Sequential design optimization of high-variability additive manufacturing processes using Gaussian process regression

Maher B. Alghalayini

doi:10.1016/j.jmapro.2026.01.007

Sequential design optimization of high-variability additive manufacturing processes using Gaussian process regression

2026-01-07

0

PRE

AI

M

Maher B. Alghalayini *

C

Christiaan J. J. Paredis

F

Fadi Abdeljawad

DOI：10.1016/j.jmapro.2026.01.007

Original

Original request for help

Save

Abstract

En 中文

Additive Manufacturing (AM) has undergone remarkable evolution, opening doors to unprecedented opportunities and challenges. Optimizing AM processes has emerged as a critical frontier for industry advancements. While prior research has presented various optimization approaches, the inherent complexity and variability in AM processes pose unique obstacles, making it difficult to apply existing optimization methods effectively. Here, a data-driven sequential optimization framework explicitly tailored for AM is developed. The problem is framed as a multistage stochastic programming (MSP) problem, in which a sequence of experiments is performed, each optimally informed by the results of previous experiments to efficiently identify the process parameters that lead to the most valuable material properties. The framework explicitly accounts for both inherent variability of the AM processes and uncertainty in the models and measurements. In each iteration, the selection of new experiments and their sample sizes is guided by trading off the expected value of additional experiments with the cost of continued experimentation. The solution approach is a myopic approximation of the MSP problem in which Gaussian process regression is combined with value-of-information optimization. To determine the approach’s efficacy, we apply it to synthetic problems inspired by AM of stainless-steel properties. Results show that the framework outperforms Bayesian optimization with the Expected Improvement acquisition function, achieving similar optimization quality with 55% reduction in experimental costs. In broad terms, the approach meets the challenges inherent in AM optimization, offering a swift, efficient, and value-driven approach. Such data-driven, efficient experimentation is expected to accelerate AM’s continued growth and innovation.

Sequential design optimization of high-variability additive manufacturing processes using Gaussian process regression

Abstract

Journal

Organization