Measurement, instrumentation, and sensors handbook electromagnetic, optical, radiation, chemical, and biomedical measurement ( pdfdrive com )

Instrumenta-Reflecting the current state of the art, it describes the use of ments and techniques for performing practical measurements in engineering, physics, chemistry, and the life sciences and discusses processing systems, automatic data acquisition, reduction and analy-sis, operation characteristics, accuracy, errors, calibrations, and the incorporation of standards for control purposes

Optical, Radiation, Chemical, and Biomedical Measurement

volume of the Second Edition:

• Contains contributions from field experts, new chapters, and updates to all 98 existing chapters

signal processing, displays and recorders, and optical, medical, biomedical, health, environmental, electrical, electromagnetic, and chemical variables

A concise and useful reference for engineers, scientists, academic faculty, students, designers, managers, and industry professionals involved in instrumentation and measurement research and

Hand-book, Second Edition: Electromagnetic, Optical, Radiation, Chemical, and Biomedical Measurementprovides readers with a greater understanding of advanced applications

Measurement, Instrumentation, and Sensors Handbook

Electromagnetic, Optical, Radiation, Chemical, and Biomedical Measurement

CRC Press is an imprint of the

Taylor & Francis Group, an informa business

Boca Raton London New York

does not constitute endorsement or sponsorship by The MathWorks of a particular pedagogical approach or particular use of the MATLAB® software.

CRC Press

Taylor & Francis Group

6000 Broken Sound Parkway NW, Suite 300

Boca Raton, FL 33487-2742

© 2014 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S Government works

Version Date: 20140113

International Standard Book Number-13: 978-1-4398-4893-7 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid- ity of all materials or the consequences of their use The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy- ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers.

uti-For permission to photocopy or use material electronically from this work, please access www.copyright.com (http:// www.copyright.com/) or contact the Copyright Clearance Center, Inc (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400 CCC is a not-for-profit organization that provides licenses and registration for a variety of users For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for

identification and explanation without intent to infringe.

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com

Preface xiii Acknowledgments xv Editorsxvii Contributorsxix

Part I Sensors and Sensor technology

Roberto Ambrosini, Stelio Montebugnoli, Claudio Bortolotti, and Mauro Roma

Part IV time and Frequency

Michael A Lombardi

Michael A Lombardi

John A Nousek, M.W Bautz, B.E Burke, J.A Gregory, R.E Griffiths, R.L Kraft,

H.L. Kwok, and D.H Lumb

M.M Rad, Halit Eren, and Martin Maier

Part VI Chemical Variables

Part VII Medical, Biomedical, and Health

James T Dobbins III, Sean M Hames, Bruce H Hasegawa, Timothy R DeGrado,

James A Zagzebski, and Richard Frayne

Gourab Sen Gupta

Part IX Signal Processing

Part X Displays and recorders

Introduction

ThesecondeditionofThe Measurement, Instrumentation, and Sensors

Handbookcomesintwovol-umesThiseditionreflectsthestateoftheartinmeasurement,instrumentation,andsensorsInthistwo-volumeset,allchaptershavebeenupdatedand40newchaptershavebeenincludedtoprovidethefinestpossiblereferencethatisbothconciseandusefulforengineerspracticinginindustry,scientistsandengineersinvolvedinR&D,designers,collegeanduniversitypersonnelandstudents,aswellasmanagerstogetherwithmanyotherswhoareinvolvedininstrumentationandmeasurementdesignandapplications

mentation,andsensorsItdescribestheuseofinstrumentsandtechniquesforpracticalmeasurementsrequiredinengineering,physics,chemistry,environmentalscience,andthelifesciencesItalsoexplainssensors,techniques,hardware,andtheassociatedsoftwareThehandbookincludesinformationpro-cessingsystems,automaticdataacquisition,reductionandanalysis,operationcharacteristics,accuracy,errors,calibrations,standards,andtheirincorporationforcontrolpurposesEmphasisisgivenonmod-ernintelligentinstrumentsandtechniques,wirelessnetworkoperations,humanfactors,andmoderndisplaymethodsaswellasvirtualinstruments

Thehandbookcoversanextensiverangeoftopicsthatcomprisethesubjectofmeasurement,instru-surementTheyconsistofequationstoassistengineersandscientistswhoseektodiscoverapplicationsandsolveproblemsthatariseinfieldsnotintheirspecialtyTheyalsoincludespecializedinformationneededbyinformedspecialistswhoseektolearnadvancedapplicationsofthesubject,evaluativeopinions,andpossibleareasforfuturestudyThus,thehandbookservesthereferenceneedsofthebroadestgroupofusers—fromtheadvancedhighschoolsciencestudenttoindustrialanduniversityprofessionals

Thechaptersincludedescriptiveinformationforprofessionals,students,andworkersinterestedinmea-Organization

Inthisedition,thefirstvolumehas10parts,eachconsistingofseveralchapters,foratotalof96chapterswrittenbyexpertsintheirareasItconcentratesonconceptsininstrumentationandmeasurements,spatialvariablemeasurement,displacementmeasurement,mechanicalvariablemeasurement,acous-tics,flowandspotvelocity,thermalandtemperaturemeasurement,andradiationItreflectsrecenttrendsininstrumentationandmeasurementswiththeadditionofanewpartonwirelessinstrumenta-tionConceptsincontrolsystemsandhumanfactorsaregivenasaseparatepart

Thesecondvolumehas10parts,eachhavingseveralchapters,foratotalof98chapterswrittenbyexpertsintheirareasasinvolume1Itconcentratesonsensorsandsensortechnology,electricvariablemeasurement, electromagnetic variables, time and frequency, optical measurements, chemical vari-ables,andmedical,biomedicalandhealth,andenvironmentalmeasurementSignalprocessing,anddisplaysandrecordersconstitutethelasttwopartsofthisvolume

Locating Your topic

Tofindouthowtomeasureagivenvariable,skimthetableofcontents,turntothatpart,andfindthechaptersthatdescribethedifferentmethodsofmakingthemeasurementConsiderthealternativemethodsofmakingthemeasurementandeachoftheiradvantagesanddisadvantagesSelectamethod,sensor,andsignalprocessingmethodManychapterslistanumberofvendorstocontactformoreinfor-mationYoucanalsovisithttp://wwwglobalspeccom/toobtainalistofvendors

For more detailed information, consult the index, since certain principles of measurement mayappearinmorethanonechapter

MATLAB®isaregisteredtrademarkofTheMathWorks,IncForproductinformation,pleasecontact:TheMathWorks,Inc

