제조 시스템의 대규모 탐색을 위한 이산사건–물리 기반 하이브리드 최적화 프레임워크 개발

초록

In constructing digital twins for manufacturing and logistics systems, discrete-event simulation and physics-based simulation have conflicting strengths: the former enables computationally efficient large-scale analysis, whereas the latter reproduces microscopic physical phenomena with high fidelity. This study proposes a multi-fidelity hybrid simulation framework that combines low-fidelity discrete-event simulation with high-fidelity physics-based simulation. In the proposed framework, microscopic physical variables observed in the physics-based simulation are abstracted into conditional probability distribution models and injected in advance into the discrete-event simulation. This enables large-scale iterative evaluation with reduced computational cost, allowing system-suitable parameters to be rapidly explored. The selected candidate parameters are then revalidated through physics-based simulation to determine representative operating conditions. The framework was applied to a simple robotic pick-and-place cell process, and the results confirmed that it can rapidly identify process performance thresholds while deriving final stable parameters that reflect physical risks. Therefore, the proposed framework can simultaneously secure the computational efficiency required for large-scale optimization and the physical reliability of manufacturing processes. It is expected to support cost-effective digital twin and simulation development in various manufacturing fields and contribute to digital transformation.

키워드