Speaker
Description
As one of the largest production chains in the world, the steel industry faces an ever-increasing demand for higher level of functionality and quality of final products at reduced environmental impact and manufacturing cost. The steel industry has developed an extensive range of sensors to generate data, monitor, and control steelmaking processes. Despite these advances, issues remain in the areas of data collection, storage, migration, lack of through-process data links and erroneous datasets, all of which significantly increase the complexity of the quality process control. The development of data-driven approach through advanced artificial intelligence (AI) techniques offers the opportunity to practically implement machine learning techniques to such datasets aiming to provide processing-property optimisation and identify gaps and errors in the data.
Recently, computational capabilities and algorithmic developments have significantly grown in power and complexity, accelerating the progress of process optimisation and materials defect and property prediction. However, addressing large scale industrial data process-property optimisation strategies is challenging as it involves numerous influencing factors each possessing with insufficient data. Herein, an integrated data-driven steelmaking case study is attempted with the aim of predicting and optimising the performance of final products. The key variables have been identified across multiple process chains such as steelmaking, casting, hot rolling and heat treatment. Machine learning has been used collaboratively with metallurgical knowledge, first-principal calculation, and feedback into non-linear neural network models. The integration of data mining, and machine learning generate reasonable predictions and improve productivity of the steelmaking industry. Hence, this data-driven strategy is able to provide predictive capability of composition design, processing optimisation, emerging microstructure and property.