Wewouldliketothankalltheauthorsfortheirvaluablecontributiontowardthistwo-volume-setbookWeappreciatethetimeandeffortdevotedbyallournewauthorsandthoseauthorswhowentanextramiletoreviseandupdatetheirchaptersWearegratefultotheCRCPressteamfortheirencouragementtopreparethissecondeditionThepublicationofthisbookwouldnothavebeenpossiblewithouttheirtirelessdedicationinputtingittogetherLast,butnotleast,wewouldliketothankallourreadersforselectingthisbookinadvancingtheirknowledgeandtechnicalskills

John G Webster Halit Eren

Editors

John G WebsterreceivedhisBEEfromCornellUniversity,Ithaca,NewYork,in1953andhisMSEEand

PhDfromtheUniversityofRochester,Rochester,NewYork,in1965and1967,respectively

MadisonHeservesasahighlycitedresearcheratKingAbdulazizUniversity,Jeddah,SaudiArabia

HecurrentlyservesasprofessoremeritusofbiomedicalengineeringattheUniversityofWisconsin-ductsresearchonintracranialpressuremonitoring,ECGdryelectrodes,andtactilevibrators

He alsoteachesundergraduateandgraduatecoursesinthefieldofmedicalinstrumentationandcon-Dr Webster is author of Transducers and Sensors, an IEEE/EAB Individual Learning Program (Piscataway,NJ:IEEE,1989)Heiscoauthor,withBJacobson,ofMedicine and Clinical Engineering (EnglewoodCliffs,NJ:Prentice-Hall,1977);withRPallas-Areny,ofSensors and Signal Conditioning,

Second Edition(NewYork:Wiley,2001)andAnalog Signal Conditioning(NewYork:Wiley,1999)He

istheeditoroftheEncyclopedia of Medical Devices and Instrumentation, Second Edition(NewYork: Wiley,2006),Tactile Sensors for Robotics and Medicine(NewYork:Wiley,1988),Electrical Impedance

Tomography(Bristol,UK:AdamHilger,1990),Teaching Design in Electrical Engineering(Piscataway,

NJ:EducationalActivitiesBoard,IEEE,1990),Prevention of Pressure Sores: Engineering and Clinical

Aspects(Bristol,UK:AdamHilger,1991),Design of Cardiac Pacemakers(Piscataway,NJ:IEEEPress,

1995), Design of Pulse Oximeters (Bristol, UK: IOP Publishing, 1997), Medical Instrumentation:

Application and Design, Fourth Edition (Hoboken NJ: Wiley, 2010), Encyclopedia of Electrical and Electronics Engineering(NewYork,Wiley,1999),Minimally Invasive Medical Technology(Bristol,UK:

IOP Publishing, 2001), and Bioinstrumentation (Hoboken NJ: Wiley, 2004) He is also  the  coeditor, with AMCook,ofClinical Engineering: Principles and Practices(EnglewoodCliffs,NJ:Prentice-Hall, 1979)andTherapeutic Medical Devices: Application and Design(EnglewoodCliffs,NJ:Prentice-Hall, 1982);withWJTompkins,ofDesign of Microcomputer-Based Medical Instrumentation(Englewood Cliffs,NJ:Prentice-Hall,1981)andInterfacing Sensors to the IBM PC(EnglewoodCliffs,NJ:Prentice Hall,1988);andwithAMCook,WJTompkins,andGCVanderheiden,ofElectronic Devices for

Rehabilitation(London:Chapman&Hall,1985)

DrWebsterhasbeenamemberoftheIEEE-EMBSAdministrativeCommitteeandtheNIHSurgeryandBioengineeringStudySectionHeisafellowoftheInstituteofElectricalandElectronicsEngineers,theInstrumentSocietyofAmerica,theAmericanInstituteofMedicalandBiologicalEngineering,theBiomedicalEngineeringSociety,andtheInstituteofPhysicsHeistherecipientoftheIEEEEMBSCareerAchievementAward

Halit ErenreceivedhisBEngin1973,hisMEngin1975,andhisPhDin1978fromtheUniversityof

Sheffield,UnitedKingdomHeobtainedanMBAfromCurtinUniversityin1999

Afterhisgraduation,DrErenworkedinEtibank(aminingandmetallurgycompanyinTurkey)asaninstrumentationengineerfortwoyearsHeheldthepostofanassistantprofessoratHacettepeUniversityin 1980–1981 and at Middle East Technical University in 1982 He has been at Curtin University

DrErenwasappointedasavisitingassociateprofessoratPolytechnicUniversity,HongKong,in2004andservedasavisitingprofessorattheUniversityofWisconsin,USA,in2013Heisaseniormem-berofIEEEandparticipatesinRegion10activities,invariouscommitteesfororganizingconferences,andmemberofeditorshipinseveraltransactionsDrErenhasover180publicationsinconference

proceedings,books,andtransactionsHeistheauthorofElectronic Portable Instruments—Design and

Applications(BocaRaton,FL:CRCPress,2004)andWireless Sensors and Instruments—Networks, Design and Applications(BocaRaton,FL:CRCPress,2006)Hehascoedited,withBelaLiptak,Instruments Engineers’ handbook—Process Software and Digital Networks,Vol3,edn4(BocaRaton,FL:CRCPress,

2011)HeisinvolvedinwritinganumberofbooksinthefieldofinstrumentationandmeasurementDrErenresearchesandpublishesonintelligentsensors,wirelessinstrumentation,wirelesssensornet-works,wirelessbiomedicaldevices,automationandcontrol,andverylargecontrolsystems

L Basano

DepartmentofPhysicsUniversitàdiGenovaGenova,Italy

M.W Bautz

ThePennsylvaniaStateUniversityUniversityPark,Pennsylvania

David M Beams

DepartmentofElectricalEngineeringTheUniversityofTexasatTylerTyler,Texas

K Beilenhoff

InstitutfürHochfiequenztechnikTechnischeUniversitätDarmstadtMunich,Germany

Michael Bennett

CentreforAviation,Transportandthe

EnvironmentManchesterMetropolitanUniversityManchester,UnitedKingdom

Tushar Kanti Bera

Robert M Crovella

NVIDIACorporationPlano,Texas

Timothy R DeGrado

DukeUniversityMedicalCenterDurham,NorthCarolina

Alfons Dehé

InstitutfürHochfiequenztechnikTechnischeUniversitätDarmstadtMunich,Germany

Alessandro Dionisi

DipartimentodiIngegneriadell’InformazioneUniversityofBrescia

Brescia,Italy

James T Dobbins III

DukeUniversityMedicalCenterDurham,NorthCarolina

F Domingue

Laboratoiredemicrosystèmeset

télécommunicationsLMST-UniversitéduQuébecàTrois-RivièresTrois-Rivières,Québec,Canada

