Investigation of the effective parameters causing stickers in cold rolling mill steel by knowledge discovery from process data

Improving the quality of products by optimizing production procedures is a continuous goal for all industrial-manufacturing companies. Reducing waste and defective products through controlling factors in production processes is always a goal of managers’ steel companies. One of the surface defects appears in steel coils is sticker issue, which is a kind of layers welding that occurs in the production line during the opening of the annealed coil when the force required to open the coil is greater than the Surrender Strength. To reduce the stickers’ issue, operational data of the production process of cold rolling analyzed and presented in a fashion that reflects the effective features create this defect. For this purpose, data mining techniques can be exploited, because these techniques can retrieve knowledge and extract operational rules from a dataset. The dataset under examine was analyzed using neural network, logistic regression, support vector machines, and decision trees. Experimental results reveal the highest accuracy is related to decision tree C5.0. Accordingly, from this decision tree, rules were extracted which were checked and controlled by the experts. The results showed that the use of data mining to analyze the parameters affecting sticker defects can lead to improved quality, because this approach can be used to adjust the operational parameters of the manufacturing processes.