The defect on the surface of the hot-rolled strips is difficult to be observed especially on coiling, this research presents a prediction method for suchlike fault detection by an artificial intelligent instead of a complicated mechanical model. First, an autoregressive model is utilized to represent the multi-feedback signals from the coiler in hot strip mill. Then the coefficients of this autoregressive model are used to compute signals' power spectrum, which allows to eliminate the bias and variation resulted from the ordinary way. The combination of multi-way principal component analysis and power spectrum is proposed, which allows to identify the characteristics of normal and abnormal coils by the computation of T{sup}2 and squared prediction error. Finally, the proposed algorithms are implemented with coiling data of China Steel, which shows that the proposed method is able to detect the coils with surface defects.
展开▼
