Modeling, Diagnostics and Process Control Józef Korbicz and Jan Maciej Kos´cielny (Eds.) Modeling, Diagnostics and Process Control Implementation in the DiaSter System ABC Prof.JózefKorbicz UniversityofZielonaGóra, InstituteofControl&ComputationEngineering ul.Podgórna50, 65-246ZielonaGóra Poland E-mail:[email protected] Prof.JanMaciejKos´cielny WarsawUniversityofTechnology, InstituteofAutomaticControlandRobotics Ul.Chodkiewicza8 02-525Warszawa Poland E-mail:[email protected] ISBN978-3-642-16652-5 e-ISBN978-3-642-16653-2 DOI10.1007/978-3-642-16653-2 LibraryofCongressControlNumber:2010938641 (cid:2)c 2010Springer-VerlagBerlinHeidelberg Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthemate- rialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation, broadcasting, reproduction onmicrofilmor inanyother way, andstorage indatabanks. Dupli- cationofthispublicationorpartsthereof ispermittedonlyunder theprovisions oftheGerman CopyrightLawofSeptember9,1965,initscurrentversion,andpermissionforusemustalways beobtainedfromSpringer.ViolationsareliabletoprosecutionundertheGermanCopyrightLaw. Theuseofgeneraldescriptivenames,registerednames,trademarks,etc.inthispublicationdoes notimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Typesetting:Datasuppliedbytheauthors CoverDesign:ScientificPublishingServicesPvt.Ltd.,Chennai,India Printedonacid-freepaper 987654321 springer.com Preface The continuous increase in the complexity of modern industrial systems and objects as well as growing reliability demands regarding their operation and control quality are serious challenges for further development of the theory and practice of control and technical diagnostics. Thus modern control sys- tems are complex in the sense of implementing numerous functions, such as processvariableprocessing,digitalcontrol,processmonitoringandalarmin- dication, graphic visualization of the course of a process, or data exchange with other systems or databases. Moreover, modern control systems are in- tegrated with management systems, which very often cover production and corporate management problems. Hardware and software structures of con- trol systems of complex processes are decentralized and space distributed. Decentralization consists in dividing system operation into many function unitsworkingsimultaneously.Inintegratedsystems,softwareforcontroland visualization creates one information system with a common database. The maindrivingforcebehindthedevelopmentofmoderncontrolsystemsisrapid evolvementofcomputertechniques,whichhaveforcedthestandardizationof fieldnetworksandprogramminglanguagesofcontrolsystems.Theenormous possibilities of technical and program realization of control systems permit significant extension of their functions and tasks, including the introduction of advanced algorithms of process modeling, control and diagnostics. The present book conveys a description of the developed DiaSter system as well as characteristics of advanced original methods of modeling, knowl- edge discovery, simulator construction, diagnostics, control, and supervision controlappliedinthesystem.Thesystemgivesthepossibilityofearlyrecog- nition of abnormal states of industrial processes and faults or malfunctions of actuators as well as technological and measuring units. The universality of solutions assumed in DiaSter allows its broad application, for example, in power, chemical, pharmaceutical, metallurgical and food industries. The system is a world-scale unique solution, and due to its open architecture it can be connected practically with any other control systems. VI Preface Inthemainpartofthebook,analyticalandartificialintelligencemethods implemented in the DiaSter system are discussed. One of the first chapters is devoted to problems of physical process modeling, fundamental in mod- ern control,diagnostics, or designing and searching for alternative solutions. The known analytical as well as neural, fuzzy and neuro-fuzzy models are presented. Particularly the latter, that is, artificial intelligence methods, are attractive for controlsystems as they give possibilities for describing nonlin- ear processesin quite an easy way.In turn, taking into accountthe fact that modern Distributed Control Systems (DCSs) as well as Supervision Control And Data Acquisition (SCADA) systems allow collecting a huge amount of process data coming from different sources, in another chapter methods and algorithms of knowledge discovery in databases are considered. In DiaSter, process data can be a source of diagnostic knowledge,which—obtained with the techniques of knowledge discovery—can be used for fault detection and localization of processes and systems. An extensive chapter is devoted to describing diagnostics methods of pro- cesses and systems, which can be found useful in industry and were imple- mented in the DiaSter system. First of all, robustfault detection algorithms designed by employing the adaptive decision threshold approach are dis- cussed.Suchthresholdswereassignedforfaultdetectionsystemswithdynam- ical neural models such as multilayer perceptrons and the Group Method of DataHandling(GMDH).Methodsoffaultlocalizationareconsideredmainly with the application of fuzzy logic. The proposed approach is based on in- ference rules robustwith respectto structure changesofdiagnosedprocesses or