AI-Driven prediction of surface roughness in laser-polished LPBF Ti6Al4V: a sustainable proof-of-concept

Abstract

Additive manufacturing (AM) of Ti-6Al-4V often yields high surface roughness (Ra > 10 µm), requiring post-processing for aerospace and biomedical use. This study proposes a lightweight physics-informed artificial neural network (ANN) trained on 20 experimental and ~5,000 surrogate samples derived from an overlap-adjusted energy–roughness relation. Incorporating both process parameters and physics-guided descriptors, the model achieved R² ≈ 0.99 with RMSE < 0.1 µm, predicting polishing outcomes with minimal data. The framework demonstrates that physics-assisted AI can reduce experimental trials from >30 to ~5, enabling sustainable, data-efficient optimization for intelligent laser polishing of AM components.

Topics & Concepts