Data Science to Determine Mechanical Properties of Low Carbon Steel During In-Process Inspections

Abstract

Carbon steel is a widely used category of engineering metal, mainly due to its attractive mechanical and fabrication properties and low cost. The chemical composition, physical parameters, and mechanical properties of carbon steel are maintained as per the specified standards, and local steel should be complied with Sri Lankan Standard 375: 2009. Generally, the chemical composition is tested during melt stages, and mechanical properties are tested for finished products. Since it is necessary to ensure products comply with the standard, mechanical properties are tested during in-process inspections as well. When the results are not within the acceptable range, a considerable amount of production has to be rejected, causing a loss to the manufacturers. If the results of the in-process inspection are instant, it will help make suitable adjustments to process conditions and thereby prevent rejection of products, while reducing quality assurance costs, as well. Therefore, the objective of this study is to predict tensile properties with chemical composition, as input variables, to be used for in-process inspections. Forty mechanical test reports were collected from a steel manufacturing factory for 12 mm diameter, thermo-mechanically treated (TMT) steel bars. Each test report is of 15 samples from the respective batch, and consists of corresponding chemical composition and physical parameters. Multiple linear regression analysis was applied to each batch, using Statistical Package for the Social Sciences (SPSS) software, to predict yield strength (YS), ultimate tensile strength (UTS), elongation at break (EB) with carbon equivalent value (CEQ) and percentage of Sulphur as inputs. Relationships between variables were not significant, even though those relationships can be used to predict tensile properties. The predictions may not be reliable, due to the limited conditions of the study and assumptions made. It is therefore recommended to apply multivariate regression analysis or Artificial Neural Network (ANN) techniques, with other chemical elements, process temperature and water flow rate etc. also as input variables.