Real-Time Fault Detection for Arduino Sensor Networks Using Lightweight ML Models

Marwa Awni Kamal

doi:10.23918/eajse.v11i3p21

Authors

Marwa Awni Kamal Computer Engineering Department,Faculty of Engineering, Tishk International University, Erbil, Iraq. https://orcid.org/0000-0002-0657-0691

DOI:

https://doi.org/10.23918/eajse.v11i3p21

Keywords:

Arduino-Based Embedded Systems, Real-Time Fault Detection, Bonsai Algorithm, Lightweight Machine Learning, Sensor Fault Diagnosis, Tiny ML, Edge AI

Abstract

In embedded systems based on the Arduino platform, real-time fault detection is difficult because of the small computational power, memory, and power restrictions. Traditional threshold-based methods lack flexibility and fail to detect subtle fault patterns, types of faults like gradual drift or intermittent faults, and most machine learning methods are currently too costly to run on microcontrollers. The objective of the study is to come up with an effective fault detection model that can attain high accuracy at low latency and memory consumption on constrained-resource embedded systems. This article suggests a lightweight time-based fault detection system that uses low-complexity time-domain features and an edge-optimized Bonsai machine learning classifier to perform real-time on-device inference. A controlled experimental setup with simulated fault conditions such as bias, drift, spike anomalies, and signal loss is used to generate multivariate Arduino sensor data, which includes temperature, vibration, current, and voltage data. The data is divided into 50-sample sliding windows. Model optimisation methods, including pruning and fixed-point quantisation, including pruning and fixed-point quantisation, can trim the model down to 1.76 KB and can achieve inference latency as low as 105 µs using hardware on the Arduino-class. The experimental findings indicate that the proposed method has a fault detection rate of 0.991, which is better than the k-Nearest Neighbors (k-NN), both in terms of latency (approximately 1632 ) and memory (5686 KB) utilization, but it has a similar performance as a Decision Tree classifier that has an accuracy of 1.000 at a higher cost. The framework achieves a balance between diagnostic accuracy and resource efficiency that can be used in real-time fault detection of a low-cost embedded sensor network. The findings reveal that the suggested Bonsai-based model provides a better accuracy-latency-memory trade-off than traditional machine learning models, and proves to be efficient in detecting faults in real-time on resource-limited embedded systems.

Downloads

Download data is not yet available.

References

Reis MJ. Lightweight Signal Processing and Edge AI for Real-Time Anomaly Detection in IoT Sensor Networks. Sensors. 2025. https://doi.org/10.3390/s25216629

[2] Samuel ET, Polonelli T, Magno M, Esan A. TinyML Anomaly Detection and Fault Prediction for Industrial Applications. 2025. https://doi.org/10.21203/rs.3.rs-8379288/v1

[3] Attarha S. A framework for sensor fault detection and management in low-power IoT edge devices: Universität Bremen (Germany); 2025.

[4] Fowdur TP, Ajageer L. A hybrid online learning-based predictive maintenance system for Industry 4.0. Computing. 2025. https://doi.org/10.1007/s00607-025-0142-x.

[5] Zheng Z, Liu L, Liu P, Yu Z. Long-term mechanical behaviour of CRTS III ballastless track-simply supported girder system under permanent load coupling action. Alexandria Engineering Journal. 2025;117:276-88. https://doi.org/10.1016/j.aej.2025.01.045.

[6] Kok CL, Heng JB, Koh YY, Teo TH. Energy-, Cost-, and Resource-Efficient IoT Hazard Detection System with Adaptive Monitoring. Sensors. 2025. https://doi.org/10.3390/s25061761

[7] Caleb S, Padmapriya G, Nandhini T, Latha R. Revolutionizing fault detection in self-healing network via multi-serial cascaded and adaptive network. Knowledge-Based Systems. 2025. https://doi.org/10.1016/j.knosys.2024.112732

[8] Grujev M, Stojanovic D, Milosavljevic A, Ilic M, Prodanovic V. An efficient real-time apple disease classification approach using a novel lightweight neural network on an Arduino edge device. Internet of Things. 2025. https://doi.org/10.1016/j.iot.2025.101833.

[9] Al-Libawy H. Efficient implementation of a tiny deep learning classifier based on‎ vibration-related fault detection using limited-resource hardware. Journal of the University of Babylon for Engineering Sciences. 2025. https://doi.org/10.29196/770r2493

[10] Gupta S, Shivhare SN. Embedded TinyML for Predictive Maintenance: Vibration Analysis on ESP32 with Real-Time Fault Detection in Industrial Equipment. International Journal on Computational Modelling Applications. 2025. https://doi.org/10.63503/j.ijcma.2025.114

[11] Moharam, M. H., Ashraf, K., Alaa, H., Ahmed, M., & El-Hakim, H. A. (2025). Real-time detection of Wi-Fi attacks using hybrid deep learning models on NodeMCU. Scientific Reports, 15(1), 32544.

[12] Gochhayat, P. C., Afam, M., Tanvir Alam, S. K., Mallick, G. K., Sarker, K., & Paramanik, S. (2025). Edge Ai-Enabled Dynamic Power Factor Correction Using Tinyml, Blockchain, and IoT for Real-Time Smart Grid Optimization And Industrial Applications. I-Manager's Journal On Electrical Engineering, 18(4).

[13] Akande, J. O., & Chukwunweike, J. Developing Scalable Data Pipelines For Real-Time Anomaly Detection In Industrial IoT Sensor Networks.

[14] Sudharson, D., Divya, P., Saranya, K., Dubey, A. K., Vijay, S., & Mayuri, G. (2023, September). A Novel AI Framework for Assuring Data Sustainability in Health Care Dataset. In 2023 Third International Conference on Ubiquitous Computing and Intelligent Information Systems (ICUIS) (pp. 161-166). IEEE.

[15] Park, D., Kim, S., An, Y., & Jung, J. Y. (2018). LiReD: A light-weight real-time fault detection system for edge computing using LSTM recurrent neural networks. Sensors, 18(7), 2110.

[16] https://github.com/DEEPAK-KHAIRWAL/SDN-IMPLEMENTATION-OF-CLOUD-AND-IOT-BASED-LIQUID-SPILL-AND-LEAKAGE-DETECTION-SYSTEM