2026· E3S Web of Conferences· 0 citations· 12 references
TL;DR
A practical, real-time arc fault detection system that integrates a comprehensive two-branch household dataset, lightweight handcrafted features, and classical machine learning for deployment on an STM32F407 microcontroller (MCU) and confirms its feasibility for low-cost, real-time protection devices.
Abstract
AC series arc faults in household appliances pose significant safety risks but remain challenging to detect due to load-dependent signatures and complex interactions in mixed-branch operations. This paper presents a practical, real-time arc fault detection system that integrates a comprehensive two-branch household dataset, lightweight handcrafted features, and classical machine learning for deployment on an STM32F407 microcontroller (MCU). Current signals were sampled at 10 kHz across 32 scenarios, encompassing 13 singleload and 19 two-branch mixed-load cases. Each signal window is represented by a 133-dimensional feature vector, comprising five statistical descriptors and 128 FFT magnitude coefficients, optimized for resource-constrained environments. Among various evaluated classifiers, Random Forest was selected for its superior trade-off between recognition performance and deployment efficiency. The proposed model achieved 89.18% accuracy, 91.34% recall, and 86.96% F1-score. Furthermore, a leave-one-scenario-out (LOSO) evaluation was employed to systematically analyze the masking effect in multi-branch environments, identifying challenges in which high-power loads obscure weak arc signatures. The complete firmware requires 896 KB of Flash and 102 KB of RAM, with a total processing time of only 7.02 ms per cycle, significantly within the update interval and confirming its feasibility for low-cost, real-time protection devices.
This study develops a low-complexity data-driven method for real-time fault classification, and localization on a 20 kV radial feeder, while accounting for the computational and memory constraints of embedded hardware. The proposed approach is motivated by the changes introduced by distributed generation (DG), which alters fault-current behavior and reduces the reliability of conventional impedance-based protection. These limitations are mainly caused by reverse power flows and the additional infeed contribution from DG units during fault conditions. The proposed hybrid architecture processes time- and frequency-domain features, including symmetrical components and wavelet energy, through a dual-stage inference chain. The first stage uses a Random Forest classifier for fault-type identification, while the second stage uses a Multi-Layer Perceptron (MLP) distance estimator for fault localization. Evaluation on 23,778 MATLAB/Simulink fault scenarios, was performed using a grouped scenario-level holdout split to reduce overlap between training and test cases. The proposed approach reduces the localization error compared with the conventional impedance method, which produced errors exceeding 14 km under DG operation. The Random Forest classifier achieved 99.44% fault-type accuracy, and the MLP estimator reached a Mean Absolute Error (MAE) of 93.8 m, with localization accuracy of 99.51% within 1 km. Analytical resource profiling estimates less than 300 KB Flash and under 50 µs execution time, assuming a Cortex-M4F/M7-class target, pending hardware-in-the-loop validation.
M. Lemkharbech, S. Sarih, Z. Boulghasoul et al.· International Conference on...· 0 citations
Within the framework of Industry 4.0, the implementation of real-time and economical fault diagnostics for rotating machinery is vital for predictive upkeep. Conventional cloud-reliant strategies grapple with drawbacks like excessive latency and substantial bandwidth needs. At the same time, specialized diagnostic apparatus is typically too expensive for smaller enterprises. This work introduces VibeSense, a remarkably inexpensive edge AI platform designed for the instantaneous identification of imbalance faults in rotating machinery. The described setup employs an ESP32-S3 microcontroller paired with an MPU6050 MEMS accelerometer, keeping the entire hardware expenditure below 50 RMB (approx. 7 USD), which is, to the best of our knowledge, one of the lowest-cost edge AI solutions for fault diagnosis reported to date. The system executes on-device Fast Fourier Transform (FFT) for spectral vibration evaluation and derives an 11-parameter feature array covering time-domain and frequency-domain attributes. A streamlined Random Forest classifier, containing 50 decision trees, undergoes training and is subsequently installed straight onto the ESP32-S3 via translation into C-based if-else branching code. This tactic considerably lowers memory consumption relative to conventional TinyML solutions. Testing outcomes indicate a fault identification success rate of 100% on artificial datasets and upwards of 95% during actual fan-based trials. The minimal expense, straightforward implementation, and precision of the platform underscore its promise as a feasible and expandable edge AI instrument for industrial predictive servicing. All source code and documentation have been open-sourced and are publicly available at https://github.com/vibezhang/ VibeSense.
Yisong Deng· International Conference on...· 0 citations
A progressive three-stage time–frequency learning framework that identifies series arc faults directly from normalized current waveforms and provides robust discrimination across unseen measurement sessions within the evaluated load categories and operating conditions is presented.
Seoyoung Jeon, Won-Kyu Choi, Sungsoo Kwon et al.· Italian National Conference...· 0 citations
Measurement-level feature attribution shows that the DL-Xformer decision is driven by physically meaningful current and voltage channels at the attacked measurement location, which motivates a layered protection architecture for next-generation inverter-dominated smart grids.
Emad Abukhousa, Saman A. Zonouz, A. P. S. Meliopoulos· 0 citations
A hybrid two-stage machine learning pipeline that decouples detection from classification is proposed, and the direction of the zero-sequence signature is found to be system-dependent, motivating a learned decision boundary in place of a fixed relay threshold.
Sahil Manikshete, A. Gujarathi, Thanh Long Vu et al.· 0 citations
Hydromachinery is vital for clean and sustainable power generation, where reliable and efficient operation directly supports the stability of hydropower plants. To achieve this, real-time performance tracking and fault monitoring are becoming increasingly important. This review summarizes recent techniques and technologies used for monitoring turbines and their components in operation. Key areas include sensor-based data collection, modern signal processing tools, and artificial intelligence methods for detecting issues such as cavitation, vibration irregularities, pressure fluctuations, and mechanical wear. Methods like wavelet analysis, principal component analysis (PCA), support vector machines (SVM), and digital twins are discussed for their roles in fault diagnosis and performance evaluation. Advances in IoT-enabled monitoring and predictive maintenance are also highlighted, demonstrating their potential to enhance reliability and minimize downtime. The paper further outlines challenges such as harsh operating conditions, large data handling, and the need for accurate predictive models. Future directions are suggested, focusing on hybrid machine learning approaches, adaptive monitoring strategies, and digital twins for smart, autonomous health management of hydro machinery.
Juhi Padma, Hemant J. Sagar· IOP Conference Series: Earth...· 0 citations