A Multi-Agent Decision Intelligence Architecture for Autonomous Production Exception Resolution in SAP-Integrated Smart Factories
Year: 2026
Type: article
Abstract: Smart factories routinely encounter material shortages, equipment failures, quality deviations, capacity conflicts, and late supplier deliveries. Although SAP enterprise resource planning systems provide transactional visibility, conventional exception handling remains fragmented across planning, maintenance, procurement, and shop-floor teams. This paper proposes a multi-agent decision intelligence architecture that combines event-driven SAP integration, specialized autonomous agents, a shared digital production state, constrained optimization, and human-governed execution. A supervisory agent coordinates detection, diagnosis, negotiation, policy verification, and recovery while maintaining auditable action provenance. The methodology defines a formal exception representation, risk scoring, agent utility, negotiation protocol, feasibility constraints, and a closed-loop learning mechanism. An illustrative, explicitly synthetic simulation of 1,200 production exceptions compares manual handling, rule-based automation, a centralized AI planner, and the proposed architecture. Under the stated assumptions, the multi-agent approach reduces mean resolution time from 94 to 31 minutes, increases on-time recovery from 68.2% to 91.4%, and reduces estimated disruption cost from 100 to 58 index points relative to manual processing. These values demonstrate model behavior in a controlled scenario and are not empirical SAP deployment results. The proposed framework offers a testable design for safe, explainable, and interoperable autonomous exception resolution.
DOI: 10.66592/szfg5ry9
Authors: Ahmed, Ali, pal, Ritu
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7235931613
