Human-Knowledge-Embedded Bayesian Optimization for adaptive machine setup in customized building manufacturing

The transition to mass customization in industrialized building construction imposes significant demands on rapid machine setup, yet current practices face two critical hurdles: (1) Implicit Knowledge Discovery, as operator expertise remains difficult to quantify; and (2) Cross-Task Knowledge Transfer, as expertise is rarely reused across varying designs, forcing optimization to restart for each task. To address these challenges, we propose a Human-Knowledge-Embedded Bayesian Optimization (HK-Embedded BO) framework utilizing Shapley Additive Explanations (SHAP) to extract implicit human knowledge from historical setup data, converting tacit experience into interpretable, transferable knowledge. This extracted knowledge guides BO by dynamically shaping the search space toward high-potential regions for new tasks. Results demonstrate efficacy in both challenges. SHAP-based discovery captures parameter importance, reducing iterations by 14.6% versus human operators and 76.7% versus vanilla BO. Knowledge transfer achieves robust cross-task generalization with standard deviation reduced by 63.5% and 83.3%, proving human-AI collaboration bridges manual adaptability and automated efficiency.