Application of Deep Learning for Fault Diagnostic in Induction Machine’s Bearings

Recent developments in sensor technologies and advances in communication systems have resulted in deployment of a large number of sensors for collecting condition monitoring (CM) data in order to monitor health condition of a manufac-tring/industrial system. Efficient utilization of sensory data leads to highly accurate results in system fault diagnostics/prognostics. The exponential growth of CM data poses significant analytical challenges, due to their high variety, high dimensionality and high velocity rendering conventional health monitoring tools impractical. In this regard, the paper proposes a deep learning-based framework for fault diagnosis of an induction machine's bearing based on real data set provided by Case Western Reserve University bearing data center. In particular, we focus on deep bidirectional long short-term memory (BiD-LSTM) networks fed with raw signals for fault diagnosis to address drawbacks of conventional machine learning (ML) solutions such as support vector machines. A numerical example is provided to illustrate the complete procedure of the proposed framework, which shows the great potentials of the BiD-LSTM for detection of different types of the bearing fault with high accuracy. The effectiveness of the proposed model is demonstrated through a comparison with a recently developed deep neural network (DNN) that considers temporal coherence for the same data set. The results indicate that the proposed framework provides considerably improved performance in comparison to its counterparts.