Using the adaptive control theory and an Auto-tuning technique, various digital control algorithms for a regulation of the cell mass concentration in a CSTBR (continuous stirred tank bioreactor) and the dissolved oxygen concentration in a batch culture system have been developed and their experimental studies were investigated.
In chapter 2, two types of digital control algorithms were experimentally examined for the control of yeast cell-mass concentration in a CSTBR. One of them is a high-gain proportional control resulted from a highly output-weighted quadratic cost regulator. The other is a stochastic adaptive control named ELS-STR, which is a minimum variance controller involving an extended least-squares method as a parameter identifier. On-line estimation of the state variable, cell mass concentration, was accomplished by measuring the optical density of the fermentor broth. From experimental results, it was known that the performances of these control schemes were better than that of a conventional PID controller tested.
In chapter 3, four bilinear adaptive controllers were developed for the control of a continuous yeast cell cultivation process. The process was described by a SISO-bilinear model having only one unknown time-varying parameter. Each control algorithm consisted of an exponential data-weighting least-squares parameter estimator and a generalized one-step ahead controller, and the design variables were set to specific values. Those algorithms were named BAC (Bilinear Adaptive Control), WIBAC (Weighted-Input Bilinear Adaptive Control), SBAC (Simple Bilinear Adaptive Control), and SDC (Simple Digital Control). Experiments with a laboratory-scale CSTBR demonstrated that the developed bilinear adaptive controllers had satisfactory dynamic properties. Atheoretical proof of global convergence of the WIBAC algorithm was established.
In Chapter 4, a new adaptive DO(dissolved oxygen) concentration control algorithm considering DO electrode dynamics including a response time delay has been developed. Three types of system models with two time-varying parameters were employed to relate the DO concentration with two control input variables, air flow rate and agitation speed. Parameters of these models were estimated on-line using a regularized constant trace recursive least-squares method. An extended Kalman filter was used to remove the effect of noises from the DO concentration measurements and thus to improve control performance. A discrete one-step ahead control scheme was adopted to determine control action based on the parameter estimation results. Experimental results showed that the new adaptive DO concentration control algorithm performed better than other algorithms tested, a PID controller and adaptive algorithms without the DO electrode dynamics.
In chapter 5, a novel automatic tuning of digital PID controller parameters has been developed for better regulation of the dissolved oxygen tension in fermentation processes. Heuristic reasoning allows the PID controller to reach tuning decisions based upon the supervision of certain control performance indices in the same cognitive manner as in expert control. In an experimental study, the output and the input covariances, the offset, and the oscillation period of the response curve were used as the performance indices for the auto-tuning, which is aimed to compensate for changes in the process dynamics. They were calculated on-line using a moving window. A set of several knowledge-based rules were applied to determine the optimal value of the proportional gain and the integral and the derivative time constants based on the values of the performance indices. Experimental results indicated that the auto-tuning PID controller developed in the present study worked extremely well with no necessity for an initial tuning step of the controller gain and no man-machine interactive scheme which consists the major part of conventional auto-tuning PID control.
