Deep Learning–Based Real-Time Fault Detection in Networked Cyber-Physical Systems Using Embedded Edge Devices
TL;DR
The suggested method combines multi-modal sensor data fusion with lightweight neural models and an adaptive feedback mechanism, enabling efficient on-device inference in dynamic situations, supporting the claim that deep learning with edge computing can be significantly more responsive, flexible and energy-efficient for real-time CPS fault detection.
Abstract
The paper proposes a deep learning-based framework for real-time fault detection in the networked structure of cyber-physical systems (CPS), using embedded edge devices. The suggested method combines multi-modal sensor data fusion with lightweight neural models and an adaptive feedback mechanism, enabling efficient on-device inference in dynamic situations. According to the experimental analysis of CPS benchmarks for detection accuracy, latency, and energy consumption, there is a +19.3 % improvement in detection accuracy, a -18.6 % decrease in latency, and a -13.1 % decrease in energy consumption relative to the baseline models. Moreover, there are system reliability gains of +21.8%, giving it resilience in the noisy and time-sensitive environment. Scalable deployment in industrial automation, smart grids, and autonomous systems is supported by the architecture, which has low computational overhead. These findings substantiate the claim that deep learning with edge computing can be significantly more responsive, flexible and energy-efficient for real-time CPS fault detection.