Enrique Dorronzoro

DepartmentofElectronicTechnologyUniversityofSevilla

Seville,Spain

Achim Dreher

GermanAerospaceCenterWessling,Germany

Richard Frayne

UniversityofWisconsin,MadisonMadison,Wisconsin

K Fricke

InstitutfürHochfiequenztechnikTechnischeUniversitätDarmstadtMunich,Germany

Alessandro Gandelli

DipartimentodiElettrotecnicaPolitecnicodiMilano

Milan,Italy

Julian W Gardner

SchoolofEngineeringUniversityofWarwickCoventry,UnitedKingdom

John D Garrison

SanDiegoStateUniversitySanDiego,California

Manel Gasulla

UniversitatPolitècnicadeCatalunyaBarcelona,Catalonia,Spain

W.A Gillespie

UniversityofAbertayDundeeDundee,Scotland

Olaf Glück

InstitutfürSchicht-undIonentechnikForschungszentrumJülichGmbHJülich,Germany

Daniel Harrison

DepartmentofPhysicsJohnCarrollUniversityUniversityHeights,Ohio

H.L Hartnagel

InstitutfürHochfiequenztechnikTechnischeUniversitätDarmstadtMunich,Germany

Bruce H Hasegawa

UniversityofCalifornia,SanFranciscoSanFrancisco,California

Alireza Hassanzadeh

DepartmentofElectricalandComputer

EngineeringShahidBeheshtiUniversityTehran,Iran

Michael B Heaney

HuladyneResearchandConsultingPaloAlto,California

Albert D Helfrick

Embry–RiddleAeronauticalUniversityDaytonBeach,Florida

David A Hill

USDepartmentofCommerceNationalInstituteofStandardsandTechnologyBoulder,Colorado

Stanley S Ipson

DepartmentofElectronicandElectrical

EngineeringUniversityofBradfordWestYorkshire,UnitedKingdom

Rahman Jamal

NationalInstrumentsGermanyMunich,Germany

Roger Jones

PrimaryChildren’sMedicalCenterSaltLakeCity,Utah

Carmine Landi

UniversitàdeL’AquilaL’Aquila,Italy

W Marshall Leach, Jr.

SchoolofElectricalandComputerEngineeringGeorgiaInstituteofTechnology

Atlanta,Georgia

Jin-Hwan Lee

IntelCorporationRancho,NewMexico

Sungyoung Lee

DepartmentofComputerEngineeringKyungHeeUniversity

Seoul,SouthKorea

Woo Hyoung Lee

DepartmentofCivil,Environmental,andConstructionEngineering

UniversityofCentralFloridaOrlando,Florida

Kathleen M Leonard

DepartmentofCivilandEnvironmental

EngineeringTheUniversityofAlabamainHuntsvilleHuntsville,Alabama

Yufeng Li

HDDR&DCenterSamsungInformationSystemsAmericaSanJose,California

Robert G Lindquist

DepartmentofElectricalandComputer

EngineeringTheUniversityofAlabamainHuntsvilleHuntsville,Alabama

E.B Loewenstein

NationalInstrumentsAustin,Texas

Ana Verónica Medina

DepartmentofElectronicTechnologyUniversityofSevilla

Seville,Spain

Gerry H Meeten

SchlumbergerGouldResearchCambridge,UnitedKingdom

John T Mester

WWHansenExperimentalPhysicsLaboratoryStanfordUniversity

Stanford,California

Graham A Mills

SchoolofPharmacyandBiomedicalSciencesUniversityofPortsmouth

Portsmouth,UnitedKingdom

Jeffrey P Mills

IllinoisInstituteofTechnologyChicago,Illinois

Devendra K Misra

DepartmentofElectricalEngineeringandComputerScience

UniversityofWisconsin-MilwaukeeMilwaukee,Wisconsin

William C Moffatt

SandiaNationalLaboratoriesLivermore,California

Stelio Montebugnoli

InstituteofRadioastronomy,INAFBologna,Italy

Jerry Murphy

ElectronicMeasurementsDivisionHewlett-Packard

ColoradoSprings,Colorado

J Nagaraju

DepartmentofInstrumentationandAppliedPhysics

IndianInstituteofScience,BangaloreBangalore,India

V Palmisano

EuropeanCommission-JRCInstitute

forEnergyPetten,theNetherlands

Ronney B Panerai

LeicesterRoyalInfirmaryUniversityofLeicesterLeicester,UnitedKingdom

Ian Papautsky

BioMicroSystemsLaboratorySchoolofElectronicandComputingSystemsUniversityofCincinnati

Cincinnati,Ohio

E Pasero

DepartmentofElectronicsand

TelecommunicationsPolitecnicodiTorinoTurin,Italy

Maria Teresa Penella

UrbioticaSL

Barcelona,Catalonia,Spain

B.W Petley

CentreforBasicMetrologyNationalPhysicalLaboratoryMiddlesex,UnitedKingdom

Thad Pickenpaugh

AirForceResearchLaboratoryWright-PattersonAirForceBase,Ohio

Luca Podestà

SapienzaUniversityofRomeRome,Italy

Arshak Poghossian

InstituteofNano-andBiotechnologiesAachenUniversityofAppliedSciencesAachen,Germany

Terry L Rusch

DouglasScientificAlexandria,Minnesota

Giancarlo Sacerdoti

SapienzaUniversityofRomeRome,Italy

Ravi Sankar

DepartmentofElectricalEngineeringUniversityofSouthFlorida

Tampa,Florida

Sumita Santra

DepartmentofPhysicsandMeteorologyIndianInstituteofTechnology,KharagpurKharagpur,India

Emilio Sardini

DipartimentodiIngegneriadell’InformazioneUniversityofBrescia

Brescia,Italy

Kalluri R Sarma

DepartmentofAdvancedDisplaysHoneywell,Inc

Phoenix,Arizona

Michael J Schöning

InstituteofNano-andBiotechnologiesAachenUniversityofAppliedSciencesAachen,Germany

