Research On Engine Fault Prediction Technology Based On Deep Learning

With the continuous improvement of people’s requirements for engine stability and safety,engine failure prediction technology has received extensive attention.There is a contradiction between safety and high costs in the long-term maintenance of engines.How to reduce maintenance costs while ensuring safety has become a difficult problem for companies.In the context of the era of big data,fault prediction driven by big data has become an effective solution to this problem.In this paper,the time-frequency domain analysis algorithm is used to decompose the data collected by the engine sensor,the decomposed data is used as the input data,and the deep learning related algorithm is used to realize the failure prediction of the engine.And complete maintenance management and fault warning.In this paper,two decomposition algorithms are respectively studied and compared,including variational mode algorithm and complementary set empirical mode decomposition.Experiments show that the variational modal algorithm has a more effective decomposition ability for non-stationary signals.Aiming at the problem that the variational modal algorithm needs to pre-set the decomposition parameters,an adaptive variational modal decomposition algorithm based on the information divergence and Bayesian optimization algorithm is studied.The Algorithm optimizes the iterative process of finding the optimal parameters.At the same time,a fault prediction method based on attention mechanism and hybrid neural network model is studied.The convolutional network with attention mechanism is used to extract important features,and then the two-way gated cyclic unit is used to predict the remaining service life of the engine,so as to realize the failure of the engine.predict.The method in this paper is verified by using the NASA public data set,and compared with the results of the baseline model,which verifies that the prediction results of the model in this paper have high accuracy.The error is reduced by an average of 11.7%,and the scoring function is reduced by an average of 9.9%.This paper builds a set of equipment management and maintenance support system based on fault prediction technology.The system has been running in an enterprise,and the system can effectively predict engine failures and provide a reliable reference for engine maintenance.