This paper develops a model-based technique based on the Unscented Kalman Filter (UKF) for on-line condition monitoring, and applies it to the hydraulic system of an automated industrial fish processing machine. First an analytical model of the hydraulic system is developed and the system parameters are identified (determined). Then the developed UKF approach is implemented in the machine. The UKF employs an unscented transformation to select a minimal set of sample points around the mean, which are then propagated through nonlinear functions, from which the mean and covariance of the estimate are recovered. This approach is known to be more accurate for nonlinear systems. For experimental investigation of the performance of the approach, four common hydraulic faults are deliberately introduced into the machine. The four faults are external leakage in the two chambers of the hydraulic cylinder; internal leakage; and dry friction build-up at the moving support plate of the cutter carriage. Three levels of leakage are manually introduced to the system for each fault scenario using needle valves. The criteria that are considered in fault diagnosing are residual moving average of the errors, chamber pressures, and actuator characteristics. The experimental results show that the developed technique is able to accurately determine the fault conditions of the machine.