Fritz Schuermeyer

UnitedStatesAirForceWrightLaboratoryWright-PattersonAirForceBase,Ohio

Gourab Sen Gupta

SchoolofEngineeringandAdvancedTechnologyMasseyUniversity

PalmerstonNorth,NewZealand

Michal Szyper

AGHUniversityofScienceandTechnologyand

UniversityofMiningandMetallurgyKracow,Poland

Nitish V Thakor

DepartmentofBiomedicalEngineeringSchoolofMedicine

JohnsHopkinsUniversityBaltimore,Maryland

Marion Thust

InstitutfürSchicht-undIonentechnikForschungszentrumJülichGmbHJülich,Germany

Michael F Toner

NortelNetworksNepean,Ontario,Canada

A Toriono

DepartmentofElectronicsand

TelecommunicationsPolitecnicodiTorinoTurin,Italy

E.E Uzgiris

GeneralElectricResearchandDevelopmentCenter

GeneralElectricCompanySchenectady,NewYork

Ramanapathy Veerasingam

ThePennsylvianaStateUniversityUniversityPark,Pennsylvania

Gert J.W Visscher

InstituteofAgriculturalandEnvironmentalEngineering

Wageningen,theNetherlands

James A Zagzebski

DepartmentofMedicalPhysicsUniversityofWisconsin,MadisonMadison,Wisconsin

1 Soft Sensors: Software-Based Sensors Petr Kadlec1-1

Introduction • DataforSoftSensorDevelopment • Data-Driven

TechniquesforSoftSensorDevelopment • ApplicationsofSoft

Sensors • Conclusions • Acknowledgment • References

2 Implantable Systems Vincenzo Luciano, Emilio Sardini, Alessandro Dionisi,

Mauro Serpelloni, and Andrea Cadei2-1

Introduction • Architecture • ForceMeasurementinsideKneeProsthesis • Power

HarvestinginImplantableHumanTotalKneeProsthesis • Conclusions • References

3 Bio-Inspired and Life-Inspired Sensors Cesar Ortega-Sanchez and Halit Eren3-1

Introduction • Bio-InspiredSystems • Life-InspiredSystems • Semiconductor

Sensors • BiomedicalandBiologicalSensors • BiomimeticSensors • Signal

Processing • Bio-InspiredSensorsinIndustry • References

4 Image Sensors Mehdi Habibi4-1

Introduction • BasicPixelTrends • DigitalImageSensors • VisionSensorswith

ProcessingPower • FinalRemarks • References

5 Principles and Technology of SQUIDs Robert L Fagaly5-1

Superconductivity •SuperconductingQuantumInterferenceDevices(SQUID) • CryogenicRequirements • Applications • References • FurtherInformation

6 Next Generation of Smart Sensors Michael E Stanley and Stéphane

Gervais-Ducouret6-1

Introduction • NeedforSensorFusion • UseCases • Power

Considerations • IntegrationandPartitioningIssues • TrendandMarket

Segments • SummaryandConclusions • References • PartialListofSensor

Manufacturers

7 Needle-Type Environmental Sensors Woo Hyoung Lee, Jin-Hwan Lee,

Woo-Hyuck Choi, Ian Papautsky, and Paul L Bishop7-1

Introduction • MicroelectrodeSensors • AmperometricMicrosensors • Potentiometric Microsensors • DataAcquisitionand Processing • MEMS Sensor Arrays • 

Introduction • Transducers • LiquidCrystalChemicalandBiological

Sensors • Conclusion • References

9 CMOS Integrated Gas Sensors Prasanta K Guha, Sumita Santra,

and Julian W. Gardner9-1

Introduction • Microhotplate-BasedGasSensors • InterfaceElectronics • Sensing

Materials • ConclusionandFutureOutlook • References

10 Energy Harvesting for Sensors: DC Harvesters Oscar Lopez-Lapeña,

Maria Teresa Penella, and Manel Gasulla10-1

PhotovoltaicEnergyHarvesting • ThermoelectricEnergyHarvesting • References

11 Energy Harvesting for Sensors: AC Harvesters Maria Teresa Penella,

Oscar Lopez-Lapeña, and Manel Gasulla 11-1

Introduction • RadiofrequencyEnergyHarvesting • MechanicalEnergy

Sensors • Conclusion • References

13 Thermal Sensors: Flow Nam-Trung Nguyen13-1

Introduction • PrinciplesofConventionalThermalMassFlowSensors • Analytical

ModelsforCalorimetricFlowSensing • CalibrationConditions • References

14 SQUID Magnetometers Robert L Fagaly14-1

CryogenicRequirements • MagneticFieldSensing • Geophysical

Applications • NondestructiveTestandEvaluation • MedicalApplicationsof

SQUID • DefiningTerms • References • FurtherInformation

15 Sensor Networks and Communication Robert M Crovella 15-1

Introduction • CommunicationandNetworkingConcepts • NetworkTechnologies • 

ApplyingNetworkCommunications • RecentAdvances • References • FurtherInformation

1.1 Introduction

One of the major trends driving and driven by technological advancement is increasing use of thesensorsandinstrumentsintheworldaroundusExamplesofsensorsenablingtheproliferateduseofinstrumentsaroundtheglobeareingreatnumberpresentedinthischapterAconsequenceofthiseffectisthatincreasingamountofdigitaldataisavailableforfurtherprocessingandexploitationInfact,dataaredevelopingintoapreciouscommodity,whichisoftentradedatveryhighpriceThesedataaretheessenceofsoftsensorsdiscussedinthiswork,andassuchthereisastrongpressureonthequalityofthedataWithsomeexceptions(eg,[1]),algorithmsforsoftsensorsdevelopmentrequireasubstan-tialamountofhigh-qualityhistoricaldatainordertobeabletodevelopusefulsoftsensors,andyetatthesametimethequalityofreal-lifeindustrialdataisoftenverylowByanalyzingthedataobtainedfromtheirsources,oneveryoftenfindsdataimpuritieslikeoutliers,missingvalue,andmeasurementnoise[2]Thecausesoftheseissuesarenumerous:itcanbethephysicallimitsofthehardwaresensors,forexample,incaseofnoisydata,andhardwaresensorfailuresormaintenance,andincasesofmissingdataForthesereasons,thehistoricaldataoftenhavetobemanuallytreatedtoremovetheimpuritiesmentionedpreviously

tialbycreatingtheusefulinformationoutofthemThisprocessiscalledinformationextraction,anditisachievedbyapplyingmachinelearning[3]anddatamining[4]techniquestotherawdataThesemethodscaneitherbeunivariatewithasinglevariable,thatis,transformingasinglemeasurementstream(inputvariable)intoanothermeasurementstream(outputortargetvariable)Anexampleofsuchatechniqueisforecasting,wherebasedonpastvalues,thesoftsensoristryingtoforecastfuturevaluesofthesamevariableOther,morecommon,techniquesaremultivariatemethods,whichuse

Oncethedataarepretreated,theyshouldbereadytobeprocessedinordertoexploittheirfullpoten-1

Soft Sensors: Software-Based Sensors

Petr Kadlec

Evonik Industries

Inparticularintheprocessindustry,softsensorshavebeenestablishedasusefultools,whichcanprovideadditionalinformationabouttheunderlyingprocessesOnecandistinguishtwofundamentallydifferentsoftsensortypes,namely,model-drivensoftsensorsanddata-drivensoftsensors[2]Thereareseveralothertermsthatcanbeusedforthesetwotypesofmodels;theseincludeparametricvsnonpara-metricmodels,whiteboxvsblackboxmodels[5],andphenomenologicalvsempiricalmodels [6]Themodel-drivensoftsensorsusethechemicalandphysicalprinciplesunderlyingtheprocessIngeneralifsuchknowledgeisavailableandexploitablefortherequiredpurposes,itshouldbeusedTheproblemisthat,inchemicalandmanyotherindustries,oftenthisisnotthecasebecausetheprocessesaretoocom-plextobedescribedbyrigorousmodelsintheformofmathematicalorchemicalequationsAprominentexampleofsuchprocesstypeisthebiochemicalprocessessuchaspenicillinproduction[7]

