Multiconcept classification of diagnostic knowledge to manufacturing systems: analysis of incomplete data with continuous-valued attributes

The performance of a manufacturing system is largely dependent upon the con dition of its system components. By closely monitoring the condition of cri tical system components and carrying out timely system diagnosis as soon as a fault symptom is detected would help to reduce system down time as well as improving overall productivity. To achieve this, an effective diagnostic system is absolutely necessary. In recent years, computerized diagnostic s ystems such as knowledge-based systems have been developed to assist engine ers in performing system diagnosis. These computerized systems require suff icient knowledge to be acquired within a short time, which is not an easy t ask in reality, especially in the case of acquiring knowledge from imprecis e/incomplete data. Consequently, there is a need to look into ways to extra ct diagnostic rules from the raw information/data gleaned from a manufactur ing system in an efficient manner. The paper presents an approach that can extract diagnostic knowledge from incomplete data with continuous-valued at tributes. It begins with a brief discussion on the treatment of continuous- valued attributes for both twin-concept and multi-concept classification. S ubsequently, a detailed discussion on the treatment of incomplete informati on is presented. A case study is used to validate the application of the pr oposed approach. Results show that the rules induced are logical and quite consistent with those obtained from domain experts. The details of the case study and results are presented.