Assessment of performance of single-input-single-output loops:

¨      The key literature will be surveyed: Individual experts and centers of excellence for each of the topics within the course will be identified.

¨      Motivational introduction to control loop performance assessment via industrial case studies.

¨      Minimum variance as a benchmark for the lowest achievable output variance: The key concept is the comparison of the actual output variance with the minimum variance. Minimum variance is estimated by time series analysis of routine closed-loop operating data. It characterizes the most fundamental performance limitation of a system due to the existence of time-delays.

¨      Signal processing methods for estimating the minimum variance of univariate control loops: The key algorithms required will be presented and worked examples provided.

Practical implementation:

¨      Practical performance assessment benchmarks: Practically there are many other limitations on the achievable control loop performance. Performance assessment in a practical context by means of a user-defined benchmark or control action constraints will also be discussed.

¨      Control loop performance evaluation via spectral analysis of data.

¨      Survey of commercial products and services.

Assessment of performance of multivariable controllers:

¨      Extensions of performance assessment to advanced multivariable control systems.

¨      Evaluation of the interactor matrix: Extension of minimum variance control benchmarking to multivariable controllers requires knowledge of the time-delay/interactor-matrices, a fairly stringent requirement. However, simpler forms of the interactor matrix allow the performance of the multivariable closed loop system to be evaluated rather easily. The workshop will demonstrate how to estimate this term from routine operating data.

Plant-wide disturbance and root-cause diagnosis:

¨      Plant wide disturbances: A faulty control loop can cause widespread process disturbance. Examples will be presented to motivate the study of methods for detection of plant-wide disturbances and for diagnosis of the root cause.

¨      Common causes of control loop failure: Ender (1993) shocked the process industries with the extent to which valve faults caused poor performance of control loops. A new procedure for diagnosis of valve stiction will be presented.

¨      Detection of plant wide disturbances: A technique will be presented that combines principal components analysis with controller performance assessment to detect measurements at various places in a plant which are influenced by the same process disturbance.

Process Monitoring and Modeling using Multivariate Statistics

¨      Basics of Univariate Process Monitoring – Shewart Charts, CUSUM Charts etc.

The basic concepts will be reviewed. The shortcomings of the univariate approach will be pointed out and the need for a multivariate approach to process monitoring will be outlined.

¨      Introduction to Multivariate Statistical Methods: The theory and key concepts of multivariate methods such as Principal Components Analysis (PCA), Partial Least Squares (PLS) and Canonical Correlation Analysis (CCA) will be presented.

¨      Application to Process Monitoring and Modeling: This segment of the workshop will discuss applications of the above statistical methods to Multivariate Statistical Process Control (MSPC), fault detection & isolation and identification of multivariable dynamic process models from plant data. Numerous industrial case studies will be used to reinforce the theory introduced earlier.