Forthisreason,thefocusofthisworkisonmultivariatedata-drivensoftsensorsThesemodelsrelyondatamodelingtechniquesandaretrainedondatacollectedduringtheoperationoftheprocessThemostcommondata-driventechniquesappliedforsoftsensormodelingaretheprincipalcomponentregres-sion(PCR),principalcomponentanalysis(PCA),andpartialleastsquares(PLS)method,whichwillbeexplainedlaterinthesectiondedicatedtodata-driventechniquesThesetechniquesgainedpopularityduetotheirstatisticalbackgroundandeaseofinterpretabilityofthemodelandbecausetheydealefficientlywithdatacolinearity,whichiscommonamongindustrialdatasetsExamplesofsoftsensorapplicationsbasedonPCA/PLShavebeendiscussedin[8–11]MultilayerPerceptron(MLP)[12]isanotherpredictivetechniquewidelyappliedasasoftsensor[13]ThepopularityoftheLayerPerceptrons(LPs)originatesintheirabilitytomodelnonlinearfunctionsFurthermore,onecanfindarangeofothertechniqueslikesup-portvectormachines(SVM)[14]andneuro-fuzzysystems[15]appliedforsoftsensors[16,17]

Whilethefieldofapplicationofsoftsensorsisbroad,themostcommonapplicationtypeiswhatisreferredtoasonlineprediction[2]Inthecaseofthisapplicationtype,thesoftsensoristrainedusinghistoricaldatabeforeitislaunchedinthereal-lifeenvironmentwhereitstaskistoprovideanonlinepredictiononthebasisoftheincominginputdatastreamAnotherfrequentapplicationtypeofsoftsensorsistheiruseforprocessmonitoringandprocessfaultdetectionThesystemscanbeeitherappliedtodescribe/analyzethenormaloperatingstateortorecognizepossibleprocessfaultsbeforethesestatesoccurCommonly,processmonitoringtechniquesarebasedonmultivariatestatisticaltechniqueslike

PCAormorepreciselyonHotelling’sT2[18]andQ-statistics[19]Comparedtotraditionalunivariate

controlcharts,thesemeasureshave,ontheonehand,theadvantageofconsideringallinputfeaturesand,ontheotherhand,providinginformationaboutthecontributionoftheparticularfeaturestoapossibleviolationofthemonitoringstatistics[20]Softsensorsprovidingacertaindegreeoftranspar-encymayalsobeveryusefulfortheinterpretationoftheunderlyingprocessParticularlyusefularesoftsensorsbasedonPCAandPLSasthesetwomethodsdeliverameasureofhowmuchagiveninputvariablecontributes(1)toexplainingtheinputdatavariance(incaseofPCA)or(2)theoutputvariableofinterest(incaseofthePLS)SoftsensorscanalsobedevelopedtoreplicatehardwaresensorsInsuchacase,thesoftsensorpredictionsareusedincaseoffailureorunavailabilityoftheoriginalhardwaresensor,whichmayoccurduetoitsmaintenance,replacement,etc[21]

Aparticulardrawbackofmanyofcurrentsoftsensorsistheirstaticnature[2]Traditionally,themodelsarenotadaptive,andoncedeployedintothereal-lifeoperation,themodeldoesnotchange(see,eg,[21,22])Incontrasttothis,theenvironment,inwhichthesoftsensorsareapplied,isoftenchang-ing[23]Thecombinationofthestaticmodelandthechangingdataleadstoperformancedeteriorationbecausethemodelusuallyrepresentstheout-of-datestateoftheprocessasitwasobservedduringthetrainingphaseInordertocopewiththisproblem,oneofthefollowingtwoconditionshastobeful-filled:(1)thehistoricaldatahavetocontainallofthestates,whichmayoccurinthefuture,inthiscasethesoftsensorcanbetrainedtocopewithanyofthestatestobeexpectedinthefuture;and(2)ifthe

• Preprocessingandpredictivetechniques:ThestatisticalandmachinelearningtechniquesforthedatapreprocessingandfortheactualsoftsensorOnepreprocessingandtwosoftsensortech-niquesarepresentedinthesectionondata-driventechnique

• Historicaldata:ThehistoricaldataforthetrainingandparameterizationofthepreprocessingandthepredictivemethodsThecharacteristicsandroleofthehistoricaldataareinmoredetaildiscussedinthesectionondataforsoftsensordevelopment

• Expertknowledge:TheroleofexpertknowledgewasbrieflydiscussedintheintroductionofthisworkThegoldenruleforexpertknowledgeisthatithastobeusedforthesoftsensordevelop-mentwhenavailable

Allthesethreepartsareinputfortheactualsoftsensordevelopment,asshowninFigure11Amoredetailedmethodologyforsoftsensordevelopmentcanbefoundin[2]Theoutcomeofthisprocessisatrainedpredictivemodel—asoftsensor

1.1.1 Soft Sensor Operations

Oncedeveloped,thesoftsensorcanbedeployedintothereal-lifeoperationwhereitstaskistodeliverthepredictionscalculatedmeasurementoftheoutputvariablesasshowninFigure12

Inthenextsection,thecharacteristicsofthehistoricalandonlinedatarequiredforthesoftsensordevelopmentandoperationarediscussedindetail

1.2 Data for Soft Sensor Development

1.2.1 Historical Data

Usually,whendealingwithreal-lifeindustrialmodelingtasks,thereisasetofhistoricalrecordingsavailableThisformsthebasisforthedevelopmentofthesoftsensorThesedatahaveanumberofpropertiesrelevantforthesoftsensordevelopmentprocessThehistoricaldataconsistofanumber

Preprocessing and predictive techniques

Historical data

Expert knowledge

Soft sensors development process sensorSoft

FIGURE 1.1 Softsensordevelopmentprocess

FIGURE 1.2 Softsensoroperation

getvariable,inmostcasesonlythelabeledsamples(ie,datapointscontainingthetargetvalues)arerecordedwithinthehistoricaldataTherefore,thesamplingrateoftheinputvariablesandthetargetvariablecanbeassumedasequal

Furthermore,althoughthesamplingrateoftheinputdataisusuallyhigherthantheoneofthetar-AnotherrelevantandpossiblyharmingpropertyofthedataisthedelaysbetweentheinputvariablesthemselvesWithoutpriorprocessknowledge,thesetypesofdelaysaredifficulttocompensateforandhavetobedealtwithbytheapplicationofanappropriatefeatureselectionalgorithm

