High coupling exists among parameters in complex assembly processes, leading to conflicting and drifting objectives (quality, cycle time, and energy). We propose an adaptive multi-objective parameter optimization framework that integrates a tuned digital twin, a set of surrogate models aware of uncertainty, constrained expected hypervolume improvement optimization, and online drift compensation. On three publicly available manufacturing/inspection datasets and a highly simulated assembly twin, our method improves hypervolume by 4.0 to 5.0 points compared to a strong baseline, while simultaneously reducing defect rate and cycle time with fewer evaluations. Paired t-tests show significant improvements (p < 0.001) at the 95% confidence interval. Abstract: This framework is a scalable, general, and data-efficient adaptive tuning scheme for complex assemblies with uncertainty.
Xiao Han (Thu,) studied this question.