A GMDH neural network-based approach to robust fault diagnosis: Application to the DAMADICS benchmark problem

This paper focuses on the problem of designing a robust fault detection scheme with group method of data handling (GMDH) neural networks. One of the main objectives is to show how to determine the structure and parameters of the neural network as well as how to estimate the modelling uncertainty of the resulting neural model. It is shown that the algorithms proposed to tackle the above tasks make it possible to obtain a suitably accurate mathematical description of the system. This description allows developing a technique of generating an adaptive threshold that permits robust fault detection. Another objective is related with an application-oriented study regarding the proposed approaches. In particular, the paper presents the results concerning the application of the techniques and algorithms for the modelling and fault detection of a valve actuator (DAMADICS benchmark).