1.2.2 Online or real-time Data

Oncethesoftsensorbuildingphaseisfinished,themodelisappliedintheonlineoperationandneedstodealwiththeonlinedatastreamIncomparisontohistoricaldata,theonlinedatahaveslightlydif-ferentcharacteristics

Dataarearrivinginanincrementalway,thatis,onesample(orasmallbatchofsamples)afteranotherIngeneral,thesamplingratebetweentheinputandthetargetdatacandiffer

Therealvaluesofthetargetmeasurement,whichusuallyarriveatalowersamplingratethantheinputsamplesandoftenwithcertaindelays,canbeusedtoevaluatethemodelperformanceduringtheonlinepredictionphaseIfthereisanotabledeteriorationofthemodelperformance,anadaptationofthemodelneedsbeperformedusingthetargetdataThecharacteristicsofhistoricaldataandreal-timedataaresummarizedinTable11

1.2.3 Process Data Issues

cussedfurtherinthissection

Figure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis- 1Figure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-Figure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis- Missing values:Figure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-MissingFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-dataFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-areFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-singleFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-samplesFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-orFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-consequentFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-setsFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-ofFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-them,Figure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-whereFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-oneFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-orFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-moreFigure13showsexamplesofvariablesaffectedbycommonissuesofindustrialdatathatwillbedis-

variables(ie,measurements)haveavaluethatdoesnotreflecttherealstateofthephysicalmea-suredquantityTheaffectedvariablesusuallyhavevalueslike±∞,0

ureofahardwaresensor,itsmaintenanceorremovalAsalreadymentioned,industrialplantsareheavilyinstrumentedforprocesscontrolpurposes;therefore,therecordedprocessdataalsoconsistofalargenumberofdiversevariablesInsuchascenario,thereisacertainprobability

MissingvaluesinindustrialcontexthavevariouscausesThemostcommononesarethefail-TABLE 1.1 CharacteristicsoftheHistoricalandOnlineData

HistoricalData Real-TimeData

parameteroptimization Softsensoroperationandadaptation(ifrequired)

Inputvariabledelays Possiblypresent Possiblypresent

Targetvariabledelays Compensated Possiblypresent

Inputvstargetsamplingrate Equal Possiblylower

Sincemostofthetechniquesappliedtodata-drivensoftsensingcannotdealwithmissingdata,astrategyfortheirreplacementusuallyhastobeimplementedAnapproach,whichisveryprimitiveandnotrecommendedbutstillcommonlyappliedinpracticalscenarios,istoreplace the missing values with the mean values of the affected variable Another nonopti-mal approach is to skip the data samples consisting of variable or variables with the miss-ing values, that is, case deletion [25] A more efficient approach to missing values handlingtakesintoaccountthemultivariatestatisticsofthedataandthusmakesthereconstructionofthemissingvaluesdependentontheotheravailablevariablesoftheaffectedsamples(eg,maximum-likelihoodmultivariateapproachtomissingvaluesreplacement[26])ThesetypesofapproachesarerelatedtosensorfaultdetectionandreconstructionFromanotherpointofview,onecandistinguishtwodifferentapproachesfordealingwithmissingvalues[25]Theseare(1)singleimputationwherethemissingvaluesarereplacedinasinglestep(using,eg,mean/medianvalues)and(2)multipleimputation,whichareiterativetechniqueswhereseveralimputationstepsareperformed

In[27],analgorithmbasedoniterativelyreweighedleastsquaresisappliedtodealwithmissingandnoisydataThisalgorithmislimitedtotheestimationofdynamiclinearsystemparametersonlyTheauthorsshowthatthealgorithmcandealwithasituationwheretheprobabilityofmiss-ingdataislessthan50%,providedthatahighnumberofsamplesareavailable

dlingwaspresentedin[26,28]

Anothertechniquefordealingwithmultipleimputationtechniquesformissingvalueshan-(b) Var 1

Var 2 Var 3 (a)

(d) (c)

FIGURE 1.3 Commonissuesfoundinindustrialdatasets:(a)missingvalues;(b)outliers;(c)datacolinearity;

(d) measurementnoise

 2 Data

outliers:Outliersaresensorvaluesthatdeviatefromthetypical,orsometimesalsomeaning-ousoutliersandnonobviousoutliers[29]ObviousoutliersarethosevaluesthatviolatethephysicalortechnologicallimitationsForexample,theabsolutepressuremaynotreachnegativevaluesortheflowsensormaynotdelivervaluesthatexceedthetechnologicallimitationsofthesensorTobeabletodetectthistypeofoutlierefficiently,thesystemhastobeprovidedwiththelimitingvaluesintheformofaprioriinformationIncontrasttothis,nonobviousoutliersareevenhardertoidentifybecausetheydonotviolateanylimitationsbutstilldonotreflectthecorrectvariablestatesOutlier detection as part of the data preprocessing remains very critical for soft sensor

ful,rangesofthemeasuredvaluesOnecandistinguishbetweentwotypesofoutliers,namely,obvi-development because undetected outliers have a negative effect on the performance of themodels For example, the influence of a single outlier can be critical for the PCA [30–32]Anotherproblemofoutlierdetectionisthatevenwhenapplyingautomaticoutlierhandlingpreprocessingsteps,usuallytheresultshavetobevalidatedmanuallybythemodeldeveloperThegoalofthemanualinspectionistodetectanypossibleoutliermasking(ie,false-negativedetections, undetected outliers) and outlier swamping (ie, false-positive detections, correctvalueslabeledasoutliers)[33]

TypicalapproachestooutlierdetectionarebasedonthestatisticsofthehistoricaldataThemostsimpleapproachisthe3_outlierdetectionalgorithm(see,eg,[9,34]),whichisbasedonunivariateobservationsofthevariabledistributionsThismethodlabelsalldatasamplesoutof

therange_(x)_3_(x),where_(x)isthemeanvalueand_(x)thestandarddeviationofthevariable

x,asoutliersAmorerobustversionofthisapproachistheHampelidentifier[35],whichusesa

moreoutlierresistantmedianandmedianabsolutedeviation(MAD)values[33,34]tocalculatethelimits

Theinfluenceofoutliersontheidentificationoflinearandnonlinearmodelsisdiscussedin[33]Forthehandledmodels,theHampelidentifierisfoundtobeaneffectiveapproachfordealingwithoutliersIn[36],amovingwindowfilteriscombinedwiththeHampelidentifiertoobtainanoutlierdetectionandremovalsystemIncontrasttotheunivariateapproaches,themultivariatemethodsusecombinationsofmorefeaturestodetecttheoutliersAnexamplefromthisgroupbasedonthePCAistheJolliffeparameter[37,38]Atwo-stageoutlierdetectionapproachisdis-cussedin[39]ThefirststageistheapplicationofthePCA,afterthistheHotelling’sT2measure[18]canbeusedtodetectoutliercandidatesthatarelocatedoutsideofthe99%confidenceellipseThesecandidatesarethenfurtheranalyzedinthesecondstep,whereScheffe’stest[40]isappliedtothesepoints

