Researchers have developed and evaluated an artificial intelligence-driven dual-path framework designed for real-time fault diagnosis in power metering secondary circuits, reaching an accuracy of 96.5% ± 1.3% on a controlled test platform. In modern power systems, the reliable operation of power metering secondary circuits is essential for accurate energy measurement and safe power-system management. However, conventional fault diagnosis methods depend heavily on manual inspection and fixed rules, which struggle with nonlinear, transient, and composite faults.
To resolve the limitations of traditional inspection rules, the proposed framework processes six operational inputs: current, voltage, active power, reactive power, phase angle, and harmonic-distortion signals. These electrical signals are systematically aligned, denoised, segmented, normalized, and represented by both statistical and wavelet-packet features. Unlike conventional serial CNN–SVM schemes, the framework combines a lightweight one-dimensional convolutional neural network (1D-CNN) with a support vector machine (SVM) operating simultaneously in two distinct analytical branches.
Within the dual-path architecture, the two branches fulfill distinct and complementary diagnostic roles. The 1D-CNN branch learns local patterns directly from normalized signal windows, while the parallel SVM branch operates on physically interpretable statistical features. Rather than forcing a sequential decision pipeline, the calibrated probabilities from each branch are integrated by sample-wise confidence-adaptive fusion. This adaptive mechanism allows the system to balance localized signal features with interpretable statistical metrics for each analyzed sample.
The researchers validated the hybrid framework on a controlled platform containing 500 independently acquired records categorized into eight operating classes. Under standard conditions, grouped evaluation yielded 96.5% ± 1.3% accuracy, a 2.1% misjudgment rate, and a 0.42 s mean response time. These quantitative results confirm that the dual-path design can achieve high diagnostic precision alongside response speeds suitable for real-time monitoring under controlled experimental conditions.
To evaluate the design comprehensively, the team conducted comparative evaluations alongside standalone SVM, decision tree, 1D-CNN, BiGRU, CNN–LSTM, and a lightweight Transformer. Additional ablation and perturbation tests further supported the individual contributions of the parallel branches and the sample-wise adaptive fusion strategy. The evidence establishes a reproducible edge-oriented proof of concept, though the authors emphasize that multi-site field validation remains necessary before operational deployment in live power grids.
Subject of Research: Dual-path artificial intelligence framework for fault diagnosis in power metering secondary circuits
Article Title: An AI-driven hybrid CNN–SVM framework for real-time fault diagnosis in power metering secondary circuits
Article References: Chen, X., Zhang, C., Liang, F., Tian, R., Yan, Y., & Xu, Y. (2026). An AI-driven hybrid CNN–SVM framework for real-time fault diagnosis in power metering secondary circuits. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09921-2
Image Credits: AI Generated
DOI: 10.1007/s42452-026-09921-2
Keywords: power metering secondary circuits, fault diagnosis, one-dimensional convolutional neural network, support vector machine, confidence-adaptive fusion, wavelet-packet features, harmonic-distortion signals, active power, reactive power, phase angle, signal denoising, transient faults, composite faults, edge-oriented proof of concept, bidirectional gated recurrent units, convolutional neural network long short-term memory, lightweight transformer
News Source: Denise Maddox. (November 16, 2026). Parallel AI Model Diagnoses Power Metering Circuit Faults in Controlled Tests. Scienmag.



