Estimation for Remaining Useful Life Based on a Complex Context Aggregation Model

Abstract

Deep learning is wildly used in remaining useful life estimation of mechanical equipment. However, existing methods couldn't avoid losing useful information during the process of extracting feature. In order to extract rich feature from limited data, we proposed a prognostic model using residual network and dilated convolution to aggregat complex contextual information during training. Furthermore, time-frequency analysis is also utilized in our method to combine useful information in frequency and time domain. Experimental results represented that our method makes better results on remaining useful life estimation over other methods using deep learning.