Lightweight Acoustic Anomaly Detection Algorithm for Wireless Sensor Networks

Data-driven anomaly detection for machines in the wireless Industrial Internet of Things (IIoT) environments is a critical task in modern industrial domains for system stability and reliability. Edge computing in IIoT environments can offer advantages by reducing decision-making time and bandwidth usage. However, edge nodes have stringent constraints in memory and computing capacities. Thus, to implement an anomaly detector on edge nodes, it is necessary to develop efficient data preprocessing algorithms and a lightweight anomaly detection model. The proposed algorithm employs a parallel discrete wavelet transform that can efficiently capture and compress both low and high-frequency content in the acoustic signals, significantly reducing data preprocessing and model training time as well as memory usage. The experimental results, using real-world data collected from industrial machines, confirm the efficient use of memory and computing resources in the edge nodes.