Health indicator construction and status assessment of rotating machinery by spatio-temporal fusion of multi-domain mixed features

Rotating machinery has been applied in various industries, and weak fault feature monitoring is of great significance to constructing health indicators (HIs) and assessing their status. However, there are some challenges in HI construction and status assessment, including difficult expression of weak features, incomplete information domain, and quantification of early degradation points. To construct a novel HI of rotating machinery, this paper proposes a multi-domain features-based spatio-temporal fusion method, which integrates the spatio-temporal advantages of self-attention (SA), long short-term memory (LSTM), and an improved convolutional autoencoder (ICAE), called SALICAE. On this basis, the sooty tern optimization algorithm (STOA) is used to automatically optimize the extreme gradient boosting model (XGBoost) for assessing the status of rotating machinery accurately. The effectiveness and adaptability of the proposed method are verified by the standard bearing database from Xi’an Jiaotong University, and the average accuracy under different working conditions is approximately 85.3%. Moreover, the accuracy of the proposed method is also tested by the reducer platform organized by our lab, which is 99.3%.