Model free fault detection using one set of reference input and output data

This paper proposes a method for fault detection of control plants without using plant models. So far, mainly, there exist two methods to detect faults. The first is a statistical method, which uses statistical characteristics of input signals and fault signals. To extract the statistical characteristics, this method requires sufficiently fluctuated signals, which is not satisfied in practical applications. The second method is to use a model, such as to estimate an output in normal state of the plant using an observer and compares the estimated output to a measured output of the plant to be detected. But usually plant models are not available in industry. Hence, a method for fault detection without using signals of sufficient richness and also without using plant models is required. The method in this paper does not require sufficient richness of signals and plant models, hence it is practically useful. The method uses a pair of input and output data of a normal state as reference data instead of using a model. From the data, the method derives an impulse response sequence and compares the output estimated from the sequence with the measured output and detects a fault from the difference of the outputs. A simulation example of a water level control experimental plant is given to show the effectiveness of the method.