Fault severity recognition of aviation piston pump based on feature extraction of EEMD paving and optimized support vector regression model

Recognizing an early fault for aviation hydraulic pump and evaluating its size is essential in this industrial application. This paper proposes a new method which combines ensemble empirical mode decomposition (EEMD) paving and optimized support vector regression (SVR) model to detect faults and estimate the fault sizes of a piston pump. Different from other feature extraction methods in which the information of intrinsic mode functions (IMFs) is not being fully utilized, the collected pressure signals are first decomposed by EEMD, and then some useful IMFs are selected by calculating the correlation coefficients between the signals reconstructed by the chosen IMFs and the original signals. These selected IMFs are referred to as EEMD paving. Subsequently, some new fault features considering time domain, frequency domain, and time–frequency domain are extracted from the paving of EEMD. To acquire the most sensitive fault features, principal component analysis (PCA) is then employed to reduce the dimensionality of the original feature vectors. Finally, SVR model is constructed to identify different fault sizes of aviation pump. To achieve higher recognition accuracy, a new method combining genetic algorithm (GA) with grid search is adopted to optimize the parameters of the SVR model. The effectiveness of the proposed method is verified by two datasets collected from a test rig under different conditions. The results demonstrate that the fault features based on the proposed method can be used to characterize the pump fault severity more accurately, and the constructed SVR model has higher recognition accuracy and better prediction ability when compared with previously published methods. The proposed method can also be readily used in other industrial applications.