Fault Diagnosis of Hydraulic Support Control Circuit Based on Bidirectional Long Short Term Memory with Probabilistic Sparse Self Attention Network

In recent days, fault diagnosis in hydraulic support control circuits is essential for ensuring the resilient process of machines in industries such as construction and mining, where hydraulic systems are widely used. Previous researchers suggested various techniques but still there were issues such as valve malfunctions, pressure anomalies, and actuator failures, which were challenging to detect and predict through conventional monitoring techniques. Therefore, this research proposes a novel approach for fault diagnosis of hydraulic support control circuit based on Bidirectional Long Short Term Memory (Bi-LSTM) with Probabilistic Sparse Self-Attention Network (PSSAN) named as BiLSTM-PSSAN. Initially, the machinery faults data is collected from Collaborative Robot (cobot) dataset from University of California Irvine (UCI) machine learning repository. Then, Adaptive Synthetic sampling (ADASYN) is employed for balancing the collected data and improving overall system accuracy. Further Empirical Mode Decomposition (EMD) and Boruta algorithms are incorporated for extraction and selection optimal features. Finally, a hybrid model Bi-LSTM-PSSAN is incorporated for fault diagnosis of hydraulic support control circuit since they assess the temporal detect and enhance interpretability respectively. The proposed Bi-LSTM-PSSAN achieved greater results with an accuracy of 99.89% when compared to existing model namely Multi-Sensor Convolution Neural Network (MS-CNN).