I. Introduction
Condition monitoring of fluid power systems is earning more and more consideration to reduce the cost of maintenance and prevent the system from deteriorating further. Faults in fluid power systems and methods for detecting them have been documented in the book by Watton [1]. This paper focuses on the internal (cross-port) leakage in hydraulic actuators, which is caused by wear of the piston seal that closes the gap between the moveable piston and the cylinder wall. Internal leakage causes the hydraulic fluid to be displaced between the two chambers of the actuator, influencing the dynamic performance of the actuation, since the entire flow is not available to move the piston against the load. In general, internal leakage cannot be detected until the actuator seal is completely damaged and the actuator fails to respond to a control signal. Research on actuator internal leakage fault identification, in spite of its importance, is still limited. Tan and Sepehri [2] applied the Volterra nonlinear modeling concept to implement an online fault diagnosis scheme in hydraulic systems. By constructing a parametric space, actuator leakage faults were detected. The technique is similar to the work of Le et al. [3], who employed neural networks and dynamic feature extraction to classify the leakage type and level in hydraulic actuators. In both studies, systems with various internal leakage fault levels must be emulated (through simulations or via experiments). An and Sepehri [4] studied the feasibility of using an extended Kalman filter (EKF) to detect an actuator internal leakage fault while the actuator tracks a sinusoidal reference input. They further extended this work to include both friction and loading as unknown external disturbances [5]. Although the requirement of using a model for internal leakage was removed by An and Sepehri [4], [5], the need for knowing the model of the hydraulic actuator still remained a challenge. In order to overcome the difficulties associated with modeling nonlinear hydraulic systems, a linearized model with an adaptive threshold (to compensate for the error due to linearization) was used by Shi et al. [6] to detect an internal leakage fault.