Research On Trend Prediction For Aeroengines In Durability Test Based On Data Mining

It’s been over100years since planes are produced and it’s already come into the jet aircraftera. Nowadays, for various new types of planes are emerged and more and more widely used indifferent places and in different tasks, the work environments of engines in planes obviouslybecome more and more complicated. With the development of aeroengines in higher pressure ratio,higher temperature and higher thrust-weight ratio, as a result, the working stress level ofcomponents of aeroengines need to improve greatly. Comparing with the old engines, the modernones adopt new materials, new technologies, but due to the complication of the structures ofaeroengines and the scurviness of work environments, such situations as premature failures and airaccidents sometimes happen. Therefore, researching trend prediction technologies for aeroengineshas important theoretical significance and applied values right now.The application of data mining in trend prediction for aeroengines is researched in this paperand the works and innovation in this paper are:Firstly, one combined predicting model of Grey model and particle swarm optimization (PSO)neural network is put forward. This model combines grey system with neural network and makesthe best use of the two techniques so as to solve complicated uncertainty problems. Moreover,PSO algorithm has faster convergence and is harder to entrap into local minimum compared withBP neural network. The experimental results of comparing the combined model with the singlemethod respectively in exhaust temperature prediction and vibration trend prediction show that theshort-term effect within3-step prediction of combined model is better than that of grey model orof neural network.Secondly, one fuzzy support vector machine algorithm based on PSO parameter optimizationis proposed in this paper. In the model, fuzzy support vector machine replaces standard supportvector machine and membership function is introduced into training samples to distinguish theweights of data samples in different periods to the model. Meanwhile, in order to solve theproblem of parameter determination for fuzzy support vector machine, particle swarmoptimization is used to improve the parameters of fuzzy support vector machine model and thenthe method is experimented into exhaust temperature prediction and vibration trend predictionrespectively. As a result, not only can fuzzy support vector machine effectively predict exhausttemperature and vibration trend, but also PSO parameter optimization technology can obviouslyimprove the prediction performance.Thirdly, one multi-parameter degradation evaluation model for aeroengine is also presentedin this paper. Firstly, grey relational analysis method is adopted to analyze10major parametersaeroengine condition monitors in the model and then the subsets which have high correlationdegrees to each other are worked out and then made in the same group so that the perfectparameter, the one can best represent the performance of aeroengine, can be chosen out from all ofthem while the rest ones are abandoned as well. Finally, support vector data description model isused to train the vector sets made up with several parameters of aeroengine in health states duringprimary stage and then the health state evaluation model is formed in the end. Moreover, theexperimental results indicate that the evaluation method is able to reflect the degradation evolutionprocess of aeroengines well.