Prediction of continuous cooling transformation diagrams in steels using light gradient boosting and rule-based optimization

Continuous cooling transformation (CCT) diagram is an indispensable tool for new product development in the steel industry and it is traditionally plotted via extensive thermomechanical studies. To reduce the cycle time of product development, a machine-learning model to predict the CCT of a new grade (chemistry and processing parameters) has been developed in this work. The idea is to have a holistic predictive model suitable for integrated steel manufacturing plant, instead of having many models for different class of material compositions. A wide range of data set is prepared to this effect and explored with multiple regression-based machine-learning models. Light Gradient Boosting Machine (LGBM) demonstrates exceptional results amongst the rest, and it is topped off with a metallurgical constraint/correction-factor to obtain the final model. The model is validated against test data as well as against some well-established metallurgical correlations and is found to perform fairly well.

Automated summary

Continuous cooling transformation (CCT) diagram is a crucial tool in the steel industry for new product development and is traditionally obtained through extensive thermomechanical studies. In this work, a machine-learning model using Light Gradient Boosting Machine (LGBM) was developed to predict the CCT of a new steel grade, considering its chemistry and processing parameters. The model was validated against test data and well-established metallurgical correlations, demonstrating good performance.