关键词: 实时寿命预测, 性能退化, 小波支持向量回归机, 遗传算法

Abstract: Aiming at the fact that the present real-time lifetime prediction methods do not take full advantage of the same kind of products’ performance degradation data, as viewed from the comparability of degradation paths, a real-time lifetime prediction method was proposed on the basis of wavelet support vector regression (WSVR) optimized by genetic algorithm (GA). Firstly, GA-WSVR is employed to build the same kind of products’ performance degradation path models. Then the specific individual’s historical measure time vector is used as the benchmark, the same kind of products’ corresponded degradation measurement vectors are calculated using GA-WSVR models. The Euclid distances of the specific individual and the same kind of products are used to determine degree of membership, so the individual’s degradation path model is built on the basis of degree-of-membership weighted method. Finally, the measurement vectors, Euclid distances, degree of membership and degradation path model are updated with real-time measurement data. The proposed method is applied to fatigue crack growth data, the experimental results validate the validity.

Key words: real-time lifetime prediction, performance degradation, wavelet support vector regression(WSVR), genetic algorithm(GA)