A Real -Time Knowledge-Based System for Automated Monitoring and Fault Diagnosis of Batch Processes

Transcription

1 A Real -Time Knowledge-Based System for Automated Monitoring and Fault Diagnosis of Batch Processes Eric Tatara, Cenk Ündey, Bruce Williams, Gülnur Birol and Ali Çinar Department of Chemical and Environmental Engineering Process Modeling, Monitoring and Control Group Illinois Institute of Technology

2 2 Presentation Outline Motivation and Objectives Multivariate Statistical Process Monitoring Integration of Software Rule Base Batch Expert Conclusions

3 3 Motivation and Objectives Fundamental model development is Costly and Time - consuming Time-variant, non-linear process dynamics, highly correlated variables Batch-to-batch variation Multivariate Statistical Process Monitoring (MSPM) scheme for effective on-line monitoring KBS integration provides real-time supervision

4 4 Characteristics of Batch Processes Time-variant behavior Nonlinear, slow and irreversible process dynamics Infrequent on-line measurements Constrained operation Presence of disturbance effects Reproducibility

5 5 Statistical Process Monitoring Assess process for productivity and product quality Use process variables and implement quality control strategies SPM of Batch Processing Issues End-of-batch quality characterization Within-batch assessment at end-of-batch Within-batch prognosis in real-time

6 6 Multivariate Statistical Process Monitoring Use of in-control runs in the historical database Development of the statistical model that characterizes normal operation (NOC) Computation of control chart limits for use in monitoring future batches Simulations with and without disturbances were conducted to test the capabilities of MSPM Techniques

7 7 MSPM Tools 1. Hotelling T 2 Charts 2. Squared Prediction Error (SPE) Charts 3. Contribution Plots (to SPE and T 2 ) 1 2 3

8 Multivariate Statistical Modeling for On-line Process Monitoring 8 Multiway Principal Components Analysis (MPCA) Conventional Techniques Adaptive Hierarchical PCA

9 9 3-way Array Unfolding and Decomposition X = R? r? 1 t r? P r + E I x J x K I x 1 x 1 1 x J x K I x J x K K 1 Batches I X Variables J Time b(1) b(2)?(1) v(1)..v(j)?(2).. v(1)..v(j)..?(k) v(1)..v(j) t scores b(i) p loadings

10 10 On-line Process Monitoring Use loadings to predict scores and calculate residuals as the batch progresses X new (K? J) incomplete until the end of the operation Conventional solutions Fill the unknown observations with zeros Assume the future deviations will remain at their current values Use principal components of the reference set to predict missing values

11 11 On-line Process Monitoring Adaptive hierarchical MPCA Divide the data block X into K block of two-dimensional (I? J) arrays Develop MPCA model iteratively based on each time slice

12 12 Integration of Methods G2 KBS Development Software Create real-time intelligent applications Graphical programming / user interface Object oriented paradigm (OOP) Modularization of software Natural language programming May be edited and recompiled without downtime

13 13 Integration of Software User Top Level Monitoring KB Diagnosis KB Penicillin Model KB DAQ Off-line KB On-line KB Process Equip. KB GSI Bridge C LIB MATLAB

14 14 Rule Base General IF (T 2 OR SPE > 99% UCL) { IF (Contribution [ i ] > Threshold [ i ]) { Activate diagnosis workspace of variable i Process Specific Rules

15 15 Rule Base Process Specific Agitator Substrate High Biomass Penicillin Aeration Acid Empty DO ph High ph Substrate Temp High Base Empty Feed Rate High Temp Set Point Failure Feed Rate Volume High Volume

16 16 Rule Base Process Specific Agitator Substrate High Biomass Penicillin Aeration Acid Empty DO ph High ph Substrate Temp High Base Empty Feed Rate High Temp Set Point Failure Feed Rate Volume High Volume

17 17 Process I/O Structure Process Input Variables Feed Temperature Feed Rate Inflow Air Rate Agitator Power Input Coolant Flow Rate Coolant Temperature Output Variables Substrate Concentration Dissolved Oxygen Biomass Concentration Penicillin Concentration Volume CO 2 Concentration H + Concentration (ph) Fermenter Temperature Generated Heat

18 18 Process Variables Normal Operation

19 19 Matlab Modules Simulation and monitoring code prototyped in Matlab Matlab files converted to C with Matlab C Compiler G2 Standard Interface (GSI) bridge: Bridge between G2 and external C functions Provides network communications Portable code

20 20 BatchExpert TM Software On-line process monitoring in real-time End-of-batch process monitoring Fault detection and diagnosis Phase detection MSPM and physiological-based alarming Advising on the corrective actions Flexible, modularized structure allowing the addition of new modules

21 21 Conclusions An effective integrated on-line monitoring system developed Combined implementation of heuristics and statistical inference Flexible modular software structure for further extensions

22 22 Contact Prof. Ali Cinar Eric Tatara Department of Chemical and Environmental Engineering 10 W 33 rd Street Chicago, IL tel : (312) fax : (312) chee.iit.edu/~control