Szczegóły publikacji
Opis bibliograficzny
Real-time generation of synthetic partial discharge images for AI-assisted diagnostics / Marek FLORKOWSKI, Paweł MIKRUT // Expert Systems with Applications ; ISSN 0957-4174 . — 2026 — vol. 331 art. no. 133263, s. 1–12. — Bibliogr. s. 11–12, Abstr. — Publikacja dostępna online od: 2026-06-13
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 168576 |
|---|---|
| Data dodania do BaDAP | 2026-06-29 |
| Tekst źródłowy | URL |
| DOI | 10.1016/j.eswa.2026.133263 |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | Expert Systems with Applications |
Abstract
As electricity is increasingly recognized as the dominant energy carrier of the future, high-voltage (HV) electrical insulation has become a critical enabling technology. Since partial discharge (PD) activity often precedes catastrophic insulation failure, its early and accurate detection is essential for ensuring system reliability, minimizing unplanned outages, and extending asset lifetime. AI-assisted PD diagnostics is therefore strategically important for enabling scalable monitoring, data-driven decision-making, and predictive maintenance. However, the application of AI techniques to PD diagnostics in HV insulation systems is often limited by the lack of sufficiently large and high-quality measurement datasets representing insulation systems across multiple application domains. In addition, the infrequent occurrence of certain defect categories leads to class imbalance, which adversely affects the generalization capability of data-driven diagnostic models. In this context, synthetic data generation has emerged as an effective means of augmenting PD datasets and supporting AI-assisted diagnostic methods. To address these challenges, this study focuses on the real-time synthetic generation of phase-resolved PD (PRPD) images. The proposed approach integrates physics-based and stochastic models of PD phenomena with hardware-based PD pulse generation implemented on a real-time simulation platform (OPAL-RT). The presented approach is general and is capable of simulating a wide range of defect types associated with insulation systems across different classes of power equipment. This methodology is particularly relevant for bridging the gap between laboratory-scale PD studies and real-world operating conditions, where signal variability and noise significantly affect diagnostic performance. The experimental results are demonstrated for the most typical defect in the form of a gaseous inclusion. Notably, the proposed method enables controllable and repeatable synthesis of statistically consistent PD patterns, which is essential for rigorous benchmarking of AI-assisted diagnostic models. The key advantage of the proposed real-time, hardware-supported synthetic PD data generation approach lies in the inclusion of the complete PD detection and acquisition chain within the generation loop, thereby going beyond purely software-based methods. The resulting synthetic datasets can be employed for data augmentation, pretraining of deep learning models, and systematic evaluation of diagnostic robustness under varying operating conditions. Consequently, the presented synthetic data generation methodology constitutes a crucial enabler for reliable and scalable AI-assisted partial discharge diagnostics.
