, P., and D Bronisz 1989 GRUYEX: an expert system for assistance in improving the

Gruyere cheese technology In Proceedings of Fifth International Expert Systems Conference,

Lon-don, England, June 1989, pp 193-201.

31 Olsen, R L 1986 Evaluation of artificial intelligence expert systems in cheese defect analysis.

J Dairy Sci 69 (Suppl 1)88.

32 Brown, G., R Nesaratnam, and P Rodriguez 1989 Computer modeling for the control of

sterili-zation processes In Technical Memorandum, Campden Food & Drink Research Association, No.

442, p 124.

33 Richardson, P S., P T Kelly, and S D Holdsworth 1989 Computer modeling for the control of

sterilization process In Technical Memorandum, Campden Food & Drink Research Association,

No 518, p 75.

34 Hayakawa, K I., P de Massaguer, and R J Trout 1988 Statistical variability of thermal process

lethality in conduction heating food—computerized simulation / Food Sci 53:1887-1893.

35 Hachigian, J 1989 An experimental design for determination of D-values describing inactivation

kinetics of bacterial spores: design parameters selected using computer simulation J Food Sci.

54:720-726.

36 Kay, R 1984 A computer-based management information system for spray drying plants N Z J.

Dairy Sci Technol 19:173-176.

37 Fryer, P 1989 The uses of fouling models in the design of food process plant / Soc Dairy Technol.

42:23-39.

38 Pool, R 1989 Is it real, or is it cray?" Science 244:1438-1440.

39 Milanowski, J 1988 Comparative simulation studies of energy consumption in potato storage Acta

Aliment Pol 14:131-138;

40 Doe, P E., and E S Heruwati 1988 A model for the prediction of the microbial spoilage of

sun-dried tropical fish J Food Engin 8:47-72.

41 Mermelstein, N H 1990 Computer modeling service helps control metal ions in foods Food

Technol 44:119.

42 Swientek, R J 1990 Simulation modeling: a powerful tool for optimizing plant design and

oper-ations Food Process 51:99-102.

43 Dziezak, J D 1990 Taking the gamble out of product development Food Technol 44:110-117.

44 Winn, R L 1988 Touch screen system for sensory evaluation Food Technol 42:68-70.

45 King, A J., and A Morzenti 1988 Response freedom in computerized and manual modes of sensory

scoring Food Technol 42:150-160.

46 Hossenlopp, J., G Trystram, and B Heyd 1989 Design and development of an apparatus for

automated sensory testing of liquid products 5c/ Aliments 9:613-631.

47 Jones, L R 1990 Incorporation of hypertext into an expert system for extension education / Dairy

Sci 73 (Suppl 1):186.

48 LaBeIl, F 1991 Expert systems extract insights from the information explosion Food Process.

52:38-46.

4

Dairy Equipment and Supplies

Thomas Gilmore and Jim Shell

4.1 Dairy Equipment and Supplies, 156

4.2 Equipment Common to All Dairies, 160

4.2.3.3 Positive Displacement Pumps, 181

4.2.3.4 Pump Selection Factors, 187

4.2.3.5 Pump Efficiency, 190

4.2.4 Pipe, Valves, and Fittings, 195

4.2.4.1 Sanitary Piping and Tubing, 195

4.2.7.3 Cleaning and Sanitizing, 218

4.2.7.4 Mechanical Cleaning Systems, 219

4.2.7.5 Sanitary Criteria for Processing Equipment, 226

4.2.7.6 The Relation of pH to Cleaning, 234

4.2.7.7 Types of Cleaners, 236

4.2.7.8 Types of Sanitizers, 237

4.2.7.9 Safe Chemical Handling Check List, 238

4.2.7.10 Elements of Chemical Use Control, 240

4.2.7.11 Manual Cleaning and Clean-Out-Of-Place, 241

4.3.4 Concentration and Drying, 261

4.3.5 Cottage Cheese and Other Cultured Products, 277

4.3.5.1 Cottage Cheese, 277

4.3.5.2 Yogurt, 279

4.3.5.3 Fermented Milk Products, 281

4.3.5.4 Green Cheese Products, 281

4.3.6 High-Temperature Processes, 281

4.3.7 Membrane Separation, 288

4.1 Dairy Equipment and Supplies

In all dairies the raw product, milk, moves through or by various pieces of equipmentduring processing prior to packaging as a finished product for the consumer Itemsfound in all dairies include tanks; heat exchangers; pumps; centrifuges; homogeniz-ers; clean-in-place (CIP) systems; refrigeration systems; boilers; and pipes, valves,and fittings tying all of this equipment together

Some equipment is particular to specialty dairy plants such as ice cream freezers

to ice cream plants, churns to butter plants, vats to cheese plants, and evaporators toconcentration plants, to name a few All the major items specific to specialty plantswill be discussed in the middle section of the chapter

Equipment found in all dairies as well as specialty equipment requires a method

of control and recording of the process The controls and method of tracking can bevery simple or quite high tech The types and degree of automation/control will bereviewed in the final section

It would be remiss at this point not to discuss the sanitary standards for dairyequipment, that is, what they are and how they are set and enforced The PMO orGrade A Pasteurized Milk Ordinance is prepared by the U.S Department of Healthand Human Services with the assistance of Milk Sanitation and regulatory agencies

at various federal, state, and local government departments The PMO is generallyrecognized and accepted by the dairy industry and U.S public health agencies as anational standard for milk sanitation The 3-A Sanitary Standards and 3-A AcceptedPractices for processing dairy foods are the main documents used for determining ifequipment design and construction meets the sanitary requirements to ensure such

Figure 4.1 Plot plan (Courtesy of The Omega Company, Janesville, WI, U.S.A.)

MlLKAEC

Of FIC€ AREAF\JT

1.0 STORAGE

FUTURE 1.0 STORAGE

items will not cause contamination to the dairy product The 3-A Sanitary StandardsProgram is a voluntary and self-regulated tripartate, cooperative program within theindustry that working together with state and federal regulators has provided equip-ment manufacturers with defined standards for equipment and the processors amethod of assuring the sanitary condition of equipment they purchase

Although Chapter 5 in this volume discusses in detail the designs of a dairyprocessing plant, some basic illustrations of typical plants are provided here.Although there are many differences in floor plans and equipment juxtaposition,the following diagrams are illustrative of a generic plant They are based on a plantthat is mid to large size, processing about 1 million pounds of milk a day Theproducts manufactured include milk and ice cream

Figure 4.1 is a plot plan of 15 acres and is a minimum for this size facility Allexpansions to double the capacity are shown and expansion to double the capacitywill not interrupt current production Traffic patterns keep trucks backing to thedriver's side—a safety consideration

Figure 4.2 is a main floor plan showing a continuous production flow from rawproducts and dry storage through process to load-out A separate machine room forhomogenizers and separators segregates noise from the rest of the plant With thetendency toward products with extended shelf life, a room pressurized with filtered