AutoFlow-MO: An Automated Planning Method of Experimental Process Flows for Multi-Object Demands

Abstract

The intelligent planning and scheduling of complex Test Support Systems (TSS) are critical for the efficient and safe operation of large-scale experimental facilities. However, traditional manual scheduling and static rule-based paradigms struggle to manage the high-dimensional, non-linear coupling between multi-object concurrent demands and stringent physical resource constraints. To address this NP-hard problem, this paper proposes a novel automated process planning framework driven by multi-dimensional physical constraints and data-driven heuristics. First, a knowledge-driven formalization mechanism utilizing an improved K-Means++ clustering strategy with Mahalanobis distance is established to accurately map heterogeneous, unstructured test demands into computable mathematical models. Subsequently, a hierarchical dependency model and a multi-dimensional resource configuration conflict matrix are constructed. Combined with domain-knowledge-guided heuristic traversal algorithms, this formulation successfully decouples severe spatial-temporal competition and eliminates resource deadlocks among coupled subsystems. Finally, to navigate the massive constraint space, a Space-Time Cooperative A* algorithm integrated with Latin Hypercube Sampling (LHS) is proposed. This approach effectively filters physically infeasible paths and consistently locates Pareto optimal scheduling schemes that balance minimum task response delays with optimal energy consumption. Ultimately, the proposed methodology fundamentally shifts experimental process planning towards a highly autonomous paradigm, providing robust decision-making support for the efficiency and safety enhancement of complex test facility clusters.