A self-learning diagnosis system for grinding vibration.

Abstract

The diagnostic procedure for grinding vibration may be carried out by using the symptoms found on the ground surface, the observations made during grinding, and the results of previous diagnostic experiments. Research has been. conducted for the purpose of obtaining a systematic procedure for grinding vibration diagnosis in which a computer-based methodology featuring knowledge engineering is used. The system consists of the production rule type inference engine whereby diagnostic censors are used to rule out the impossible causes in parallel with the separate identification of alternative causes by diagnostic rules. For the cases when no cause can be induced by the rules, the system includes another method of inferring alternative causes by making comparison with prior successful diagnoses. It also includes a self-learning function which generalizes from positive examples and negative examples. This self-learning process can produce new rules and censors, and in addition the old rules can be maintained by the production of the new censors resulting in system autonomously improving its own performance. A case study and a series of simulated diagnoses have demonstrated the effectiveness of these functions.