Prediction of Ultimate Tensile Strength of Heat-Treated Additively Manufactured Ti6Al4V Using Machine Learning Regression
摘要
Engineering applications of high strength-to-weight ratio normally employ titanium alloys, i.e., Ti6Al4V alloy. This research investigates Ultimate Tensile Strength of printed specimens of Ti6Al4V fabricated using Laser Powder Bed Fusion with respect to key printing parameters and annealing treatment after post-processing. Taguchi method was employed for optimization of printing parameters through signal-to-noise (S/N) ratio analysis with constant layer thickness. Heat treatment at 850 °C for two hours and furnace cooling significantly improved UTS, as revealed by tensile tests on a computerized Universal Testing Machine. With 300 W laser power, 40 µs exposure time, and 0.08 mm hatch spacing, the maximum tensile strength of 986.9 MPa was attained and the most influential parameter as per the Anova analysis is exposure time for both as build and after heat treatment. Energy-Dispersive x-ray Spectroscopy and Scanning Electron Microscopy microstructural analysis revealed that there was a relationship between better tensile properties, fewer defects and heat treatment. Heat gradients in course-by-course processing accounted for differences in composition between the processed specimens and the initial powder, as indicated by x-ray diffraction analysis. To make predictions of UTS and minimize the necessity for slow experimental runs, Machine Learning algorithms were also explored. K-Nearest Neighbor gave the best result among all models with R2 of 0.922, RMSE of 67.53 MPa, and MAE of 49.71 MPa and is therefore an appropriate nondestructive method for mechanical property estimation. Support Vector Regression has performed very poorly among the six models considered. Optimization due to the streamlined process is achievable with the addition of machine learning in additive manufacturing, which reduces the cost and time consumed to produce high-performance parts.