Research On Gas Path Fault Diagnosis Methond For Civil Aero-engine Based On Deep Belief Networks

At present,civil aero-engine fault diagnosis method often needs to combine the experience of engineers when dealing with complex engine data,resulting in low efficiency of the model and low diagnostic accuracy.Deep belief networks(DBN)is one of the typical deep learning models that can mine deep features of data.Therefore,based on the characteristics of engine data,this paper introduces DBN into the process of civil aero-engine fault diagnosis,and proposes three kinds of DBN-based civil aero-engine gas path fault diagnosis methods.According to the characteristics of engine data,a model based on DBN and support vector machine(SVM)for fault classification is proposed.The DBN-SVM model uses DBN to extract feature of data and uses SVM to classify the features..To further verify the effectiveness of the DBN-SVM model,experiments are performed using small deviations in engine performance parameters.The experimental results show that DBN-SVM model can be applied to the civil aero-engine gas path fault diagnosis process,and the fault diagnostic result is good.Based on the characteristics of Aircraft Communications Addressing and Reporting System(ACARS)data,a multi-dimensional time series information-driven civil aero-engine gas path fault diagnosis model is built.According to the engine customer notification report,ACARS data is collected.Considering the gross error and random noise of ACARS data,the Grubbs criterion and exponential smoothing method are used for data preprocessing.Then,considering that ACARS data belongs to multi-dimensional time series,this paper uses wavelet packet transformation and dynamic time warping algorithm to extract time series information and related information between parameters.The two parts of information are vectorized and input to the DBN-SVM model for fault classification.The experimental results show that the model can classify ACARS data and has a high classification accuracy.Based on the above research,the influence of the imbalance between the size of normal samples and fault samples is considered,combined with sampling technology and integrated learning technology,an unbalanced sample-driven civil aero-engine gas path fault diagnosis framework is proposed.By merging the previous chapter’s model into this framework,the point form unbalanced sample-driven civil aero-engine gas path fault diagnosis method and the sequence form unbalanced sample-driven civil aero-engine gas path fault diagnosis method can be obtained.The experimental results show that the combination of the previous model and the unbalanced sample-driven civil aero-engine fault diagnosis framework improves the overall classification accuracy of the sample s and the classification accuracy of the fault samples.Based on the civil aero-engine gas path fault diagnosis method studied in this paper,combined with the airline’s demand for engine gas path fault diagnosis method,the civil aero-engine gas path fault diagnosis prototype system is designed and developed to provide technical and system support for airlines to work on engine gas path fault diagnosis.