A Hybrid Neuro Symbolic Multi Agent Framework for Autonomous Manufacturing Workflow Execution and Constraint Aware Resource Optimization in SAP ERP
Year: 2026
Type: article
Abstract: Manufacturing enterprises require adaptive workflow orchestration while preserving enterprise resource planning (ERP) transactional integrity, hard production constraints, and traceable decisions. This paper proposes a hybrid neuro-symbolic multi-agent framework that combines predictive neural models, declarative constraint reasoning, and distributed negotiation for SAP-integrated manufacturing execution. A planning agent decomposes production orders, resource agents propose feasible capacity allocations, a symbolic validator enforces material, precedence, labor, and maintenance rules, and an execution agent commits approved actions through governed SAP interfaces. A discrete-event simulation design compares rule-based dispatch, predictive dispatch, unconstrained multi-agent negotiation, and the proposed hybrid architecture under variable demand and disruptions. Illustrative, explicitly synthetic outcomes show a reduction in mean tardiness from 18.4 to 10.2 hours and an increase in on-time completion from 78.0% to 91.5% relative to rule-based dispatch. The framework emphasizes feasibility certificates, exception handling, audit logs, human approval thresholds, and rollback mechanisms. Results are intended as a reproducible demonstration rather than evidence of a live SAP deployment. The study offers an implementable architecture and an evaluation protocol for future industrial validation.
DOI: 10.66592/ty26iu21
Authors: Sánchez, Irma, Addu, Sowjanya
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7236262097
