Study and prediction of the zinc coating weight inside a galvanizing line

Abstract

Inside a galvanization line, a thin layer of zinc is applied on a steel strip. The thickness of this deposit depends on the galvanized steel's application and must thus be controlled precisely. This master thesis studies and predicts the resulting zinc coating weight given the set of the line's parameters. Based on real data, multiple models using machine learning and deep learning algorithms were designed to infer the deposit. The best results were obtained with the Extremely Randomized Trees regressor and this model achieves to reach less than 1\% of relative error in its predictions. After that, the trained model was used to create a tool to study the sensitivity of the zinc coating weight when the parameters of the line are modified. A static and a dynamic analysis were implemented. The last one allows an operator to enter the different values of each parameter and receive the predicted coating weight associated to them.

Along this work, this master thesis also presents another problem which is the surface prediction. In this case, only the coating weight measures and their positions are given to the model. This one must then infer the profile of the deposit, the surface of the steel strip.