Machine Learning Driven Prediction and GUI Based Optimization of Quasi-Static Mechanical Properties in SLM Fabricated Ti6Al4V Alloy

Overview

Selective laser melting can print Ti6Al4V parts for aerospace and biomedical use, but the tensile properties that come out depend on a tangle of process settings, and finding the right settings by experiment is slow and expensive. This paper asks whether machine learning can learn that relationship from results that have already been published.

A dataset of 201 results was built from the literature, covering six SLM process parameters and three tensile properties — yield strength, ultimate tensile strength and elongation. Six classical regressors were compared with an artificial neural network, and SHAP analysis was used to show which inputs drive the network's predictions and how that lines up with the process physics.

The network was then trained in reverse and wrapped in a graphical interface: an engineer enters the mechanical properties a part needs, and the tool suggests the process parameters to reach them — a practical check before anything goes into production.

Key Findings

• —201 results from published literature: six SLM process parameters against yield strength, ultimate tensile strength and elongation.
• —Support Vector Regression, Random Forest, K-Nearest Neighbors, Gradient Boosting, Gaussian Process Regression and Decision Trees reached only moderate accuracy.
• —An artificial neural network captured the relationships best, with R² up to 0.84 across all three properties.
• —SHAP analysis explains the relative importance and physical influence of each process parameter.
• —A reverse-trained ANN behind a GUI returns process parameters for the properties a part requires.

This is our summary of the paper. The full text, with its abstract, figures and data, is on the publisher’s site.