Development of AI based analysis tools for online monitoring of steel-making process

Abstract

Mechanical properties of steel are closely monitored during the manufacturing process through various non-destructive techniques. The IMPOC device implements one of them and is used by steel manufacturers at several locations of the production line. It provides a fast electromagnetic measurement that is strongly correlated with the microstructural state of the steel, hence its use to check online the uniformity of the semi-final or final product. Tata Steel, CEA LIST and SSSA have worked together to analyse IMPOC datasets corresponding to various situations in terms of steel coils and process conditions. A data driven model has been built from field measurements using a specific neural network architecture. This model proves useful to investigate the complex relations between numerous process parameters and the IMPOC value. This communication presents the approach followed to build the data driven model using neural networks and to check its relevance. The relationship between the measurement on one side and the process parameters and material composition on the other side are then investigated by means of several sensitivity analysis and feature importance techniques.