Toward super-clean bearing steel by a novel physical-data integrated design strategy

Abstract

The cleanliness of fabricated ingots is crucial for the quality and properties of bearing steel. To address this issue, a physical-data integrated design strategy was developed to optimize vacuum arc remelting (VAR) parameters, combining numerical simulation, machine learning (ML), and experimental validation. Initially, a multi-phase, multi-physics coupled model was developed to predict the movement and distribution of inclusions during the VAR process. Furthermore, five ML algorithms were utilized to predict the cleanliness assessment score (CAS) based on inclusion size and distribution data from various VAR processing parameters, with gradient boosting regression (GBR) showing the best performance. Finally, a systematic framework based on a genetic algorithm was proposed to select the optimal combination of CAS. Here, the ML-optimized processing parameters comprised current of 4255 A, helium pressure of 0.69 kPa, and melting rate of 2.5 kg/min. Intriguingly, the number density of small inclusions at the center of the ingot decreased by 58.2 % and that of large inclusions reduced by 13.3 %. This was mainly caused by the appropriate maximum flow velocity of 2.6–2.8 cm/s during the steady-state stage of the molten pool. This study highlights a common and novel method for fabricating bearing steel with other superalloys via a physical-data integrated strategy.

Keywords

• Vacuum arc remelting (VAR)
• Inclusions
• Multi-phase model
• Genetic algorithm (GA)
• Physical-data integrated design strategy