knowledge transfer


An Intelligent Adaptive Controller for Bioreactors

Submitted by Robert Babuska, 11-12-2001, TE


Bioreactors are widely used in food and pharmaceutical industries to cultivate microorganisms. To ensure an optimal environment for the microorganisms, the pH value, temperature and dissolved oxygen concentration in the reactor must be controlled within tight bounds. Ideally, one controller should be able to ensure the required performance for a whole variety of processes (different microorganisms, batch, fed-batch or continuous operation), different scales (volumes ranging from 1 liter to 10,000 liters) and throughout the entire process run. The main control challenges are the dependence of the process parameters on the process type and scale, and the time-varying nature of the dynamics due to gradual changes of operating conditions. Industrial experience shows that controllers with fixed parameters cannot fulfill the requirements. Adaptive control has therefore been chosen as an alternative approach.


A model-based adaptive control scheme has been implemented. An important requirement is the robustness of the entire system in order to ensure safe and stable operation under all circumstances. This is achieved by combining well-proven linear identification and control design methods with a fuzzy expert system that supervises the entire process and initiates appropriate actions whenever needed (such as the identification of a new model, re-tuning of the controller, etc.). The control system consists of two levels: feedback loops at the real-time control level and the fuzzy expert system at the supervisory level. The low-level loops are based on standard digital PID controllers whose parameters can be adjusted on line. At request, a test signal is added to the reference or to the control input, in order to ensure proper excitation for model identification. In order to minimize the disturbance of normal process operation, the test signal is adaptively (re-)designed and is only injected when strictly needed. The process input-output data are continuously being collected for performance monitoring purposes.

Status and results

In order to validate the self-tuning scheme, extensive simulations were first done under Matlab/Simulink, using models identified from real-time process data. Then, a large number of fermentations were run in a 20 l glass bioreactor controlled by the Applikon ADI 1065. Currently, a prototype of the control system is implemented in standard computer hardware. The low-level control loops run on a personal computer under the TwinCAT real-time extension of Windows NT. The supervisory system runs on another personal computer under Matlab/Simulink/Stateflow. These two computers communicate via network, using an Active X server. Temperature, pH and dissolved oxygen control loops were tested in several settings. The experimental results proved that even with a poorly tuned initial controller, good performance can be achieved after two or three system identification and controller tuning iterations. The adaptive control scheme is also able to keep track of changes in process operating conditions, such as an increased or decreased feed. It clearly outperforms standard fixed-parameter controller algorithms that are currently used in commercial biocontrollers.

Adaptivity and portability

Adaptivity is an essential feature of the developed system, using a combination of well-proven linear methods and a knowledge-based supervisory system. The implementation of the supervisor as a fuzzy rule-based system proved useful. It is transparent, easy to manage, adjust and extend for the different control lops within the considered process and even for other applications.

More information

This research project has been a cooperation between Applikon Dependable Instruments B.V., Schiedam, Faculty of Electrical Engineering, Eindhoven University of Technology and Faculty of Information Technology and Systems and Kluyver Laboratory for Biotechnology, both at Delft University of Technology. The project was in part sponsored by Senter, project number BTS98083. For more information, please, contact Robert Babuska (R.Babuska@ITS.TUDelft.NL).

technical reports & case studies

executive summaries

scientific papers




search eunite with