Edge-Native Resilient Orchestration: Bridging the Semantic IT-OT Gap in SAP-Integrated IIoT via Multi-Agent DRL

The integration of Operational Technology (OT) with Information Technology (IT) in Industry ۴.۰ relies heavily on Enterprise Resource Planning (ERP) systems like SAP S/۴HANA for real-time decision-making. However, this convergence exposes a critical "semantic gap" at the edge-cloud continuum, where compromised IIoT nodes inject syntactically valid but operationally malicious data (e.g., "slow poisoning" of production parameters). Traditional Byzantine Fault Tolerance (BFT) protocols are computationally prohibitive for industrial edge gateways, and standard firewalls lack business logic awareness. To address this, we propose a lightweight, Edge-Native Multi-Agent Deep Reinforcement Learning (MADRL) framework for resilient transaction orchestration. Instead of relying on static consensus, we model the validation process as a dynamic resource competition between attackers and validators. Our agent, deployed on NVIDIA Jetson Nano hardware, uses an LSTM-based Context Encoder to map sensor streams directly to SAP BAPI constraints. Experimental results prove that our approach enhances operational resilience, supporting a ۹۹.۲% Transactional Integrity Ratio (TIR) while reducing inference latency to ۱۲.۴ ms, making it ۸.۴x faster than traditional BFT and suitable for real-time industrial control loops.

کلیدواژه ها:

IT-OT Convergence ، Industrial IoT ، SAP S/۴HANA ، Edge Intelligence ، Resilient Orchestration ، MADRL