Fault Detection for Aircraft Turbofan Engine Using a Modified Moving Window KPCA

As a typical data-driven fault detection approach, the moving window kernel principal component analysis (MWKPCA) method has attracted attention for fault detection of turbofan engines considering the presence of component degradation, but the conventional MWKPCA method uses a fixed step to update the KPCA model periodically until anomaly data is detected, this will increase the amount of calculation. To address this weakness, a modified MWKPCA method is proposed based on an adaptive updating mechanism for the KPCA model in this study. To realize the capability of updating KPCA model adaptively, k-means clustering method is utilized to divide a certain amount of newly acquired sampling data and the same amount of oldest data in the current time window into two categories, and calculates the Mahalanobis distance of the two clustering centers. Then the distance is compared with a prescribed threshold to determine whether to update the KPCA model. The proposed method is applied to an illustrative case, and the fault detection results under normal condition and sensor faults condition show that compared with the conventional MWKPCA algorithm, the modified MWKPCA method does not weaken the ability of fault detection and has better performances in terms of computation efficiency than the conventional MWKPCA algorithm with a fixed moving step.