Research On Gas Path Data Mining Method In Aeroengine Condition Prognostics And Health Management

Maintenance after failure and time scheduled maintenance are out of fashion for aeroengines. Many defects of these two maintenance methods such as low efficiency, high maintenance costs, and weak guarantee for flight safety and reliability have been exposed in the practical application of aviation engineering. Compared with traditional maintenance methods, maintenance after failure or time scheduled maintenance have been changed into condition based maintenance by Aeroengine Prognostics and Health Management (EPHM). It could be possible for engineers to exactly place the points on the engine in particular time where the potential failures might happen, and then do active maintenance. EPHM will make contributions to enhancing maintaining efficiency, flight safety, aircraft reliability, and lowing maintenance cost.Taking the gas path system of Rolls-Royce Company’ Trent700engine as an example, in this article, data mining method, which is one of the key issue of EPHM, has been deeply studied. At the beginning of this research, based on the gas path system status data of this engine, information mining study has been carried out. Taking Turbine Gas Temperature (TGT) as an example, the baseline model of the status parameter, which could reflect engine gas path performance, has been made and tested. The test result showed that the baseline’s accuracy was enough, and the foundation of engine gas path condition monitoring has been laid. Then based on the gas path parameter deviations which were calculated in the monitoring process, the informations of engine gas path system’s performance degradation which were contained in these deviations also have been mined in this research. The regression forecast model which was based on Support Vector Machine (SVM) has been built for the five gas path parameters deviations’single point prediction. In addition, to develop these regression forecast models, Fuzzy Information Granulation Theory (FIG) has been tried to be incorporated into the SVM prediction model to make a new prediction model called Granular Support Vector Machines (GrSVM). A range of several future time series points’ variation could be forecast by GrSVM model. In the end, a simulation experiment was made to test and analyze the models’ performance. The test results of the single point prediction model base on SVM and the range prediction model base on GrSVM showed that these two models’ accuracies were satisfied, and this work could provide a reference for EPHM’s trend prediction research.