Application of Machine Learning to predict the mechanical properties of high strength steel at elevated temperatures based on the chemical composition. (June 2023)

• Record Type: Journal Article
• Authors: Shaheen, Mohamed A.; Presswood, Rebecca; Afshan, Sheida
• Is Part Of: Structures. Volume 52(2023)
• Page Start: 17
• Page End: 29
• Publication Date: 2023-06
• Subjects: Constitutive model -- Elevated temperature -- High strength steel -- Machine Learning -- Chemical Composition -- Deep Learning; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2023.03.085
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: This paper presents the results obtained using Machine Learning (ML) algorithms to predict the mechanical properties, including ultimate tensile strength, yield strength, 0.2% proof strength and elastic modulus, of high strength steel plate material at elevated temperatures. High strength steels are increasingly used in several areas of construction offering efficient structural solutions with a high strength-to-weight ratio. Safe fire design of these structures relies heavily on accurate prediction of mechanical properties of the material with temperature. The data on elevated temperature mechanical properties collected from the literature experimental tests show a high degree of scatter, implying that they are influenced significantly by various factors, most notably the testing method, manufacturing process and chemical composition. However, the current methods for predicting the mechanical properties of high strength steels at elevated temperatures by using 'reduction factors' as adopted by the structural design codes do not consider these effects and may lead to inaccurate predictions. To overcome these deficiencies, a ML-based prediction method that uses temperature and chemical composition as input parameters is developed in this paper. Deep Neural Networks are trained and validated on the basis of elevated temperature material data collated from the literature test programmes. The analysis of the results show that the trained algorithm gives an excellent correlation coefficient with very small error value in predicting the strength and stiffness reduction factors of HSS.