systems. It includes inference algorithms implemented in DiaSter for sin- gle and multiple faults, different ways of dealing with symptoms occurrence delays, or reference algorithms in a hierarchical structure. Moreover, taking into account advantagesand disadvantages of methods of symptom diagnos- ticsandthemodelbasedapproach,inthebookabelief-network-basedmodel is presented as well. It is a heuristical model that permits sequential appli- cation of methods characteristic for both classes of diagnostic investigation. In another chapter, methods of supervision control implemented in the system are discussed. Elementary structures and algorithms of predictive control, the so-called MPC (Model-based Predictive Control), including the DMC (Dynamic Model Control) algorithm for linear models, are presented. Also,fundamentalmethods ofautomaticadjustmentofthe PIDcontrolloop followed by precision adjustment and adaptation are considered. Describing asetpointcontrolsystem,the wayoftransmittingthesetamountgenerated in the optimization layer of the process working point to the control loop in the direct control layer is presented. The last chapter illustrates the operation of different functions of DiaSter with a simple system of three tanks. The chosen plain example has many educational advantages and gives an excellent possibility for exact study- ing of fundamental characteristics and possibilities of the DiaSter system. By using system tools, a simulator of the three-tank system, dynamical Preface VII models ofthe GMDH andthe multilayerperceptrontype aswellasthe TSK (Takagi–Sugeno–Kang)model were built. The realizationof diagnostic tasks was shownboth for systematic diagnostics ofabruptfaults as wellas the de- tectionandtrackingofthe developmentofslowlyincreasingfaults.Here,the control systems used in the object, that is, traditional PID controllers with automatic adjustment of settings and predictive controllers, are presented as well. The present monograph is in a sense a continuation of our earlier book entitled Fault Diagnosis. Models, Artificial Intelligence, Applications (Springer,2004),anditsfirsteditionwaspublishedinPolishbyWydawnictwa Naukowo-Techniczne, WNT, in 2009, (Warsaw, Poland). This english lan- guage version is not merely a translation of the original—many chapters contain some significant improvements, such as new or extended parts and examples, found especially in Chapter 7. The book presents theoretical and practical results of research into fault diagnosis and control conducted over manyyearswithincooperationbetweenPolishresearchteamsfromthe War- saw University of Technology, the University of Zielona Go´ra, the Silesian UniversityofTechnologyinGliwice,andtheTechnicalUniversityofRzeszo´w. In the years2007–2009,the above-mentionedconsortiumof universities con- ducted a developmental project entitled Intelligent diagnostic and control assistance system for industrial processes DiaSter. The editors wish to express their gratitude to all authors for preparing jointchaptersandforveryfruitfulcollaborationduringtheeditorialprocess. July 2010 Jo´zef Korbicz Zielona Go´ra/Warsaw Jan Maciej Ko´scielny Contents 1 Introduction............................................. 1 Jan Maciej Ko´scielny 1.1 Control System Structures.............................. 1 1.2 Trends in the Development of Modern Automatic Control Systems .............................................. 5 1.3 New Functions of Advanced Automatic Control Systems ... 7 2 Introduction to the DiaSter System ..................... 15 Jan Maciej Ko´scielny, Michal(cid:2) Syfert, Pawel(cid:2) Wnuk 2.1 Introduction .......................................... 15 2.2 System Structure and Tasks ............................ 15 2.2.1 Main Uses of the System ......................... 15 2.2.2 System Functionality ............................ 17 2.2.3 System Structure................................ 25 2.3 Software Platform ..................................... 27 2.3.1 Information Model and the System Configuration ... 29 2.3.2 Central Archival Database and User Databases...... 32 2.3.3 Data Exchange.................................. 37 2.3.4 Modeling Module................................ 39 2.3.5 On-line Calculation Module....................... 44 2.3.6 Visualization Module ............................ 48 3 Process Modeling........................................ 55 Krzysztof Janiszowski, Jo´zef Korbicz, Krzysztof Patan, Marcin Witczak 3.1 Introduction .......................................... 55 3.2 Analytical Models and Modeling ........................ 57 3.2.1 Basic Relations for Balance Dependencies .......... 58 3.2.2 Integration Methods ............................. 62 X Contents 3.2.3 Pneumatic Cylinder: A Balance Model ............. 64 3.2.4 Pneumatic Cylinder: A Block Model ............... 70 3.3 Linear Models: Local Approximation of Dynamic Properties ............................................ 72 3.3.1 Dynamic Model Linearization..................... 72 3.3.2 Pneumatic Cylinder: A Linear Model .............. 74 3.3.3 Pneumatic Cylinder: An Optimized Linear Model ... 78 3.4 Parametric Models .................................... 82 3.4.1 Discrete Linear Parametric Models ................ 83 3.4.2 Identification of the Coefficients of Parametric Models......................................... 