Machine learning-powered analysis of hot isostatic pressing for Ti-6Al-4 V powder

This study focuses on developing a machine learning (ML) model capable of predicting relative density and equivalent strain in samples produced through hot isostatic pressing (HIP) of Ti-6Al-4 V powders. The model is trained using data from numerical simulations, incorporating processing parameters and powder size and distribution as input features. Results demonstrate strong predictive performance, with R${}^2$ values of 0.951 and 0.911 for relative density and equivalent strain, respectively. The findings also reveal that the effectiveness of ML predictions is greatly influenced by the weight functions assigned to processing parameters as input features, while the impact of powder size and distribution weighting on optimal prediction is comparatively minimal. This suggests that particle behavior may demonstrate a higher level of consistency in response to the HIP process compared to the variability created by processing parameters. The outcomes of the ML predictions are further utilized to provide a detailed discussion on how variations in temperature, pressure, and powder size and distribution impact changes in relative density and equivalent strain in a specimen.