Cover ERCIM News 56

This issue in pdf
(64 pages; 8,7 Mb)

Cover ERCIM News 55
previous issue
Number 55
October 2003:
Special theme:
Machine Perception

all previous issues

Next issue:
April 2004

Next Special theme:
Games Techlology

About ERCIM News

Valid HTML 4.01!

< Contents ERCIM News No. 56, January 2004

Integrated Multivariate Statistical Process Control and Condition Monitoring for Fed-batch Fermentation Processes

by Hongwei Zhang

Multivariate Statistical Process Control is seen by many as the key to real-time process monitoring which results in significant improvements in operational safety and efficiency. Members of the Control Engineering Research Group (CERG) at Manchester School of Engineering are developing MSPC technologies, including Principal Component Analysis (PCA), Partial Least Squares (PLS) and Kernel Density Estimation (KDE), and combining them to form an integrated robust process control and monitoring package. Recently, research has been carried out at CERG with particular interests in the applications to fed-batch fermentation processes.

As fermentation processes are responsible for the production of high value added substances such as enzymes, antibiotics and vitamins, they have great importance in the biotechnology industry. From the control engineer's point of view, the fed-batch processes present the greatest challenge. At the same time, there is an enormous economic incentive behind the work on control, optimisation and condition monitoring of fed-batch fermentation processes.

Figure 1: Data Unfolding.
Figure 1: Data Unfolding.

To achieve high performance operation, the optimisation of the primary factors influencing fed-batch fermentation processes is an important task. The ability of optimal control techniques to assist in this optimisation has been investigated. Optimal control techniques rely upon an accurate model of the process and for many years mechanistic models have been used to develop optimal control strategies for fed-batch processes. However, mechanistic models of fed-batch processes are usually very difficult to develop due to the complexity and nonlinear nature of the processes. An attractive alternative to developing a mechanistic model of a process is to build an empirical model using input-output data collected from the process. Among all the existing methods, the multivariate statistical process control techniques, such as PCA and PLS have been proved to be very promising approaches for application to fed-batch fermentation systems. Multi-way Partial Least Squares (MPLS) is used to identify empirical models for fed-batch fermentation processes. The objective function of the optimal control problem for a batch or fed-batch process needs to reflect the performance of the fermenter. Maximisation of yield, selectivity or conversion and minimisation of batch cycle time are some examples of possible objectives. As the optimal control of fed-batch fermentation processes is a challenging dynamic optimisation problem, it is usually difficult to solve because of the nonlinear system dynamics and the constraints on the control and state variables. Typical choices for the objective function in the optimisation of a fed-batch fermentation process include the maximisation of either biomass or metabolite production with respect to the substrate feed rate. Based on the techniques of MPLS and FGA, members at CERG have developed optimal control algorithms that are suitable for application to batch processes. The algorithm utilise linear models of the process that are identified using the multi-way partial least squares technique. The linear models are used within optimal control law to regulate the productivity of the batch by manipulating the substrates that are fed into the fermentation vessel. The proposed control algorithms are successfully applied to different fed-batch fermentation process and their performances compared with alternative control algorithms that have been applied to the applications in the past. The proposed controllers are found to compare favourably with these alternative algorithms.

Figure 2 Figure 1: Data Unfolding.
Figure 2: Detection of a pH sensor fault and pH measurement inferred by the PLS model.

As the objective of control systems for batch and fed-batch processes is always to ensure high quality of the products. A further issue in the successful operation of fed-batch fermentation processes is the early and accurate detection of fault conditions, such as sensor failures or drifts that may occur. The early detection of fault conditions is of great benefit in fed-batch fermentation processes since the earlier that a fault can be detected and acted upon, the lower its impact will be on the process. In some situations this can be critical, for example, a drift on a pH sensor could have catastrophic results on biomass growth if this measurement is used within a feedback control scheme. PCA and PLS have also been seen as very promising approaches for fault detection and isolation. In previous studies condition monitoring and control of fed-batch fermentation processes have been viewed as independent problems. In the research carried out at CERG, these problems are considered together and integrated fault detection and process control schemes are being developed. These schemes rely heavily on the successful development of a PLS model to provide soft-sensing capabilities in a fed-batch fermentation processes. In research implemented at CERG, it is demonstrated through several different case studies that PLS models can be used, not only in a predictive control framework to regulate the growth of biomass within the fermenter, but to also provide fault detection and isolation capabilities. It is further shown from simulations that the integration of the predictive controller and fault detection capabilities provides a useful diagnostic tool for the control system.

Figure 3
Figure 3: Integrated Process Control and Condition Monitoring.

This research so far has seen some very good results in various simulations, efforts now are being made to verify these techniques through lab experiments. Future research also includes the integration of artificial intelligence methods such as expert systems and the multivariate statistical process control methods. This project is funded by EPSRC (Grant Number GR/N24858).

Please contact:
Barry Lennox, University of Manchester, UK
Tel: +44 161 275 4324

Hongwei Zhang, Northeast Wales Institute
of Higher Education, UK
Tel: +44 1978 293 392