Engine Fault Diagnosis Method Based on Convolution Neural Network and BiLSTM

Aiming at the complex characteristics of engine fault and fuzzy boundary of fault area, a method of engine fault diagnosis based on convolution neural network and bidirectional LSTM is proposed. To explore the causes of common engine faults, a convolutional neural network feature extraction model is established, the modified linear unit is used to improve the over fitting problem of the model, and the SoftMax classifier is used to lock the fault area, and the long short-term is established Memory rural network model, the introduction of remaining life tag, using min max method to complete data preprocessing, adding a dropout layer in the bidirectional LSTM layer and full connection layer, removing the output value of hidden layer, obtaining the final fault diagnosis results, and using root mean square error and asymmetric score function to evaluate the reliability of the method. The simulation results show that the fault diagnosis has fast convergence rate, high diagnosis accuracy and strong applicability, which provides valuable technical reference for engine maintenance and maintenance.

I. Introduction

The engine is the core component of the automobile, how the performance of the engine will directly affect the use of the car, the other performance of the car and the life of the car. The engine of the automobile works in a less stable environment, the structure of the whole engine is very complex, and with the use of the automobile, the engine is always in the condition of high -speed operation and work. Some parts in the engine also need to be challenged in high pressure, high temperature and impact environment. According to the previous investigation, we found that the engine fault accounts for the highest proportion of the whole automobile fault, so it is of great significance to diagnose the automobile engine fault problem and put forward effective solutions. The problem of engine failure is very important [1]–[2].