Methodologies for a real-time intelligent supervisory system for a hot strip mill finisher

Abstract

This paper describes methodologies and the architecture used in a prototype intelligent supervisory system for hot strip finishing mills in steel manufacturing.The prototype system incorporates a knowledge-based supervisory layer in its top level. The supervisor gathers information from critical areas and warns the operator on abnormalities. The system takes advantage of an emerging artificial intelligence (AI) toolset in a virtually parallel processing environment and couples symbolic and numeric computation in real time. The toolset allows the user to construct a hierarchy of Autonomous Communicating Objects (ACOs) [10], each of which may have some intelligent supervisory function and communicate with each other.The supervisory system is a prototype system using a simulated plant. The system has two separate diagnostic modules which are activated by messages from the simulated plant The modules detect sensor failures using traditional signal processing schemes and inference by rules representing the knowledge of the process. A sensor is not considered to be failed until detected by both modules. This can solve some practical difficulties in implementing adequately sensitive fault diagnosis.The supervisory system has a result interpreter which analyzes the plant behavior when no serious sensor failure is detected. The interpreter consists of heuristic rules which pursue forward chaining, and subordinate modules for primitive diagnostic algorithms which can be executed concurrently. It has a fault propagation digraph for each physical perspective of the plant, a category, and invokes subordinate modules by supplying digraphs. Supervising these insufficiently informed but “functionally accurate” modules, the diagnostic module determines an essential category without which the current plant problems can not be explained. The combination of the rule-based deduction and the reachability matrix-based approach offers a practical methodology for the plant diagnosis.