Supervising a DCS - Controlled Batch Process

I n the past half century, developments in control theory have been successfully applied to the lower level control of industrial processes [1]-[5]. In contrast with conventional machine control, however, industrial process control requirements concern product performance instead of the accuracy of the lower level control loop. The process performance is an overall index, including technical and economic specifications, which are often difficult for the lower level control system to handle directly and solely. Higher level supervision is often required to adjust the process control system from time to time. Since many industrial processes are of a complex nature (e.g., highly nonlinear, seriously coupled, higher order, and time varying), it is difficult to develop a closed-loop control model for this higher level supervision [6], [7], even though some advances have been achieved in hybrid control design theory [8]. Thus, the human operator is often required to provide online adjustment, which makes the process performance greatly dependent on the experience of the individual operator. Furthermore, process control has become a case-by-case phenomenon [9], [10] and has less referential value. It would be extremely useful if some kind of systematic methodology can be developed for the process control model that is suited to one kind of industrial process. Traditionally, there are two ways to develop process control methods. The first is the knowledge-driven (white-box) method. In this method, mechanistic models are derived from physicochemical laws, which can include qualitative information in the form of expert and/or linguistic knowledge. This kind of method is well suited to a wide range of process operations [11] and is still being widely used in industry, but its case-by-case characteristics can make its application difficult and time consuming for industrially relevant control systems. The second is the datadriven (blackbox) method motivated by development of identification and intelligent techniques. Linear and nonlinear ARMAX-type models are widely used to model dynamic systems [12], and intelligent methods such as fuzzy logic [13], neural networks [14], genetic algorithms [15], rule-based expert systems [16], and even multivariate statistical partial least squares [17] have been proposed to model the dynamics of industrial processes. However, this type of data-driven model requires extensive data or experience and may not be suited for newly