Multiclass Fault Diagnosis in Power Transformers Using Dissolved Gas Analysis and Grid Search-Optimized Machine Learning
Bibliographic record
Abstract
Dissolved gas analysis remains the most widely utilized non-intrusive diagnostic method for detecting incipient faults in insulating liquid-immersed transformers. Despite their prevalence, conventional ratio-based methods often suffer from ambiguity and limited potential for automation applicrations. To address these limitations, this study proposes a unified multiclass classification model that integrates traditional gas ratio features with supervised machine learning algorithms to enhance fault diagnosis accuracy. The performance of six machine learning classifiers was systematically evaluated using training and testing data generated through four widely recognized gas ratio schemes. Grid search optimization was employed to fine-tune the hyperparameters of each model, while model evaluation was conducted using 10-fold cross-validation and six performance metrics. Across all the diagnostic approaches, ensemble models, namely random forest, XGBoost, and LightGBM, consistently outperformed non-ensemble models. Notably, random forest and LightGBM classifiers demonstrated the most robust and superior performance across all schemes, achieving accuracy, precision, recall, and F1 scores between 0.99 and 1, along with Matthew correlation coefficient values exceeding 0.98 in all cases. This robustness suggests that ensemble models are effective at capturing complex decision boundaries and relationships among gas ratio features. Furthermore, beyond numerical classification, the integration of physicochemical and dielectric properties in this study revealed degradation signatures that strongly correlate with thermal fault indicators. Particularly, the CIGRÉ-based classification using a random forest classifier demonstrated high sensitivity in detecting thermally stressed units, corroborating trends observed in chemical deterioration parameters such as interfacial tension and CO2/CO ratios. Access to over 80 years of operational data provides a rare and invaluable perspective on the long-term performance and degradation of power equipment. This extended dataset enables a more accurate assessment of ageing trends, enhances the reliability of predictive maintenance models, and supports informed decision-making for asset management in legacy power systems.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".