Convolutional AutoEncoder and Bidirectional Long Short-Term Memory to Estimate Remaining Useful Life for Condition Based Maintenance

Abstract

Intelligent prognosis and Condition-Based Maintenance (CBM) strategies accurately estimate the Remaining Useful Life (RUL). To achieve this task, various kinds of neural networks have been applied. Bidirectional long short-term memories were preferred for their ability to identify patterns of temporal sequences independently, while Convolutional Autoencoders were used in extracting features. In an attempt to combine the advantages of these two types of deep learning, a hybrid model is proposed in this paper. Based on these two networks, this study tries to find the best alternative by testing several hyperpa-rameters and studying their effects on the model performances. Finally, to study the approach efficiency, it is compared with other similar models.