Modeling Process–Microstructure Relations in PBF-LB/M Laser Treatment using Gaussian Process Surrogates and Bayesian Optimization

A key challenge in additive manufacturing is the precise manipulation of microstructural properties to mitigate issues such as residual stresses and poor mechanical performance. In this context, laser treatments such as laser heat treatment and laser remelting offer a promising approach to influence microstructure in the process of powder bed fusion with laser beam melting (PBF-LB/M). However, the complex process-microstructure relationship remains insufficiently characterized for systematic process control. This work presents a novel approach for efficient modeling of this relationship using Bayesian optimization (BO) with Gaussian process (GP) surrogate models. It addresses the mentioned issues by integrating BO with on-machine eddy current (EC) sensing, where the EC phase angle serves as an indirect metric for microstructural changes, such as the retained austenite content in the H13 tool steel used in this work. The BO algorithm adaptively proposes laser treatment parameters based on the GP surrogate model and an Upper Confidence Bound (UCB) acquisition function, iteratively refining the process-microstructure mapping. The effectiveness of this approach was validated through two experiments that successfully manipulated the EC angle and thereby the retained austenite content, as confirmed by X-ray diffraction reference measurements. The trained GP model achieved high predictive accuracy with R² values up to 0.95, demonstrating its capability for closed-loop process control of microstructure modification via laser treatment in PBF-LB/M machines.