Research On Fault Detection Methods Of Aircraft Engine Based On Class Imbalance Learning

Because the fault detection of aircraft engine is suitable for class imbalance learning,this article establishs algorithms of one-class support vector machine and least squares support vector machine(LSSVM)based on class imbalance learning.Furthermore,in order to meet the requirements of the robustness and the real time of fault detection schemes,the relevant studies are done in enhancing the robustness of the algorithms for outliers and noise,the real time of the algorithms and so on.The main arrangements are as follows:First of all,one-class support vector machine is established through the idea of one-class classification in class imbalance learning.In order to solve its sensitivity to outliers,a kind of one-class support vector machine based on rescaled hinge loss function is constructed.Different from the hinge loss function,the rescaled hinge loss function is a boundness function that decreases the loss caused by the outliers.In addition,from the perspective of assigned weights,the rescaled hinge loss function improves the robustness of the algorithm to outliers.Experimental results show that the algorithm effectively ensures the performance of fault detection of aircraft engine and improves the robustness.Secondly,LSSVM effectively deals with the classification of balanced datasets.However,the performance of the algorithm is poor when dealing with the class imbalance learning(CIL)problem,such as fault detection of aircraft engine.In order to solve the problem,we introduce the LSSVM for CIL(LSSVM-CIL),which assigns different regularization parameters to the misclassification costs.As a result,LSSVM-CIL enhances the performance of fault detection of aircraft engine.Furthermore,similar to the LSSVM,LSSVM-CIL lowers the computational complexity by solving the linear equations,but losses the parsimoniousness.As a result,it is not conducive to meeting the real time of fault detection of aircraft engine.Hence,utilizing the method of combining iterative and reduced strategies,the number of support vectors selected by this method is small.Then,after realizing the parsimoniousness,the testing time decreases.As a result,the real time is enhanced.Besides,solving the inverses of matrices via Cholesky factorization,the algorithm enhances the stability of fault detection of aircraft engine.At the same time,it maintains the real time.Finally,there exist outliers and noise in the engine datasets.Thus,we propose a scheme combining fuzzy membership function and LSSVM-CIL(FLSSVMCIL),which can effectively solve the problem and enhance the robustness.In addition,the classifiers established by ensemble learning perform better than weak classifier.By combining the strategy of Bagging,the novel algorithm called BFLSSVMCIL further enhances the performance.The experimental results show that the algorithm that has excellent performance is suitable for fault detection of aircraft engine.