A method of neuro-fuzzy computing for effective fault diagnosis

The purpose of this paper is to present results that were obtained in fault diagnosis of an industrial process. The diagnosis algorithm is based on a three-layer neuro-fuzzy network theory. We present a new technique for the treatment of overlaps among adjoining fuzzy sets. Inputs of the network are the process I/O data, such as pressure and temperature, parameters estimated by EKF, and state values calculated by dynamic equations, while outputs of the network are process fault situations. The model combines the learning capabilities of neural networks with fuzzy computing and has been simulated on a digital computer, using MATLAB programming language and training data from a local process plant. The running test results show that the strategy appears to be better suited to diagnose faults of such an industrial process