David J. Sandoz

The application of principal component analysis and kernel density estimation to enhance process monitoring

Abstract This paper discusses the application of kernel density estimation (KDE) and principal component analysis (PCA) to provide enhanced monitoring of multivariate processes. Different KDE algorithms are studied and assessed in depth in the context of practical applications so that one bandwidth selection algorithm is recommended for process monitoring. The results of the case studies clearly demonstrate the power and advantages of the KDE approach over parametric density estimation which is still widely used. Statistical summary charts are suggested to raise early warning of faults and locate the physical variables which are the prime indicators of the faults.

Real-time monitoring of an industrial batch process

This paper describes the development of a real-time monitoring system for a batch process operated by Aroma and Fine Chemicals Limited. The process shares many similarities with other batch processes in that cycle times can vary considerably, instrumentation is limited and inefficient laboratory assays are required to determine the end-point of each batch. The aim of the work conducted in this study was to develop a data driven system to accurately identify the end-point of the batch. This information can then be used to reduce the overall cycle time of the process. Novel approaches based upon multivariate statistical techniques are shown to provide a soft sensor that can estimate the product quality throughout the batch and provide a long-term estimate of the likely cycle time. This system has been implemented on-line and initial results indicate that it offers potential to significantly reduce operating costs.

Fault detection in continuous processes using multivariate statistical methods

The approach to process monitoring known as multivariate statistical process control (MSPC) has developed as a distinct technology, closely related to the field of fault detection and isolation. A body of technical research and industrial applications indicate a unique applicability to complex large-scale processes, but has paid relatively little attention to generic live process issues. In this paper, the impact of various classes of generic abnormality in the operation of continuous process plants on MSPC monitoring is investigated. It is shown how the effectiveness of the MSPC approach may be understood in terms of model and signal-based fault detection methods, and how the multivariate tools may be configured to maximize their effectiveness. A brief review of MSPC for the process industries is also presented, indicating the current state of the art.

Minimising conservatism in infinite-horizon LQR control

This paper studies the formulation of the constrained infinite horizon linear quadratic regulator control law (CIHLQR). Results from recent studies in this area are extended to show that conditions used in the standard formulation of the CIHLQR law are not necessary, but merely sufficient. Through the use of a novel proof it is shown that for a general SISO system with input constraints and certain conditions imposed, saturated LQR provides the same control sequence as CIHLQR. It is further shown that saturated LQR is equivalent to the CIHLQR in the case of first-order systems, subject to both state and control constraints. Finally, the region of constrained stabilisability is characterised for the case of open-loop unstable first-order systems.

Brief papers: Computer aided control system design applied to milk drying plants

This paper reviews a computer aided control system design facility that has been used to develop a set of control systems that have been implemented on a pilot scale industrial process. The facility caters for a broad range of control situations including those with interactions, time delays and disturbances. Online interactive graphics is used as a design aid for the identification of plant dynamics and for the assessment of control system performance. Control systems may be developed systematically to be structured in a hierarchical configuration on the plant. Particular applications to the plant, an evaporator and a spray drier, are discussed in detail.

Analysis of multivariate statistical methods for continuous systems

Abstract The approach to process monitoring known as multivariate statistical process control (MSPC) has developed as a distinct technology, closely related to the field of fault detection and isolation. A body of technical research and industrial applications indicate a unique applicability to complex large scale processes, but has paid relatively little attention to generic live process issues. In this paper, the impact of various classes of generic abnormality in the operation of continuous process plants on MSPC monitoring is investigated. It is shown how the effectiveness of the MSPC approach may be understood in terms of model and signal-based fault detection methods, and how the multivariate tools may be configured to maximise their effectiveness.

Robust inferential control using kernel density methods

Abstract The use of kernel density estimation (KDE) methods to address the issue of control under process uncertainty and unreliability is investigated. It is shown how the KDE-derived joint probability density function of plant operational data can be used to assist in this task. It is also shown how the estimated density function can be used to support robust inference of important plant variables in addition to the detection and isolation of faults.

Dynamic Programming Approach for Constrained Model Predictive Control

Abstract Constraints are always present in practical control engineering applications. Control of system dynamics without due regard to the constraints can lead to sub-optimal behaviour and may even result in instability. Dynamic Programming provides an obvious means of handling process constraints. However the resulting algorithm is computationally demanding, making it incompatible for real-time implementation. This paper describes the Dynamic Programming solution of constrained model predictive control problem. A novel model structure is proposed such that the solution of the optimisation problem is not required at each controller interval making the method compatible for real-time implementation. The model has input-output structure with incremental actuation and incorporates the system constraints information. The optimal controller is shown to satisfy the constraints on both the actuations and controlled variables.

PLS and its application within model predictive controllers

Abstract Many system identification techniques have been proposed over the last few decades, including ordinary and recursive least squares. Recently, Partial Least Squares (PLS) has become a popular tool in the chemometric community and is beginning to be applied to solve complex industrial process control problems. These studies have tended to ignore the issue of bias with this form of model and it is this issue that is addressed in this article. The paper describes the development of an unbiased recursive PLS algorithm that is successfully applied to two simulated processes.

An Application of Computer Aided Control System Design to Milk Drying Plant

Abstract This paper reviews a computer aided control system design facility that uses recursive least-squares and dynamic programming. A generalised model structure has been established to cater for a broad range of linear systems, extending from the simplest with a single input and output, to full multivariable with time delays and disturbances. Online interactive graphics is utilised as a design aid for the structuring of hierarchical control systems for application to industrial plant. Application to control processes associated with a milk evaporator and a spray drier is described.