Aeroengine Performance Parameters Prediction Using Information Fusion Based On Discreted Process Neural Network

The realization of aeroengine on-condition maintenance depends on accurate prediction of the aviation engine performance parameters. The technolo gy of traditional single parameter prediction is able to achieve the prediction of performance parameter to some extent. But its defects are not accurate enough, and the computing speed is not fast enough either. So it can no longer satisfy the parameters prediction for the mainstream airlines. In order to find the reason, in addition that the traditional BP neural network algorithm which the general forecasting system used has certain limitations, the ignoring of the relationship between the various performance parameters also causes a bottleneck in enhancing the predictive accuracy. This paper intends to select the appropriate optimization method to input parameter fusion, and to use a new discrete entrance of neural network prediction in establishing the model at the same time, using this way to predict its aviation engine performance parameters in the end.Firstly, the paper uses the method of density-based identifying outlier to identify the abnormal points of aeroengine performance parameter. In order to solve the problems in inconvenience and inaccuracy that occurred during its operation, it proposes a method of identifying outliers which is based on uncertain distance. Compared with the outliers based on uncertain distance, it can be found that the method is effective. In order to overcome the shortcomings of exponential smoothing, People select the smooth function called COMPASS R-R provide of on the performance parameters of time series smoothing to smooth the time series of performance parameters, then to verify the validity of this method and apply it to smoothing parameters of engine performance.By doing quantitative analysis of the correlation among aviation parameters of engine performance so as to perform fusion prediction, this paper proposes the method based on experimental results of three point progressive inquiry, which can quickly query the relevant maximum point of delay,then to calculate the relevant coefficient.In order to avoid the relevant analytical errors only caused by calculating the linear correlation coefficient, the paper will use related Detrended Cross theories(based on fractal theory) in nonlinear correlation quantitative analysis of aeroengine correlation between different performance parameters to calculate the cross-related Hurst Exponent then to determine the magnitude of nonlinear correlation among different parameters. By comparing the results of two methods, the larger relevant parameters is selected as the basis of the fusion prediction.It aims to solve the problem of slow operation, low precision, and changes in performance parameter that can not reacted very well in single-parameter prediction. After the reconstruction of inputting variables in the phase space and making the division of ACO and the analysis of principal component, it input the sample space of reducing dimensionality into the model of processing neural network in prediction and fusion. By comparing the predicted result with the result of comparison in Single inputted parameter, it proves that the validation of algorithm is more effective.According to the information provided by the airlines based on the theoretical research, under the adequate analysis of needs, it develops a software system of the prediction of aeroengine performance parameters, which has the function of pre-processing with a performance parameters, the correlated analysis of performance parameters, the fusion prediction of performance parameters and the management of database, and it provides references to strategic support to the aeroengine on-condition maintenance.