Fault Diagnose Technology Of Civil Aero Engine Based On Artificial Neural Network

Aero engine is the important part of a civil aircraft and the maintenance cost of the aircraft accounts for 10% to 20% of the total cost.Therefore,the diagnosis and monitoring of the engine failure is the center of civil aircraft safety and economy.In the engine fault diagnosis,the engine is abstracted as a typical complex mechanical system,due to the complex structure of the engine,the non-linear model,diverse diagnostic methods,influence of flight factors and noise factors on fault diagnosis and other reasons,making engine fault diagnosis modeling difficult.Therefore,this paper mainly studies the problems of engine fault diagnosis,including the extraction of engine monitoring status data and the establishment of accurate fault diagnosis model to predict fault occurrence according to the monitoring state parameters.in order to improve the robustness of engine fault diagnosis,this paper designs a reliable monitoring state data extraction process,and then uses the artificial neural network to train the engine fault diagnosis model to predict the failure mode of unknown sample data.The main work is as follows:(1)Extract engine monitoring status data to build sample space.The data from ACARS equipment is not enough to support the engine failure of online training and real-time diagnosis,and QAR record information is complete and high frequency.For the QAR data frame structure,the flight parameters are mainly discrete,digital and analog three types.For these different air parameters,this paper designs the corresponding decoding algorithm to achieve the storage device binary raw data engineering translation.For the ACARS data frame structure,according to ARINC-620 protocol,this paper designs the class library used to describe the specific parameters,and then design the corresponding decoding function.(2)Preprocessing and multi-source fusion of sample data.The pretreatment of gas path parameters includes data smoothing and time registration.This paper increases the data and sample availability by using moving average smoothing.And when multi-source parameter processing is performed,the sampling frequency of the data frame is different,the observation data obtained by each channel and the symptom data are not synchronized.The cubic spline interpolation method is used to reconstruct the frame data,to get the interpolation point.(3)Using artificial neural network to diagnose engine faults.In order to improve the robustness of system fault diagnosis,the manual design network is optimized,and reasonable sampling space is constructed.Aiming at the failure of gas path parameters and mechanical failure,the individual neural network is established,and the optimal fault diagnosis model is given by sample training in the sample space.The test results show that the two individual neural networks can achieve high diagnostic accuracy and its real-time performance is good.In order to further improve the fault diagnosis effect,the adaptive weighting method is used to integrate multiple neural networks,and the output of all the neural networks is weighted averagely as the final fault diagnosis output,which further improves the fault diagnosis ability in the fault environment.