1st Edition
Chemical Process Performance Evaluation
The latest advances in process monitoring, data analysis, and control systems are increasingly useful for maintaining the safety, flexibility, and environmental compliance of industrial manufacturing operations.
Focusing on continuous, multivariate processes, Chemical Process Performance Evaluation introduces statistical methods and modeling techniques for process monitoring, performance evaluation, and fault diagnosis.
This book introduces practical multivariate statistical methods and empirical modeling development techniques, such as principal components regression, partial least squares regression, input-output modeling, state-space modeling, and modeling process signals for trend analysis. Then the authors examine fault diagnosis techniques based on episodes, hidden Markov models, contribution plots, discriminant analysis, and support vector machines. They address controller process evaluation and sensor failure detection, including methods for differentiating between sensor failures and process upset. The book concludes with an extensive discussion on the use of data analysis techniques for the special case of web and sheet processes. Case studies illustrate the implementation of methods presented throughout the book.
Emphasizing the balance between practice and theory, Chemical Process Performance Evaluation is an excellent tool for comparing alternative techniques for process monitoring, signal modeling, and process diagnosis. The unique integration of process and controller monitoring and fault diagnosis facilitates the practical implementation of unified and automated monitoring and diagnosis technologies.
Nomenclature
INTRODUCTION
Motivation and Historical Perspective
Outline
UNIVARIATE SPM
Statistics Concepts
Univariate SPM Techniques
Monitoring Tools for Autocorrelated Data
Limitations of Univariate SPM Methods
STATISTICAL METHODS FOR PERFORMANCE EVALUATION
Principal Components Analysis
Canonical Variates Analysis
Independent Component Analysis
Contribution Plots
Linear Methods for Diagnosis
Nonlinear Methods for Diagnosis
EMPIRICAL MODEL DEVELOPMENT
Regression Models
PCA Models
PLS Regression Models
Input-Output Models of Dynamic Processes
State-Space Models
MONITORING OF MULTIVARIATE PROCESSES
SPM Methods Based on PCA
SPM Methods Based on PLS
SPM Using Dynamic Process Models
Other MSPM Techniques
CHARACTERIZATION OF PROCESS SIGNALS
Wavelets
Filtering and Outlier Detection
Signal Representation by Fuzzy Triangular Episodes
Development of Markovian Models
Wavelet-Domain Hidden Markov Models
PROCESS FAULT DIAGNOSIS
Fault Diagnosis Using Triangular Episodes and HMMs
Fault Diagnosis Using Wavelet-Domain HMMs
Fault Diagnosis Using HMMs
Fault Diagnosis Using Contribution Plots
Fault Diagnosis with Statistical Methods
Fault Diagnosis Using SVM
Fault Diagnosis with Robust Techniques
SENSOR FAILURE DETECTION AND DIAGNOSIS
Sensor FDD Using PLS and CVSS Models
Real-Time Sensor FDD Using PCA-Based Techniques
CONTROLLER PERFORMANCE MONITORING
Single-Loop CPM
Multivariable Controller Performance Monitoring
CPM for MPC
WEB AND SHEET PROCESSES
Traditional Data Analysis
Orthogonal Decomposition of Profile Data
Controller Performance
Bibliography
Index
*Each Chapter Contains a Summary Section
Biography
Ahmet Palazoglu, Ali Cinar, Ferhan Kayihan
"Most texts that attempt to combine SPC or SPM (statistical process monitoring) with automated control methods fail to incorporate multivariate methods as well. This text does an excellent job of covering all the bases in that regard . . . I highly recommend this text for chemical engineers and statisticians interested in learning how statistical methods can be integrated with process control methods."
– Dean V. Neubauer, Corning Inc., in Technometrics, February 2008, Vol. 50, No. 1