Mesh of Things (MoT) Network-Driven Anomaly Detection in Connected Objects

TL;DR

This paper develops a hybrid performance model for Mesh of Things networks to detect and localize anomalies by analyzing packet delivery ratio and latency, validated through experimental data in an air cargo monitoring scenario.

Contribution

It introduces a Bayesian network-based approach to model link uncertainties and detect anomalies in hybrid BLE mesh-PLC networks, a novel application in IoT security.

Findings

Effective anomaly detection in hybrid mesh networks

Validation with experimental PDR and latency data

Identification of link uncertainties in BLE and PLC

Abstract

This paper presents a hybrid Mesh of Things (MoT) network performance model to evaluate the end-to-end Packet Delivery Ratio (PDR) and latency. These PDR and latency measures are used to identify both a de-tangled mesh as well as to track the mesh successfully. A de-tangled mesh is a mesh with an anomaly where one or more nodes are separated from the rest of the mesh network. We demonstrate the performance model of a hybrid BLE mesh-PLC network by considering an air cargo monitoring application and validate with experimental PDR and latency data. The link uncertainty in Bluetooth Low Energy (BLE) mesh may be attributed to (a) RF interference,(b)~Transmitter's vicinity range, and (c) Receiver sensitivity. In contrast, the link uncertainty in Power Line Communication (PLC) may be attributed to: (a) Colored background noise, (b)~Channel frequency response, and (c) Impulse noise appearing due to load state as well as variations in the powerline. In our work, we construct an equivalent Bayesian network for the mesh to be tracked, capture the uncertainty within the mesh links using the Noisy-OR and the Noisy-Integer addition model and perform belief propagation to detect and localize a network anomaly.