Neural networks for modelling and fault detection of the inter-stand strip tension of a cold tandem mill

This paper deals with the multilayered approach of the high-order neural network applied in a robust fault detection scheme. To introduce dynamic properties in these networks, a dynamic high-order neural unit is presented. It is shown that these networks can approximate any function with less parameters than in the case of multi-layer perceptron neural network. Such networks have good modelling properties, which make them useful for designing residuals in fault detection of dynamic processes. A method of computing a variable threshold derived from the confidence interval prediction is applied in order to obtain robustness in the fault detection process. Application of these networks for system identification and robust fault detection of the inter-stand strip tension of a continuous five stands cold mill is presented in the final part.