Uncertainty-Aware Resilience-State Inference for Industrial Wireless Networks

Motivation

While the transition to wireless unlocks unprecedented flexibility, it exposes industrial control loops to the inherent volatility of shared radio frequency environments. In such environments, the wireless channel is continuously subjected to interference from competing devices sharing the unlicensed band, as well as dynamic electromagnetic interference (EMI) and transient fading from heavy metallic machinery.

To manage this unpredictability, modern IIoT architectures increasingly integrate machine learning-based diagnostic engines to detect anomalies and forecast channel degradation before packets are lost. When standard neural networks are presented with novel, out-of-distribution (OOD) interference patterns such as unprecedented jamming signatures or abrupt environmental noise spikes they do not possess the mathematical capacity to output "I do not know." Instead, they force a confident but potentially incorrect binary classification, oscillating between "Operational" and "Disrupted".

Topic

This deterministic overconfidence directly undermines industrial safety and resource efficiency. In an adaptive multi-connectivity system, a false positive triggered by a brief, benign signal dip forces the network into an immediate hard failover, wasting spectrum resources. Conversely, missing a genuine disruption leads to application packet delays that exceed the established application bounds, triggering emergency mechanical halts on the factory floor. Therefore , there is a necessity to transition from binary, point-estimate network diagnostics to probabilistic, uncertainty-aware reasoning.

Skills required

• Python, Git
• Linux experience

What you will learn

• Networking in Linux
• Artificial Intelligence modelling