| Abstract: |
This paper describes an empirical study on a safety layer in a fog-computing enabled Internet of Things (IoT) for multi-robot industrial environments that facilitates real-time hazard detection and risk mitigation. The architecture distributes computation into three tiers edge fog nodes, intermediate fog nodes, and cloud back-ends to gain sub-10 millisecond safety response times that are not possible with traditional cloud-only architectures. Over a nine month experimental period, 664 GB of validated operational data was generated across a heterogeneous sensor suite deployed over a 12-robot assembly cell (LiDAR, Force/Torque, RGB-D, Proximity, Inertial Measurement Unit (IMU), and vibration). A hybrid AI detection method combining Long Short-Term Memory (LSTM) networks and fuzzy inference obtains a hazard detection F1 score of 97.3%, a false alarm rate of 0.9% and a mean end-to-end safety response latency of 6.2 ms—an order of magnitude faster than the ISO 10218-1 recommended reaction time budgets. The ability of multivariate linear regression (R2 = 0.926, F(5, 394) = 241.8, p < 0.001) to select AI inference confidence as the most predictive feature of classifier accuracy and one-way ANOVA (F(3, 476) = 892.4, p < 0.001, eta2 = 0.849) to reveal statistically significant latency benefits of the proposed fog-AI architecture vs cloud only, edge only, and non-AI fog alternatives. The proposed system demonstrates production-grade safety infrastructure for Industry 4.0 deployment with a system uptime of 99.94% and a mean time between failures (MTBF) of 6,940 h. |
