Multi-mode acid concentration prediction models of cold-rolled strip steel pickling process

In the steel strip pickling process, the control of the acid concentration is an important part for ensuring the strip surface quality. Now only X-ray method is used to detect the acid concentration online, but the price is very high and the maintenance is very hard. The acid concentration is not measured in most of the steel strip pickling lines online. In this paper, a soft measurement of acid concentration is developed. The pressure differential, conductivity and temperature are used to calculate the acid concentration including ferrous chloride (FeCl2) and hydrogen chloride (HCl) concentration. The real pickling process is under a multi-mode condition. First, the spectral clustering based on geodesic distance is used to cluster the data into some groups. There are clearly linear relationship between the condition variables and the acid concentration. Then, orthogonal signal correction-iteratively reweighted least squares (OSC-IRLS) models based on the cluster data set are built to predict the acid concentration. The real field data set from cold-rolled strip steel pickling process is used to validate the model. The results demonstrate that the clustering method can improve the prediction result.