The Development of a Real-Time Process Control Batch/Semi-Batch Expert Advisory System

During the past 30 years, over 180 runaway incidents in industrial batch/semi-batch reactors have been ° reported to the Health and Safety Executive. Barton and Nolan?! 1987. 198°. 19°) have analyzed a large number of such incidents. This data provides an insight into the typical problems relating to reactor plant operation, which can lead to serious situations occurring. In the recorded incidents, there were eleven major chemical unit processes involved, with polymerisation featuring in a large number of the incidents. Almost 20% of the polymerisation incidents involved Phenol-formaldehyde condensations. As a direct result of these incidents in the earlier years featured in the analysis, the British Plastics Federation®"’°” published a set of guidelines for the industry; "Guidance for the safe production of phenolic resins". Despite these and other measures, further incidents involving exothermic runaway chemical reactions in batch reactors have continued to occur. The causes of such incidents include; inadequate instrumentation, poor process instructions, reactor control failures and catalyst overcharging. Errors due to inadequate operator control and training are also apparent. Given the complex nature of batch reactor operations, and the threat of runaway, it appears that process operators would benefit from the availability of a diagnostic expert system, which could act as a consultant/tutor to locate the causes of observed process malfunctions and propose the appropriate corrective actions. The application of expert system technology to fault diagnosis can improve the quality of a process operator's diagnostic and trouble-shooting skills. Thus the system can act in an advisory capacity to inter alia: increase safe operation; improve the education of process operators; reduce reliance on emergency relief systems e.g. venting; improve product yield; improve process efficiency; reduce by-products and reduce process down-time. The development of computer based tools, to assist process plant operators in their task of fault diagnosis, has received much attention over the last twenty-five years. However, very few of these techniques have ever reached the prototype stage of actual experimenting with application to process plants. The reason may be due to problems which arise during the implementation of the new technologies required. These concern : (i) the capability to respond in real time; (ii) the ability to communicate with existing control systems and (iii) the Human Computer Interfaces for both the plant operator and the programmer. Problems also occur in the development of systems. These include : (i) the time and effort required to understand and model the fault propagation behaviour of each considered process; (ii) the efficiency of the knowledge based system and (iii) the ability to schedule activities. Studies into a number of well established and widely used hazard assessment methods, show that, a combination of HAZOP (and action error) and Fault-tree analysis can provide adequate information in a structured format, for the development of a knowledge based system (KBS) of an expert system shell. HAZOP studies provide information on the areas prone to major failures, as well as the effect of a deviation from intent, while Fault-trees provide a way of mapping the propagation of faults leading to major failures, and with a diagrammatic representation such trees provide a visual logic, that is easily traceable. Together with "COGSYS" a high level expert system shell, it has been possible to implement a process control strategy with real-time capability. This project investigates the feasibility of using an experimental knowledge based system in the process control of multi-purpose batch/semi-batch reactors, with the capability to monitor the status of process equipment in real-time, both to aid and to teach the operators in the running of the plant, and provide early detection of fault conditions. The system will: i. Attempt to attribute an observed problem in the process to : operator error; equipment failure and system disturbance ii. Offer prescriptive remedies to restore the process to normal operation, i.e. propose corrective actions upon discovery of a malfunction in the plant iii. Retain all knowledge in a structured manner. In an attempt to minimise the modelling effort, the KBS has been designed to infer a diagnosis using a HAZOP and Fault-tree representation of the behaviour of the process. The KBS itself was embedded in the real-time process control expert system shell - "COGSYS"©O°S¥S0™), A batch process, the manufacture of Phenol-Formaldehyde resins, was selected as an illustrative example process.