Expert diagnostic systems for industrial plants: a case study in the aluminum industry

An expert system designed for diagnosis locates or identifies malfunctions within a biological, electronic, industrial or any other system. We claim BEST (Blackboard-based Expert System Toolkit) to be a very adequate environment for diagnostic tasks, since it provides natural means for simulating a human expert diagnostician´s behavior. The diagnostic function deals with the generation and evaluation hypotheses. Using gathered data (symptoms) and a "forward-chaining" control strategy, the diagnostician generates a hypothesis (possible diagnosis) and then, using "backward chaining", acquires more data (measurements, laboratory data, etc) and proves or rejects the hypothesis. Apart from a combined control strategy, BEST offers model-based reasoning and hypothetical reasoning, i.e. parallel exploration of different hypothetical diagnoses, which is a rather difficult task for a human diagnostician. An illustrative example of the BEST-based expert diagnostic system for a bauxite-ore mill unit in aluminum industry is described in the paper.