Inhaltszusammenfassung

Automated process planning in variant-rich manufacturing requires the discovery of suitable resource capabilities in heterogeneous production environments. Existing approaches rely either on rigid vocabulary matching with limited flexibility, or on AI-driven methods lacking transparency and real-time applicability. This paper proposes a multi-stage capability discovery architecture that balances matching precision against computational cost through explicit escalation logic. The three stages ...Automated process planning in variant-rich manufacturing requires the discovery of suitable resource capabilities in heterogeneous production environments. Existing approaches rely either on rigid vocabulary matching with limited flexibility, or on AI-driven methods lacking transparency and real-time applicability. This paper proposes a multi-stage capability discovery architecture that balances matching precision against computational cost through explicit escalation logic. The three stages of the process are applied sequentially, and each stage is invoked only if the preceding stages fail to produce sufficient results. They are as follows: a deterministic ECLASS-based lookup within Asset Administration Shell (AAS) submodels, knowledge-graph reasoning via DIN 8580 taxonomy traversal, and sentence-embedding similarity search. Escalation is governed by match confidence, result completeness, and infrastructure availability. The validation of discovered capabilities is achieved through property constraint checking. The architecture is evaluated on robotic deburring scenarios under varying requirement specificity. On a 69-capability deburring testbed, the full pipeline achieves P@1 = 0.875 and MRR = 0.910 on 88 previously unseen queries, while the staged design keeps median latency below 1 ms for known-vocabulary queries and below 400 ms for the full pipeline on CPU.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
4.11-06 - Produktionsautomatisierung und Montagetechnik

DDC Sachgruppe:
Ingenieurwissenschaften