Configuring Industrial Wireless Mesh Network via Dual-Mind Reasoning
Conference
Ma, A, Luo, D, Maatouk, A et al. (2026). Configuring Industrial Wireless Mesh Network via Dual-Mind Reasoning
. 113-120. 10.1109/DCOSS-IoT69657.2026.00021
Ma, A, Luo, D, Maatouk, A et al. (2026). Configuring Industrial Wireless Mesh Network via Dual-Mind Reasoning
. 113-120. 10.1109/DCOSS-IoT69657.2026.00021
Configuring industrial Wireless Mesh Networks (WMNs) requires jointly optimizing many settings, such as transmission power, channel allocation, and retransmission policy, to meet stringent performance requirements. Existing approaches rely on simulation-based optimization, but configurations that perform well in simulators often fail after deployment due to unmodeled interference, hardware diversity, and environmental dynamics. Prior domain adaptation techniques mitigate this simulation-to-reality gap by incorporating physical measurements but require extensive data collection in industrial facilities, imposing substantial data collection overhead. We present LLMNET, a framework that configures industrial WMNs without requiring expensive data from real-world deployments. LLMNET introduces three key innovations: a multipath self-consistent Chain-of-Thought mechanism with consensus aggregation that stabilizes configuration decisions across diverse network conditions; a domain knowledge editing pipeline that corrects mismatched feature correlations between simulation and real deployments using expert-derived priors; and a cognitive-inspired dual-mind architecture comprising a lightweight Fast Mind for rapid heuristic inference and a deliberative Slow Mind for structured optimization, enabling adaptive reasoning under resource constraints. We evaluate LLMNETusing four widely used simulators (NS-3, Cooja, TOSSIM, and OMNeT++) and a 50-node WirelessHART testbed. Experimental results show that LLMNETimproves configuration accuracy by up to 40% over state-of-the-art baselines.
