Robust Parameter Optimization of Multi-Objective Variables in Laser Metal Deposition Using Machine Learning

Abstract

Rapid establishment of optimized production parameters is crucial in manufacturing technology. However, to achieve the desired processing results, complex production technologies such as laser metal deposition (LMD) require evaluating numerous evaluation points and processing parameters. Thus, recent studies have explored the integration of artificial intelligence, especially machine learning, into manufacturing technology for fine-tuning the processing parameters. Nonetheless, laser processing engineers encounter challenges in interpreting the processing parameters identified by machine learning, and the robustness of the parameters enabling stable production is unclarified. To address these issues, this study developed a method to visualize the robustness conditions of key processing parameters. Fundamentally, the approach aims to scalarize and transform multiple objective variables related to processing outcomes into a singular evaluation index via weighted scaling. Thereafter, this index is explored in a two-step process employing both local and global search methods. The local search focuses on regions where the evaluation index is stable and favorable, whereas the global search attends to the remaining parameters where this region reaches its maximum. The proposed method exhibits minimal sensitivity to the initial weight settings and regional thresholds, thereby rendering it relatively straightforward for practical application.