Artificial intelligence for the prediction of tensile properties by using microstructural parameters in high strength steels
Im Doo Jung,Da Seul Shin,Doohee Kim,Jungsub Lee,Min Sik Lee,Hye Jin Son,N.S. Reddy,Moobum Kim,Seung Ki Moon,Kyung Tae Kim,Ji-Hun Yu,Sangshik Kim,Seong Jin Park,Hyokyung Sung
DOI: https://doi.org/10.1016/j.mtla.2020.100699
2020-06-01
Materialia
Abstract:Artificial intelligence is widely employed in metallurgy for its ability to solve complex phenomena, which are associated with the learning process of previously obtained experimental data. Although numerous physical modeling techniques have been implemented for the prediction of mechanical strength using equations, several empirical efforts are necessitated to evaluate specific constants for different models. To address this issue, numerous recent studies have employed artificial neural networks for the prediction of mechanical properties based on the material composition and process conditions; however, majority of these works have been limited applications due to the extensive number of input parameter combinations of chemical compositions. Microstructure is a good feature to understand mechanical properties because it incorporates the effects of material composition and process conditions. The complex combination of material composition and process parameters determines the microstructure of steel. In this study, the information on microstructural volume fraction is utilized for the prediction of tensile strength, yield strength, and yield ratio via artificial neural networking. Various combinations of PF (polygonal ferrite), AF (acicular ferrite), GB (granular bainite), BF (bainitic ferrite), and M (martensite) are investigated for the prediction of yield strength, ultimate tensile strength, and yield of high strength steel via back-propagation linear regression and neural network based algorithm. The effects of each microstructure on the three mechanical properties were successfully predicted by employing back-propagation linear regression. A deep learning algorithm with hyper-parameter tuning and cross-validation enabled high accuracy in predicting experimental data with mean absolute percentage errors of 6.59% and 10.78% for the validation and test sets, respectively. These studies can open a new avenue for applying the microstructural design effects to find optimum yield strength, tensile strength, and yield ratio of high strength steel.
English Else