Multi-output sparse Gaussian process based fault detection for a variable displacement pump under random time-variant working conditions

The fault diagnosis of variable displacement axial piston pumps has attracted huge attention since they are the power source of the hydraulic system. A notable superiority of the variable displacement pump is the capability of changing the swashplate angle to meet the requirements of different system loads. In most current research on the fault diagnosis of the variable displacement pump, however, the fixed displacement is usually postulated in a single test. Actually, timevariant displacement arouses the complex dynamic response, as well as unanticipated system state variations, which may confuse the classifier. In this paper, the fault detection of a variable displacement pump under random time-variant working conditions is investigated for the first time. The unscented Kalman filter with unknown input is utilized to estimate the system state and calculate the residual. The residual dynamic is then modelled by the sparse variational Gaussian process regression model. The extreme function theory gives a suitable threshold for determining whether the sample is faulty or not. The experimental investigation examines five fault types while tracking the random time-variant signal. Results with different fault sizes and noise levels validate the effectiveness of the proposed method. The comparative study demonstrates the proposed method achieves superior classification performance by an optimal trade-off between fault sensitivity and false alarm rate.