A Customised LSTM-Based Deep Learning Framework for Transformer Predictive Maintenance: Performance Analysis
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Transformers are critical and costly components of power systems whose health deteriorates over time due to factors such as poor cooling and heavy loading. Consequently, predictive maintenance is emerging as an effective alternative to conventional corrective maintenance, enabling continuous monitoring and early fault detection. To enhance the effectiveness of predictive maintenance for power transformers under limited Dissolved Gas Analysis (DGA) data conditions, this study proposes a customised Long Short-Term Memory (C-LSTM) deep learning model. The developed C-LSTM architecture is specifically designed to address the limitations of conventional LSTM networks, which often exhibit higher classification error rates when trained on small datasets and may underperform compared to traditional machine learning approaches. A comprehensive performance evaluation was conducted by comparing the proposed C-LSTM model with several well-established traditional machine learning algorithms using multiple metrics, including validation accuracy, test accuracy, precision, recall, and F1-score. Additionally, the diagnostic capability of the model was rigorously assessed across seven transformer fault categories, including low- and high-energy discharges, partial discharge, electrical and thermal faults, and low-, medium-, and high-temperature thermal faults. The experimental results demonstrate the superior classification and diagnostic performance of the proposed C-LSTM model, achieving a validation accuracy of 100% and a test accuracy of 98.57%, significantly outperforming conventional machine learning techniques. These findings confirm that the proposed C-LSTM framework offers a robust and reliable solution for transformer fault diagnosis and predictive maintenance, particularly in scenarios characterised by scarce DGA datasets.
Fields of Study & Subjects
Power Transformer Diagnostics and Insulation
Machine Fault Diagnosis Techniques
High voltage insulation and dielectric phenomena
Publication details
- Journal / source
- Book Publisher International (a part of SCIENCEDOMAIN International)
- Pages
- 98–127
- Publication date
- 14 March 2026
- Language
- EN
- Document type
- Book chapter