Edge AI for Real-Time Fault Detection in Embedded Systems
DOI:
https://doi.org/10.56472/ICCSAIML25-129Keywords:
Edge AI, Fault Detection, Embedded Systems, Model Optimization, Real-Time Monitoring, Predictive Maintenance, Resilient System DesignAbstract
The increasing complexity and criticality of industrial and automotive systems demand rapid and reliable fault detection mechanisms to ensure operational safety, reduce downtime, and improve system resilience. Traditional cloud-based approaches to fault detection often introduce latency, raise security concerns, and depend on reliable connectivity factors that are unsuitable for many real-time applications. This research explores the deployment of lightweight artificial intelligence (AI) models on embedded systems to enable real-time fault detection at the edge. We investigate model optimization techniques such as quantization, pruning, and knowledge distillation to adapt state-of-the-art AI algorithms for constrained hardware environments without compromising performance. The proposed framework is evaluated on representative industrial and automotive datasets using embedded platforms like ARM Cortex-M and NVIDIA Jetson Nano. Results demonstrate that edge AI models can achieve high fault detection accuracy with low inference latency and energy consumption, making them viable for real-world deployment. This study highlights the potential of edge intelligence to revolutionize safety monitoring and predictive maintenance in embedded systems
Downloads
References
pp. 923–928, 2019.
[2] I. A. Gheyas and L. S. Smith, “A review of fault diagnosis methods for industrial systems,” Artificial Intelligence Review, vol. 44, no. 2, pp. 217–248, 2015.
[3] W. Zhang et al., “Data-driven methods for predictive maintenance of industrial equipment: A survey,” IEEE Systems Journal, vol. 13, no. 3, pp. 2213–2227, 2019.
[4] A. Dey and M. M. M. Hassan, “A survey on edge computing in the industrial internet of things,” IEEE Access, vol. 8, pp. 143828–143850, 2020.
[5] H. Esmaeilzadeh et al., “Edge AI: On-demand deep learning model co-inference with device-edge synergy,” IEEE Computer Architecture Letters, vol. 18, no. 1, pp. 42–45, 2019.
[6] M. Han et al., “TinyONet: A lightweight and efficient CNN model for edge-computing-based industrial fault diagnosis,” Sensors, vol. 21, no. 8, 2021.
[7] Y. Choi, M. El-Khamy, and J. Lee, “Towards the limit of network quantization,” Proceedings of ICLR, 2017.
[8] G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531, 2015.
