Graph Driven Reinforcement Learning for Fault Tolerant TSN Routing in Underground Mines

Abstract

Time Sensitive Networking (TSN) is currently highly needed in cyber-physical and Industrial Internet of Things (IIoT) domains. However, deploying it in harsh and changing environments such as underground mines is very difficult due to frequent node and link failures, interference, and unpredictable changes in the network structure. This research presents GDRL-MORP, a Graph-Driven Reinforcement Learning framework for Multi-Objective redundant path planning. The proposed model uses a Graph Neural Network (GNN) combined with Deep Reinforcement Learning (DRL) to create redundant paths to improve latency, jitter, reliability, and energy usage. All the communication flows are treated as constrained Markov Decision Processes (MDP), which allows the agent to choose the best routes and backup options based on real-time network conditions. Simulation studies in a TSN-enabled mining setting show that GDRL-MORP greatly outperforms deterministic scheduling and frame replication methods, which results in $10 \%$ higher reliability, quicker convergence, and $20 \%$ less energy usage. The proposed framework provides an excellent platform to deploy TSN in critical underground mining environment.