Simulation-based multi-objective decision support for Comprehensive Digital Twins of smart manufacturing systems

Decision support in Smart Manufacturing Systems (SMSs) often requires balancing conflicting objectives related to multiple system dimensions, such as maximizing the number of completed products while minimizing energy consumption and CO₂ emissions. Comprehensive Digital Twins (CDTs) integrate multiple system dimensions within one executable model to support multidimensional analysis and multi-objective decision-making. However, an automatically extracted, simulation-ready CDT is not inherently optimization-ready because controllable parameters, admissible ranges, optimization objectives, and decision-to-model mappings remain unspecified. This paper addresses this gap by introducing a semi-automated framework that prepares an automatically extracted CDT for simulation-based multi-objective optimization. Starting from multidimensional event logs, Multi-Flow Process Mining extracts dimension-specific Stochastic Petri Nets and integrates them into an executable Multidimensional Stochastic Petri Net represented in Multidimensional Petri Net Markup Language. An optimization-preparation layer combines model parameter information with expert-defined selections, bounds, and objectives, and maps decision vectors to executable model configurations evaluated through simulation. The framework is demonstrated using an illustrative manufacturing line for robotic assembly and welding with rework. The Nondominated Sorting Genetic Algorithm II (NSGA-II) is applied as the primary Pareto-based optimizer, while equal-weight weighted-sum scalarization is evaluated as a comparative baseline. The NSGA-II Pareto-front approximation supports post-search scenario selection. In a 100-replication reference-model evaluation, the selected scenario reduced CO₂ emissions by 3.81%, while completed products and grid-energy consumption changed only marginally ( − 0.29 % and + 0.17 % , respectively). These results support the proof-of-concept feasibility of extending an automatically extracted multidimensional simulation model to multi-objective decision support.