Abstract
In the field of PHM (Prognostic and Health Management), HI (Health Indicator) play a very important pole. It can not only reflect the health status of the machine in real time, but also provide some help for RUL (Remaining Useful Life) prediction. At present, HI are often constructed by statistical methods, which require a certain amount of expert experience and cannot mine deep features in the signal. Therefore, this paper uses an unsupervised method, CAE (Convolutional Autoencoder), to extract the deep features in the signal. Then, use the criteria of the monotonicity, trend, autocorrelation to perform feature sorting and feature selection, and input the selected features into FC (the Fully Connected neural network) for regression training, after which HI can be got. The experimental results show that, compared with traditionally statistical features, the deep features extracted by CAE can construct better HI.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