86 3.4.3 Pneumatic Cylinder: A ParametricLinear Model .... 89 3.5 Fuzzy Parametric Models............................... 92 3.5.1 Fuzzy Parametric TSK Models.................... 92 3.5.2 Estimation of Fuzzy TSK Model Coefficients........ 94 3.5.3 Pneumatic Cylinder: A TSK Fuzzy Model .......... 96 3.6 Neural Models ........................................ 99 3.7 Neural Networks with External Dynamics ................ 100 3.7.1 Recurrent Networks ............................. 101 3.7.2 State Space Neural Networks ..................... 103 3.7.3 Locally Recurrent Networks ...................... 104 3.7.4 GMDH Neural Networks ......................... 111 3.7.5 Implementation of Neural Models in the DiaSter System......................................... 118 4 Knowledge Discovery in Databases ...................... 119 Wojciech Moczulski, Robert Szulim, Piotr Tomasik, Dominik Wachla 4.1 Introduction .......................................... 119 4.2 Selection of Input Variables of Models.................... 121 4.2.1 Correlation-BasedFeature Selection ............... 122 4.2.2 Measures Based on Correlation.................... 123 4.2.3 Searching through the Feature Space............... 124 4.3 Discovery of Qualitative Dependencies ................... 125 4.4 Discovery of Quantitative Dependencies .................. 128 4.4.1 Support Vector Machines......................... 128 4.4.2 Methods Involving Case-BasedReasoning .......... 134 4.5 Conclusion ........................................... 152 5 Diagnostic Methods ..................................... 153 Wojciech Cholewa, Jo´zef Korbicz, Jan Maciej Ko´scielny, Krzysztof Patan, Tomasz Rogala, Michal(cid:2) Syfert, Marcin Witczak 5.1 Introduction .......................................... 153 5.2 Specificity of the Diagnostics of Industrial Processes ....... 154 5.3 Fault Detection Methods ............................... 155 Contents XI 5.4 Robust Fault Diagnosis ................................ 160 5.4.1 Robust Neural Model: The Passive Approach ....... 161 5.4.2 Fuzzy Adaptive Threshold: The Passive Approach... 164 5.4.3 Robust Dynamic Model: The Active Approach ...... 166 5.4.4 Robust Model Design Examples................... 169 5.4.5 Implementation of Neural Models in the DiaSter System......................................... 175 5.5 Process Fault Isolation with the Use of Fuzzy Logic........ 179 5.5.1 Forms of Diagnostic Relation Notation............. 179 5.5.2 Reasoning Algorithm for Single and Multiple Faults.......................................... 184 5.5.3 Algorithms of Reasoning in a Hierarchical Structure....................................... 195 5.6 Application of Belief Networks in Technical Diagnostics .... 206 5.6.1 Introduction .................................... 207 5.6.2 Belief-Network-BasedDiagnostic Model ............ 210 5.6.3 Input Data Images .............................. 213 5.6.4 Additional Variables ............................. 219 5.6.5 Belief Networks ................................. 222 5.6.6 Model Identification and Tuning .................. 229 5.6.7 Implementation in the DiaSter Environment ........ 231 6 Supervisory Control and Optimization................... 233 Piotr Tatjewski, Leszek Trybus, MaciejL(cid:2)awryn´czuk, Piotr Marusak, Zbigniew S´wider, Andrzej Stec 6.1 Predictive Control and Process Set-Point Optimization..... 234 6.1.1 Principle of Model-Based Predictive Control ........ 235 6.1.2 Dynamic Matrix Control Algorithm ............... 240 6.1.3 Generalized Predictive Control Algorithm .......... 247 6.1.4 Non-linear Predictive Control..................... 250 6.1.5 Optimization of Set-Points ....................... 256 6.1.6 Examples ...................................... 259 6.2 Self-tuning and Adaptation of Control Loops.............. 267 6.2.1 Step Response Method........................... 267 6.2.2 Relay Self-tuning................................ 276 6.2.3 Loop Adaptation................................ 282 6.2.4 Function Blocks................................. 291 7 Application of the DiaSter System....................... 295 Michal(cid:2) Syfert, Pawel(cid:2) Chrzanowski, Bartl(cid:2)omiej Fajdek, MaciejL(cid:2)awryn´czuk, Piotr Marusak, Krzysztof Patan, Tomasz Rogala, Andrzej Stec, Robert Szulim, Piotr Tomasik, Dominik Wachla, Marcin Witczak 7.1 Introduction .......................................... 295 7.2 System of Automatic Control and Diagnostics............. 296 XII Contents 7.3 Process Information Model in the DiaSter Platform........ 298 7.4 Applications of the DiaSter System Packages.............. 302 7.4.1 Process Simulator ............................... 303 7.4.2 Self-tuning: Selection of PID Settings .............. 308 7.4.3 Reconstructing Process Variables with TSK Models......................................... 313 7.4.4 Process Modeling with Neural Networks............ 319 7.4.5 Incipient Fault Tracking.......................... 326 7.4.6 On-Line Diagnostics with Fuzzy Reasoning ......... 328 7.4.7 Belief Networks in a Diagnostic System ............ 338 7.4.8 Knowledge Discovery in Databases ................ 349 7.4.9 Model Predictive Control with Constraints ......... 362 References................................................... 369 Index........................................................ 381