Operating-Condition-Aware Fault Diagnosis of Variable-Speed Aviation Piston Pumps Using Interpretable Multi-Domain Features

This is the authors' abstract. We don't add key points for this paper.

In variable-speed aviation piston pumps, changes in rotational speed, flow demand, and load can shift fault-related spectral content and make fixed-frequency features sensitive to operating condition. This study evaluates an operating-condition-aware and physically interpretable diagnostic procedure using open experimental data from a nine-piston swashplate-type aviation piston pump. The dataset contains asynchronous pressure, flow, vibration, speed, and torque measurements recorded under healthy operation and induced wear faults affecting the servo piston, pressure regulation valve, and swashplate bearing bush. Pressure and vibration responses are analysed in non-overlapping five-second operating windows and represented by time-domain descriptors, operating-condition variables, approximate shaft-order features, and a limited set of wavelet-domain vibration indicators. The order-domain features normalize spectral content with respect to rotational speed and focus on mechanically meaningful components, including shaft-synchronous response, the 9X piston-passing order, harmonics, and sideband regions. To reduce file-level leakage, leave-one-file-out validation keeps all windows from the same experimental run in the same fold rather than splitting correlated windows at random. Under this grouped validation scheme, the fused feature set reached a macro-F1 score of 0.874 with a random forest classifier. The result suggests that combining rotation-normalized descriptors with operating-condition variables and selected non-stationary vibration indicators improves controlled fault-mode discrimination, while the small number of independent runs limits generalization beyond the bench-test setting.