Advanced Condition Monitoring and Fault Analysis in Industrial Electronics

Modern industrial electronic and electrical systems—converters, controllers, and electric motor drives among them—are exposed to electrical, mechanical, and thermal stresses that can lead to unexpected failures, production losses, and, in the worst cases, safety and environmental incidents. As industrial electronics are optimized for higher efficiency across wide ranges of speed and power, individual components are often driven closer to their operational limits, which in turn raises the likelihood of faults. Condition monitoring, fault diagnostics, and prognostics have therefore become central to keeping these systems reliable, available, and safe. The integration of virtual sensing, artificial intelligence, the Internet of Things, and cloud and edge computing has opened new opportunities to detect incipient faults early, to interpret their signatures, and to act on them before they escalate.

This Special Issue, “Advanced Condition Monitoring and Fault Analysis in Industrial Electronics,” was launched to bring together recent advances across this landscape, from device- and signal-level diagnostics to system-level architectures for resilient operation. The six contributions collected here span rolling-bearing fault diagnosis, series arc-fault detection, photovoltaic fault classification, visual anomaly detection in smart manufacturing, redundant industrial control, and non-invasive electromagnetic fault sensing. Together they reflect the breadth of the field and the increasingly tight coupling between advanced signal processing, machine learning, and practical hardware deployment.

Two contributions address diagnostics at the level of the measured signal. Wang and co-workers (Contribution 1) tackle rolling-bearing fault diagnosis, where faint fault features are readily masked by ambient noise. They propose a time- and frequency-domain blind deconvolution method based on a generalized Lp/Lq norm, using a convolutional neural network as the deconvolution filter and identifying suitable value ranges for the norm parameters in each domain. Validation on simulated signals and three real-world bearing datasets shows improved fault-feature extraction and stability relative to existing approaches. Song and co-workers (Contribution 2) turn to series arc faults in low-voltage distribution networks, whose weak characteristics are easily hidden by load currents. Combining variational mode decomposition and singular value decomposition for feature enhancement with an improved LightGBM classifier and a dynamic weighted hybrid loss, their method reaches high accuracy across a large sample set and, once deployed on embedded hardware, delivers alarm times compatible with practical protection requirements.

A further pair of contributions moves from individual signals to data-rich monitoring systems. Kull and co-workers (Contribution 3) present a hybrid diagnostic framework for photovoltaic systems that couples a Deep Belief Network for feature extraction with a LightGBM classifier, trained and validated on a real-world SCADA dataset gathered from a large operational solar plant. Their model distinguishes healthy operation from a range of fault types with high accuracy while reducing training time relative to Transformer and CNN–LSTM baselines, pointing toward scalable, AI-driven monitoring for large solar installations. Wang and co-workers (Contribution 4) address visual anomaly detection on the smart-manufacturing shop floor, where real-time video from multiple cameras must be processed under tight latency and accuracy constraints. Their edge–cloud system, built around AI accelerators and high-speed networking with key-frame extraction, lightweight detection models, and batching and microservice-level parallelism, substantially reduces end-to-end latency and is complemented by a cost-efficient alternative that still meets the required time budget.

The remaining contributions broaden the perspective to system architecture and to the sensing front end itself. Ryu and co-workers (Contribution 5) propose a heterogeneous PLC-based distributed controller with an embedded logic-monitoring “blackbox” for real-time failover and fault detection in industrial control environments. By adopting a flexible Linux-based platform rather than identical redundant hardware, integrating on-site ladder and HMI editing, and validating durability and electromagnetic compatibility through accredited testing, their approach improves mean time between failures, achieves low switchover latency, and enhances fault-detection capability under multi-vendor configurations. Finally, Lingė and Pomarnacki (Contribution 6) provide a structured review of antenna and electromagnetic spectrum-sensing techniques for non-invasive fault detection. Spanning antenna technologies, acquisition platforms, signal-processing methods, and machine-learning approaches, the review connects the full sensing pipeline for partial-discharge, arc, and insulation faults and highlights open challenges such as the lack of standardized calibration and the scarcity of open electromagnetic fault datasets.

Taken together, these contributions illustrate several themes that increasingly define the field: the fusion of physics-informed signal processing with data-driven learning; a consistent emphasis on validation using real-world data and on deployment to embedded or edge platforms; and a growing concern with system-level reliability, standardization, and reproducibility. We hope that this collection offers both a useful snapshot of current progress and a stimulus for further work on condition monitoring and fault analysis in industrial electronics. We warmly thank all of the authors for their contributions, the reviewers for their careful and constructive evaluations, and the editorial staff of Electronics for their support throughout the preparation of this Special Issue.