Remaining Useful Life Prediction of Machinery Equipment via Deep Learning Approach Based on Separable CNN and Bi-LSTM

Predictive maintenance occupies a significant role to drop the operation and maintenance costs in production systems. Remaining useful life (RUL) prediction is one of the most preferred tasks in predictive maintenance decisions. Recently, deep learning techniques are extensively employed to accurately and effectively predict remaining useful life (RUL) by examining the past deterioration data of machinery and equipment failures. In this study, a deep learning approach that includes multiple separable convolutional neural networks (CNN), a bidirectional long short-term memory (Bi-LSTM) and fully-connected layers (FCL) are proposed to ensure more effective predictive maintenance planning. Separable CNN layers are applied to learn the nonlinear and sophisticated dependencies from the raw degradation data while the Bi-LSTM layer is employed to capture the long-short temporal characteristics. Besides, the dropout method and L2 regularization are used in the training stage of the proposed deep learning approach to achieve more accurate learning. The effectiveness of the proposed approach is verified by the popular FEMTO-bearing dataset presented by NASA. Finally, it is aimed that the experimental results provide better prognostic prediction compared with the benchmark models.