Leakage fault classification in hydraulic actuators via multiple trained transformations

The application of a specific signal-based method is described in this paper to detect internal leakage faults in hydraulic actuators. By implementing an artificial leakage with various degrees of severity, pressure signals at either chambers of a laboratory-based hydraulic actuator is gathered in response to a periodic square input. The gathered data are used to train a set of transformations that play the role of a combination of parallel filters. The number of trained filters is equal to the number of introduced classes for the signals. The detection of the class of a test signal is based on the comparison of the RMS value for the filtered versions using each trained filter with some determined threshold values. Using the proposed method, the ability to introduce arbitrary numbers of classes of fault is obtained without a need to explicitly model the dynamics of the hydraulic system.