Fault Diagnosis Method with Feature Fusion Module Based on BiLSTM

Aiming at the training samples provided by industrial big data are few. In this paper, a network with a new Feature Fusion Module (FFM) combined with two-way convolution and Bidirectional Long Short-term Memory (BiLSTM) model is introduced to diagnose and classify three common types of bearing faults. We employs two-way convolution to bolster network robustness, thus extracting features efficiently. High-order information can be further extracted by FFM. BiLSTM is used for time and space feature fusion. A series of experiments show that the network has good diagnostic efficiency in the case of small amount of data.