The fault diagnosis of HAGC （Hydraulic Gauge Control） system of strip rolling mill is researched. Taking the advantage of the accompanying characteristics of the closed loop control system, rolling force forecasting model is built based on neural networks. The comparison results of the prediction and the actual signal are taken as residual signals. Wavelet transform is used to obtain the components of high and low frequency of the residual signal. Wave let decomposition results make fault feature clear and time-domain positioning accurately. Fault numerical criterion is established through Lipschitz exponent. By analyzing the varied fault features which correspond to varied fault rea sons, the fault diagnosis of HAGC system is implemented successfully.
By building mathematical model for HAGC （hydraulic automation gauge control） system of strip rolling mill, treating faults as unknown inputs induced by model uncertainty, and analyzing fault direction, an unknown input fault diagnosis observer group was designed. Fault detection and isolation were realized through making ob- server residuals robust to specific faults but sensitive to other faults. Sufficient existence conditions and design of the observers were given in detail. Diagnosis observer parameters for servo-valve, cylinder, roller and body rolling mill were obtained resoectively. The effectiveness of this diagnosis method was oroved bv actual data simulations.
