Fault Diagnosis And Reconstruction Of Engine Sensor Based On Neural Network

The sensor plays an important role in the engine’s control system,and the engine’s operating conditions change greatly during the entire operation of the aircraft.During this period,faults will inevitably occur.Compared with other components,the sensor is more prone to failure.In order to reduce engine flight accidents,while guiding the maintenance work of the aircraft,must improve the sensor fault diagnosis capability of the aero engine,and in order to improve the fault tolerance of the control system to ensure flight safety,signal reconstruction of the faulty sensor is required.Because this paper is mainly based on the experimental data of the experimental platform DGEN380.The neural network method has the characteristics of strong data processing capability.So,this paper based on the neural network method for the diagnosis and signal reconstruction of the speed sensor.The specific research contents are as follows:(1)The working principle of the sensor and the cause of the fault and the method of fault diagnosis of the sensor were analyzed,which laid a foundation for the signal prediction and fault diagnosis of the high-speed shaft speed sensor.(2)Used a data-based approach to create an engine model to meet the needs of signal reconstruction.According to the characteristics of data modeling and engine,we chose to use neural network to establish the important parameter model of the engine.In the research process,we only studied the correlation between the input and output of the system,without studying the internal principle of the system.(3)The fault sensor was predicted by the normal sensor,and the aeronautical engine speed sensor signal prediction space model was established.The BP,PSO-BP and IPSO-BP networks were used for modeling,and the modeling accuracy was compared.Then,used the historical data of the high-speed sensor to establish the time model,and used BP,PSO-BP and IPSO-BP network modeling,and compared the modeling accuracy.(4)Selected wavelet decomposition to quickly pre-diagnose the engine sensor,find the normal sensor and the sensor that may be faulty.A normal sensor was used to signal the sensor that may be abnormal through the neural network,and the actual output of the sensor and the neural network output were used to generate a residual and the magnitude of the threshold was set to determine whether there was a fault.According to expert rules,it was possible to determine all six types of faults that may occur in an aeroengine.(5)In terms of signal reconstruction,the characteristics and accuracy of the engine model,the space model and the time model were compared and analyzed.The results showed that the accuracy of the time model was higher than that of the space model and the engine model.