Fault Diagnosis For Variable Cycle Engine Gas Path:a Deep Learning Approach

As the core component of the advanced civilian and military aircraft in the future,the stable and reliable operation of Variable Cycle Engine is an important guarantee for the flight safety of an aircraft.However,due to the harsh working environment,Variable Cycle Engine will inevitably suffer from various kinds of faults,and the failure of gas path components is the main type of failure of the Variable Cycle Engine.Therefore,it is of great practical significance to diagnose the gas path fault of Variable Cycle Engine and to repair or replace the faulty parts in time.Compared with the traditional model-based fault diagnosis method,the data-based method avoids the problem caused by the inaccuracy of the Variable Cycle Engine model,and uses the data collected during engine operation to diagnose the fault directly.Traditional data-based algorithms such as BP neural network and support vector machine,which are often used for aero-engine fault diagnosis,rely on the features extracted manually,and there are still some problems such as over-fitting and parameter selection,which lead to the unsatisfactory effect of fault diagnosis.The deep learning algorithm developed rapidly in recent years,with its powerful ability of automatic feature extraction,effectively avoids the cumbersome process of extracting features manually,and has achieved extensive application and great breakthroughs in various fields.Combined with the characteristics of Variable Cycle Engine gas path faults,this paper mainly attempts to apply the deep learning algorithm to the gas path fault diagnosis of Variable Cycle Engine.In this paper,the gas path sensitivity analysis method is used to select the appropriate gas path measurement parameters,and then the collected Variable Cycle Engine fault data is subjected to pre-processing such as abnormal data elimination,smoothing and normalization.The shallow neural network,convolutional neural network(CNN),and bidirectional long-term memory network(BiLSTM)were used to diagnose the gas path of the Variable Cycle Engine.After comparing the characteristics of CNN and BiLSTM,this paper proposes the CNN-BiLSTM network,which combines the advantages of CNN and BiLSTM networks,has the local feature extraction capability of CNN network and the time feature extraction capability of BiLSTM network.It shows that the CNN-BiLSTM network has a good diagnostic capability for Variable Cycle Engine gas path faults.On this basis,in order to improve the real-time performance of the fault diagnosis system,this paper optimizes the CNN-BiLSTM network,uses the KPCA method to reduce the dimensionality of the data,and uses the adaptive momentum method(Adam)to train the network.The experiment proves that the strategy can greatly improve the learning and testing speed of the network on the basis of ensuring the accuracy of the fault diagnosis.It has certain practical value and can be applied to the Variable Cycle Engine gas path fault diagnosis system.