关键词: 空间滚动轴承, 寿命阶段识别, 双尺度柔性域感知模块, 同域泛原型学习, 两域原型迁移, 双分类器

Abstract: Aiming at the problem that the life state recognition accuracy of the existing life state recognition methods of space rolling bearings is low due to the large difference of sample distribution and the unbalanced number of samples in different life states under variable working conditions, a novel life state recognition method of space rolling bearings based on dual scale flexible prototype transfer network (DSFPTN) is proposed. In the proposed DSFPTN, a dual scale flexible domain sensing module is constructed and embedded in feature extractor to enhance the feature extractor to explore private features in different domains, thus improving the feature extractor to learn sample features in source and target domains of space rolling bearings; moreover, the same-domain imprecise prototype learning is designed to prevent the indiscriminate feature learning and incorrect clustering of cross domain samples, thus increasing the discrimination of heterogeneous samples in two domains; additionally, the two-domain prototype transfer mechanism is built to obtain the domain-invariant prototype and realize the transfer from source domain prototype to target domain prototype; finally, the dual classifiers after loading the domain-invariant prototype are used to align the distribution between two domains, and calculate the similarity between testing samples in target domain and domain-invariant prototype to complete the classification of testing samples in target domain of space rolling bearings, which can improve the recognition accuracy of various life state samples when the number of samples in different life states is unbalanced. The instances of life state recognition of space rolling bearings in space environment simulated on the ground verify the effectiveness of the proposed life state recognition method based on DSFPTN. To sum up, the constructions of the dual scale flexible domain sensing module, the same domain imprecise prototype, the two-domain prototype transfer mechanism, and the dual classifiers loaded with domain-invariant prototype make DSFPTN use only the unbalanced labeled samples in source domain of space rolling bearings to recognize the life states of testing samples in target domain with high accuracy under the large difference in sample distribution and the unbalanced number of samples in different life states.

Key words: space rolling bearings, life state recognition, dual scale flexible domain sensing module, same-domain imprecise prototype learning, two-domain prototype transfer, dual classifiers