Abstract Scope |
Many empirical methods and machine learning (ML) models have been developed to predict the as-quenched and tempered hardness of martensitic steel. These models often involve numerous variables, making them complex. However, in a production heat-treatment facility, only a few variables, such as alloy chemistry (particularly carbon content) and processing temperatures, significantly impact the outcome. This study aimed to develop an application using ML models to predict the hardness of quenched and tempered steel. A published Gaussian Process Regressor (GPR) model was employed. The chemistry of the scheduled lot was inputted, and a web interface was developed to utilize this data, adjusting the austenitizing and tempering parameters to achieve the desired hardness. The use of this web application, developed in JavaScript using TensorFlow for the GPR model, is expected to significantly reduce the guesswork during daily operations, aligning with our existing development stack at the production heat treatment facility. |