 3 Drifting data:Therearetwotypesofdriftingdata,anddependentonthecauseofthedrifts,one

candistinguishbetweensystemandsensordriftsThecausesofthesystemdriftarethechangesof the underlying systemdueto abrasion of equipment, aging of instruments, environmentalconditions,etcTheseareparticularlyprevalentinindustrieswithalargenumberofmechanicalelementsthatundergosteadyabrasionduringtheiroperationAnothercauseofsystemdriftscanalsobeexternalinfluenceslikechangingenvironmentalconditions(eg,weatherinfluence),thepurityoftheinputmaterials,andcatalystdeactivationThesefactorshavenotonlyaninfluenceonthedatabutaffecttheprocessstateaswellTherefore,thedriftsshouldberecognizedandreported,andappropriateactionsshouldbetakentoremovetheirsourceThisisdifferentinthecaseofsensordrifts,whicharecausedbychangesinthemeasuringdevicesandnotbytheprocessitselfThecriticalpointisthatthistypeofdrifts,whilestillobservedinthemeasureddata,doesnotreflectanychangesinthesystemTherefore,inthecaseofsensordrifts,theactiontobetakenshouldbetherecalibrationofthemeasurementdevicesortheadaptationofthesoftsensorIntermsoftheeffectsofdriftsonthedata,onecanobservechangesinthemeansandvariancesofthesinglevariablesaswellaschangesofthecorrelationstructureofthedata[41]

DistinguishingbetweenthetwodifferentdriftcausesdiscussedischallengingandonceagainalotofexpertknowledgeisneededinordertotakeappropriateactionAnotherchallenging

ThemostcommonapproachtodealwithdynamicsinthedataistoapplythemovingwindowtechniqueInthiscase,themodelisupdatedonaperiodicbasisusingonlyadefinednumberofthemostrecentsamplesSomeexamplesoftheapplicationofthistechniqueinthecontextofsoftsensormodelingwerepublishedinReferences[42–45]

 4 Data colinearity:Anotherchallengingissueforsoftsensingisrelatedtothestructureofthedata

Typically,thedatameasuredinheavilyinstrumentedindustriesarestronglycolinearTheseresultsfromthepartialredundancyinthesensorarrangement,forexample,twoneighboringtemperaturesensors,willdeliverstronglycorrelatedmeasurementsSuchenvironmentsareoftencalleddatarichbutinformationpoor[46]However,forsoftsensingtherequirementsaredifferent,inthiscaseonlyinformativevariablesarerequiredAnythingelseisunnecessarilyincreasingtheinputdatadimen-sionality,whichoftenhasanegativeeffectonthemodeltrainingandperformance

tipleinputvariablesintoanewreducedspacewithlesscolinearityasitisdoneinthecaseofthePCAandPLSThesetwoapproachesarethemostpopularonestodealwithdatacolinearityintheprocessindustryExamplesofapplicationswherePCAisusedare[9,43,47,48]andforthePLS[49,50,22]AnotherwaytohandlecolinearityistoselectasubsetofinputvariablesthatarelesscolinearTheseapproachesaresummarizedundertheumbrellaofvariable(orfeature)selectionmethodsinthecomputationallearningresearchAgeneralreviewofthesemethodsispresentedin[51]Somefeatureselectionmethodsinthecontextofsoftsensingarealsodiscussedin[38]Amongthediscussedapproachesarecorrelation-andpartialcorrelation-basedfeatureselectionaswellasMallows’Cpstatistics

TherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul- 5TherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-TherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul- Sampling rates and measurement delays:TherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-VariousTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-sensorsTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-usuallyTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-workTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-atTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-differentTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-samplingTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-ratesTherearetwowaystodealwiththecolinearityproblemOnewayisbytransformingthemul-

andthusonehastotakecaretosynchronizethemThesynchronizationofthedataisusuallyhandledbytherecordingdatabasesThedefinitionofsuchathresholdisanothercriticalpoint,whichinfluencesthequalityofthehistoricaldataSoftsensorsareoftenappliedinmultiratesystemswithseveraloperatingsamplingratesSuchascenariooccursinasystemwheresomeofthevariables,usuallycriticalfortheprocesscontrol,areevaluatedinlaboratoriesatamuchlowersamplingratethantherestoftheautomaticallymeasureddata

Additionalissuesofthedataaretheprocess-relateddelaysinthemeasurementsForinstance,materialsintheprocessesusuallyhaveagivenrun-timethroughtheprocess(eg,thedwellperiodwithinareactorordistillationcolumn),andthus,itisnotreasonabletorelatetwodif-ferentmeasurementstakenatthesametimeatdifferentlocationswithintheprocessInsteadofthis,thedelaysintheparticularmeasurementsshouldbecompensatedbysynchronizingthevariablesHowever,inordertoperformthesynchronization,anextensiveknowledgeaboutthesystemisrequired

 6 Measurement noise:Measurementnoiseisanothercommoneffectobservedinindustrialdata

Mostoftheapproachestosoftsensordevelopmentaretryingtocopewithmeasurementnoiseduring the preprocessing stage of the data processing This is achieved mainly by applying asmoothing(averaging)filterasapreprocessingstep

basedmethod,itcandealwithmeasurementnoiseaslongasitcanbeassumedasGaussiannoise,thatis,randomlydistributedwithzeromeanvalue[9]Inthesameworktheauthorsalsohigh-lighttheapplicationofrobuststatistics,thatis,usingthemedianinsteadofthemeanoperatorandMADinsteadofthestandarddeviation,forthenormalizationofnoisydata

ThePCAisalsoausefultoolfordealingwithmeasurementnoiseAsaleastmeansquares-ZamprognaetalhaveshowntherobustnessofthePLSmethodtowardmeasurementnoisein[22]TheauthorshaveshownthatthereareonlysmallchangesofthepredictionerrorofaPLSsoftsensorwithincreasingnoiselevelsTheexplanationofthisfactisthatthenoiseinfluencesmainlythehigher-orderlatentvariablesthatarenormallyskippedinpracticalapplication

1.3 Data-Driven techniques for Soft Sensor Development

ThissectiondescribesthethreemostcommonmethodsforthedevelopmentofsoftsensorsThePCAcanbeeitherusedasadatapreprocessingmethodorincombinationwitharegressionmethodasafullsoftsensor,insuchcasethemethodisreferredtoasPCRPLSisanotherpopularmethodforsoftsensing[2]ThismethodisinparticularveryusefulforadaptivesoftsensorsThethirdpresentedmethodistheartificialneuralnetwork(ANN)andinparticularitspopularformoftenusedfornon-linearmodelingcalledMLP

