Research On Engine Fault Prediction Method Based On Deep Learning

Aero-engines have always been the focus of aircraft safety assurance,so its accurate reliability research has great significance.The reliability of systems is getting higher than high with the application of safety concept and advanced technology,so the equipment failure data is difficult to obtain in a short time,which makes it difficult to implement fault prediction based on the traditional reliability method.At the same time,sensor technology,data acquisition methods,and databases have accumulated a large amount of state performance data for engine health management.Based on such performance degradation data,reliability analysis of engine systems and degradation fault prediction have become a trend.In this paper,the degradation data of the engine is preprocessed for first.The magnitude disturbance between performance parameters is eliminated in using data scaling method and the principal component analysis method is used for dimensionality eduction of the data.Then,the visualization of engine degraded data distribution is realized.On this basis,the data is tagged and time-serialized according to the remaining useful cycle information of degraded engines,and a deep learning model of time-serialized long short-term memory is constructed to predict the fault level of aero-engines.Then it is compared with three other prediction algorithms to verify the outstanding ability of the time-serialized long short-term memory model in dealing with aero-engine fault prediction problems.For engine samples whose failure level prediction results are severely faulty,the maximum maitenance time is determined by estimating the remaining life.By analyzing the decay mode of engines,the data corresponding to each sample is fitted into an exponential degradation model to construct a decay model library.The models in the decay model library are matched with the fault sample based on similarity principle,then the remaining life of the engine sample is estimated.Combined with engine principle and parameter trend analysis,the fault system is easy to find.At last,the failure mode and effect analysis is applied to summarize the fault cause,failure effect and maintenance decision.Since the fault is repaired,the fault prediction and health management of the engine need to be re-performed.