Robust prediction and validation of as-built density of Ti-6Al-4V parts manufactured via selective laser melting using a machine learning approach

Abstract

It is well known that slight changes in selective laser melting (SLM) process parameters may alter the outcome of mechanical and physical properties of the as-built material in a drastic and haphazard fashion. To overcome this, reliable property prediction models are most pertinent. In this study, a machine learning approach based on Gaussian Process Regression (GPR) is proposed to predict the relative density of as-built Ti-6Al-4V alloy manufactured via SLM, based on the most common input process parameters such as laser power, scanning speed, hatch spacing, and layer thickness, as well as an integrated input of volumetric energy density. A most comprehensive test dataset to train and verify GPR models was retrieved from literature papers that extensively investigated mechanical and physical properties of additively manufactured Ti-6Al-4V alloy. GPR models with four different kernel functions were analyzed and exponential GPR model with optimized hyperparameters was chosen as the most viable model for predicting as-built density of Ti-6Al-4V alloy. A parametric multiple linear regression (MLR) model was also presented and serves as a benchmark. When inferences were made on newer publication data, the GPR model and the MLR model predicted the densities with mean absolute errors (MAE) of 1.12% and 5.22% respectively. The inferior performance of the MLR model compared emphasizes the need of non-parametric supervised learning technique for SLM. To truly demonstrate the effectiveness of the proposed GPR model in real-world metal AM jobs, 22 experimental samples were printed. Predictions made on all the samples, when compared to their actual density values, resulted in MAE of 0.27%. Clearly, creation of most comprehensive mined data, kernel selection, and rigorous validation and verification of GPR model make this study one of its kind and prove the GPR model's predictive dexterity and the potential impact in the world of additive manufacturing.