1.3.1 Principal Component regression

with 1 and

1 1 1 1 1

n

[ , , , ] ,( ,

whereP∈R ,E n m× ∈R aretheloadingsandresidualsmatrices,respectivelyn m×

ThereareseveralwaystocalculatethematrixPInthecaseofthecovarianceapproach,thecorrelation matrixCoftheinputdataXhastobecalculated:

Inthissection,themostcommonsoftsensingtechniqueanditsadaptive(recursive)versionareout-andmean-centeredinputdataXandoutputdataY∈R (wherepisthenumberofoutputvariable n p×streamsthataresupposedtobepredicted)toseparatelatentvariables:

Thelatentvectors,whichareorthonormaltoeachother i.e.,t t( i T j = ∀0 i j≠ )

,areamorecompactdescrip-tionoftheinputdataThecolumnvectorsp∈Rm l×andq∈R oftheloadingmatricesPandQrepre- p l×

forthecalculationofPLSistheNIPALSalgorithm[52]TheNIPALSalgorithmcalculatesonelatentvectoraftertheotherinaniterativewayTheexplainedcovarianceisremovedfromthedataaftertheeachiteration:

 X i+ 1=X i−t p i i T  (114)

 Y i+ 1= −Y i u q i i T (115)

whichisfollowedbythecalculationofthenext,thatis,(i +1)thvectorsforthePLSouterandinner modelsusingthenewdatamatricesX i+1 andY i+1Thenumberofcalculatedlatentdimensionsisusuallyestablishedusingcross-validationorsomeotherparameteroptimizationtechniques

Inthenonadaptivemodelingscenario,thematricesP, T, Q,

U,andBarecalculatedduringthetrain-ingphasebasedonthebatchofhistoricaldataHowever,asdiscussedintheintroductionaswellas

in [44],thisapproachisoftenproblematicbecausetheprocessanditsdataarechangingoverthetimeAnadvantageofthePLSalgorithmisthatithastheabilitytoincrementallyintegratenewdataTherecursivepartialleastsquares(RPLS)hasbeenintroducedin[54]andfurtherclarifiedin[44]Itcanbeusedtoadaptthemodelinseveralways:(1)onsample-by-samplebasis,(2)byintegratinganewbatchofdata,or(3)bymergingtwoPLSmodels

new

T

i new

TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi- 1TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi- MLP:TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-AnTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-MLPTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-isTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-aTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-feed-forwardTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-ANNTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-asTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-shownTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-inTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-FigureTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-1TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-4TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-ItTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-consistsTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-ofTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-oneTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-inputTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-layer,TheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-oneTheoriginalideaofANN[12]wastobuildcomputationalmodelsmotivatedbytheoperationofbiologi-

outputlayer,andatleastonehiddenlayerTheroleoftheinputlayeristocollecttheinputdataandprovideittothehiddenlayerforfurtherprocessingThenumberofunitsintheinputlayerisequivalenttothedimensionalityoftheinputdataEachoftheinputunitsisconnectedtoeach

Wherex jinisthejthvariableoftheinputsample,w j i,hiddenistheweightbetweenthejthinputunitand

Input layer Hiddenlayer Outputlayer

1.4 applications of Soft Sensors

TheapplicationsofsoftsensorscanbefoundacrossmanyfieldsofanyindustryThemostcommonapplicationtypesarepresentedinthefollowingsectionsofthiswork;pleasenotethatthisisonlyshortextractofthelargenumberofapplicationsofsoftsensorsAcomprehensivelistofcasestudiescanbefoundin[2,24,57]

1.4.1 Online Prediction

ThemostcommonapplicationofsoftsensorsisthepredictionofvaluesthatcannotbemeasuredonlineusingautomatedmeasurementsThismaybefortechnologicalreasons(eg,thereisnoequipmentavailablefortherequiredmeasurement),economicalreasons(eg,thenecessaryequipmentistooexpensive),etcThisoftenappliestocriticalvaluesthatarerelatedtothefinalproductqualitySoftsensorscaninsuchscenariosprovideusefulinformationaboutthevaluesofinterestandinthecasewhenthesoftsensorpredictionfulfillsgivenstandards,itcanalsobeincorporatedintotheautomatedcontrolloopsoftheprocess[58]Data-drivensoftsensorshavebeenwidelyusedinfermentation,polymerization,andrefineryprocesses

1.4.2 Process Monitoring and Process Fault Detection

pervisedlearningorbinaryclassificationtaskThesystemscanbeeithertrainedtodescribe/analyzethenormaloperatingstateortorecognizepossibleprocessfaults[59–61]Commonly,processmonitoringtechniquesarebasedonmultivariatestatisticaltechniqueslikePCA,ormorepreciselyonHotelling’sT2

AnotherapplicationareaofsoftsensorsisprocessmonitoringProcessmonitoringcanbeeitheranunsu-[18]andQ-statistics[19]Thesemeasureshave,ontheonehand,theadvantageofconsideringallinput

features,thatis,usingmultivariatestatisticsand,ontheotherhand,providinginformationaboutthecontributionoftheparticularfeaturestoapossibleviolationofthemonitoringstatistics[20]

1.4.3 Hardware Sensor Backup

SoftsensorscanalsoactasbackupsensorsforhardwaresensorswithatendencytofailureorwitharequirementforperiodicmaintenanceInsuchasituationinordertopreventadisturbanceofthewholesystem,softsensorscanbedevelopedtoreplacethehardwaresensorsduringtheirunavailability[21]

1.5 Conclusions

Undoubtedly,softsensorswillgainimportanceinthenearfutureForthisreason,thisworkfocusedonthedescriptionofsoftware-basedsensorsasavaluablealternativetohardwaresensorsThegreatestpotentialofsoftsensorsliesinsituationswheretheapplicationhardwaresensorreachesitspractical,physical,oreconomicallimitsInsuchsituations,softsensorscanoftendeliverbettermeasurementsAprerequisitefortheabilitytodeliverusefulmeasurementsishowevertheavailabilityofhigh-qualityandrelevantdataforthetrainingofthesoftsensor,whichinpracticalapplicationsoftencomesatthepriceofintensivemanualdatapreprocessingonlyThisworkdescribedtheprocessofsoftsensordevel-opmentandoperationanddiscussedthemainaspectsthathavetobeconsideredduringthedevelop-mentandapplicationofasoftsensorThemostcommontechniquesappliedforsoftsensorswerealsooutlinedindetailinthiswork

acknowledgment

Parts of the study presented in this chapter were carried out during the author’s affiliation withBournemouthUniversity,UnitedKingdom