INTERVAL ARITHMETIC BASED ON GRAY MODEL FOR FAULTS DIAGNOSIS

Due to the difficulties in achieving an accurate characterization of the facilities playing important roles inoperation, maintenance is always implemented based onengin-eer's judgments and experiences. Applications of pattern recognition or pattern classification aided by fuzzy membership help to strengthen the validity of diagnosis by accumulation of experiences provided by experts involved. However numerical instances have shown that those models always result in wrong diagnosis, especially when the symptom aimed is vague oruncertain to determine whether it can be described as 揾igh?or 搇ow? or in another way, whether 1% or 10% relativeincrement can be definitely treated as increscent. Thepreviously applied 0-1 logic fuzzy membership has made these clogs, which prevent existing models from widely practicalapplica-tions. Should the input parameters and computingarithmetic be tailored to owe the ability of describing thevagueness and uncertainty, the information will be utilizedadequately to make the model more beneficial. Intervalarithmetic is suitable here. Interval numbers replaces the 0-1 logical inputs while the arithmetic is amended to gray distance judgment criteria. In another way the former neural network based fault diagnosis model has been improved on a grayinterval neural network. With these new models presented in this paper, misidentification caused by vague inputs andresem-bling membership would be eliminated to a lower level and result in a more precise diagnosis.