Título: Ensemble of Artificial Neural Networks and AutoML for Predicting Steel Properties
Autores: Krigor R. R. da Silva, Pedro Serpa, Douglas Macedo Sgrott, Fabricio Moreira Cerqueira, Fabiano Miranda, Jose Francisco Silva Filho, Rafael Stubs Parpinelli
Resumo: The design of new steel grades is a continuous pursuit in the metallurgical industry, aiming to develop lighter and stronger materials for diverse industries. This study explores the use of an ensemble of artificial neural networks, named EANN, to model the relationships between chemical composition, process parameters, and mechanical properties of five types of steels. Auto-Keras, an automated machine learning framework, is employed to identify the best configurations for the ANNs and create the ensemble model. The best ANN, named B-ANN, obtained through the same AutoML process, and the one model ANN generated by Auto-Keras, named ANN, are used as a benchmark to evaluate the E-ANN performance. The results obtained show that the E-ANN model is competitive in terms of predictive capability. Sensitivity analysis provides insights into the influence of input parameters on mechanical properties, while Shapley Value analysis highlights the relative importance of these parameters. The findings contribute to the understanding of steel behavior and provide guidance for steel design processes. This study demonstrates the effectiveness of E-ANN in predicting mechanical properties and emphasizes the value of data-driven techniques and automated machine learning in steel design
Palavras-chave: Steel, Prediction of mechanical properties, Datadriven model, Automated Machine Learning
Páginas: 7
Código DOI: 10.21528/CBIC2023-149
Artigo em pdf: CBIC_2023_paper149.pdf
Arquivo BibTeX: CBIC_2023_149.bib