Fault Diagnosis of Electromagnetic Valve Based on Bayesian Optimization and Two‐Layer Long Short‐Term Memory Neural Network

Abstract

Electromagnetic valves are widely utilized in electrohydraulic control systems because of simple structure, low cost, and fast response. Accurate diagnosis of electromagnetic valve faults is crucial to ensure high effective operation of the control systems. This paper proposes a faults diagnosis model of electromagnetic valve based on Bayesian optimization (BO) and a two‐layer long short‐term memory (LSTM) neural network. First, to fully extract information features from the fault samples, a two‐layer LSTM network structure is used as the core of the electromagnetic valve fault diagnosis framework. Second, the BO algorithm is adopted to automatically adjust the hyperparameters of the two‐layer LSTM network model. The BO algorithm restricts the search space of hyperparameters in each iteration, which enhances the algorithm's efficiency. Moreover, owing to the limited availability of fault samples, cross‐recurrence quantification analysis is utilized for data preprocessing to strengthen and expand the fault characteristic data of the electromagnetic valve. This method effectively improves the diagnostic efficiency of the model. Finally, a real case study on electromagnetic valves is conducted to validate the performance and feasibility of our proposed method. The fault diagnosis accuracy of our proposed method reaches 94.29%, which is significantly higher than other fault diagnosis methods.