Aeroengine Condition Prediction Based On Information Fusion

The monitoring and prediction of the aeroengine performance parameters is thebasis for making maintenance plan of aeroengine. Recently, the single parameterprediction model is widely used by airline companies to monitor and predict itselfwhich ignores the correlation of different parameters, and the result of prediction cannot meet the requirement of airline companies gradually. For improving the predictionaccuracy of aeroengine performance parameters, the research of aeroengine conditionprediction technology and application based on information fusion is proposed with thefusion of some different performance parameters.Statistic and density method are taken to find out the abnormal point of aeroengineperformance parameters, based on which a method of denoising that combines the EMDand wavelet threshold denoising is proposed. The process of the method is that choosingthe IMFs by comparing the evaluation coefficient after decomposing the original signalwith EMD, then denoising every IMFs with wavelet threshold method, in the endrestructuring all the signal which is denoised. The denoising method is proved to beeffective through structuring the analog signal, for that reason the method is used todenoise the performance parameters of aeroengine.According to the nonlinear and unstable characteristic of aeroengine performanceparameters time series, mutual information method and Cao method are used to figureout the time delay and embedding dimension respectively, based on which the timeseries of aeroengine performance parameters are proved to have the characteristic ofchaos with the judgment method of the biggest Lyapunov index. After the phasereconstruction of performance parameters, a method of quantify the correlation betweendifferent parameters is proposed by the change of distance in the phases, which is thebasis for the prediction based on information fusion.Aiming at the low accuracy of prediction by single parameter, the prediction modelof BP neural network based on phase reconstruction is established. Some information ofsystem evolution is added by phase reconstruction. The parameters which have thegreater correlation are chosen to be the input of the model, and the input-output sampleis structured by the chaos prediction theory. The dimension of the input is reduced bythe method of PCA, which improve the efficiency of the neural network. Aiming at the choosing the appropriate number of the neuron in the hidden layer, a method combiningexperience formula and enumeration is proposed. What is more, the genetic algorithm isused to optimize the weights and thresholds of the prediction model, which improve theaccuracy and stability of the neural network. The prediction model based on informationfusion is proved to be more effective by comparing the prediction result of long andshort term with the model of single parameter prediction.According to the above research, facing the requirement of airline companies, anprediction software of aeroengine performance parameter based on information fusion isdeveloped, which can be used to denoise the performance parameters, analyze thecorrelation between different parameters and predict the performance parameter. It canprovide the decision support for airline companies to draw up maintenance plan.