Research On Prediction Methods In Aeronautics PHM Based On Data Mining

become increasingly complicated, which brings new challenges to fault diagnosis and maintenance. As the development trend of maintenance, condition based maintenance(CBM) is a maintenance method facing the actual status and development trend of the equipment. As the basis of CBM, aviation prognostic and health management(PHM) technology can effectively enhance the maintenance capacity of aviation system, increase sorties, radically reduce maintenance expense, and effectively prevent training or combat task failure caused by catastrophic failure. Prognostic is the key technology of PHM, so it is of important theoretical significance and application value to research prognostic technology of aviation PHM.Data mining is a new research topic in the field of database system. It can analyze mass data and extract useful information from them to provide effective basis for decision-making. This dissertation is on the application of data mining prediction method in aviation PHM. Main work is as follows:Current status of failure rate prediction research in aviation PHM is analyzed. The traditional failure rate prediction models are often lack of precision, or need a lot of samples and typical probability distribution, which are not easily meted in practical work and limits its application. Two new failure rate combination forecast models, NNAG and NNCG are proposed according to the study of grey model and neural network model. The failure rates of Boeing flights in a certain airline company are forecasted using these two proposed methods. Experimental results show that these two models fully exert the advantages of gray model and neural network model, can get better prediction effects. They are effective and feasible for failure rate forecasting.Present situation of wear trend prediction research in aviation PHM is analyzed. Wear trend time series prediction model frame and the concrete implementation steps using SVM are proposed. Firstly model parameters are chosen with cut-and-try method, and the predicting results affected by each group of parameters are analyzed. Then SVM parameters selection method using ant colony algorithm is proposed, and the SVM forecasting model is build based on ant colony algorithm parameters optimization. Finally the iron elements time series of a engine is predicted using this model. Compared with neural network model, experimental result shows that the proposed model is of higher precision, and parameter optimization algorithm is accurate and efficient.