Fault Diagnosis and Prediction of Hydraulic Systems Based on Bi Directional Recurrent Neural Network With Attention Long Short-Term Memory

Since years, hydraulic systems play vital role in modern industry, but failures result in substantial economic losses and downtime. Traditional approaches for fault diagnosis and prediction of hydraulic system had faced several challenges which include sensitive to noise, lack of temporal understanding and linearity assumption. Therefore, this research proposes Bi-directional Recurrent Neural Network- Attention Long Short-Term Memory (BRNN-ALSTM) for Fault diagnosis and prediction of hydraulic systems. Initially, the data is taken from condition monitoring of hydraulic systems dataset which consists of data collected from sensors like pressure, flow rate, temperature and vibration measurements of various sensors installed in hydraulic system. This data is preprocessed by using z-score normalization which enhanced the predictive capability and facilitated the model's optimization. Then, pre-processed data optimal features are selected by utilizing variational autoencoder that efficiently eliminated irrelevant attributes from the sensors data. The selected optimal features are classified by employed BRNN for diagnosing the faults in hydraulic system which effectively captured temporal dependencies and patterns in data and finally ALSTM is employed for predicting the faults. The proposed BRNN-ALSTM achieved better accuracy of 99.67% when compared with existing Multi Sensor Convolution Neural Network (MSCNN).