Abstract:  In the field of product reliability study, identifying the reliability life distribution of a product accurately and efficiently is a key to the result of reliability modeling. The traditional Support Vector Machine (SVM) has some defects, such as unclassifiable region in multiclass classification. Aiming at those defects, a method of identifying the reliability life distribution with Multi-class Fuzzy Support Vector Machine(M-FSVM) is proposed, establishing four common models of reliability life distribution including exponential distribution, normal distribution, lognormal distribution and Weibull distribution, and is confirmed by simulation. The results show that, this method can overcome the shortcomings of traditional SVM, identify common life distribution intelligently with high recognition rate, and is convenient for the engineering application.

Key words: Multi-class Fuzzy Support Vector Machine, Life distribution, Reliability, Pattern recognition