Ensuring transformer health and accurate fault diagnosis is crucial for the reliable operation of power systems. Development of data-driven techniques for fault interpretation in mineral oil filled transformers becomes challenging due to limited availability of real-world data. The research investigates on development of a novel optimized synthetic data dataset for three different ML models—K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF) that maximizes the accuracy of these data driven algorithms without training with excessive data instances resulting in overfitting on the training dataset. Utilizing a dataset from 1135 diverse transformers for ML model training, the study introduces a novel two-step iterative and optimized methodology for generating a synthetic database. The integration of real and synthetic data enhances the overall efficacy of incipient fault identification using ML algorithms. To ensure robust evaluation and comparison of performance, the IEC TC 10 dataset is employed. With optimized dataset, the accuracy of the KNN model increased from 79.33 % to 90.26 % when the prior was trained only with real-world data. The verification of the generated synthetic data from the proposed method, compared to existing methods, demonstrated its superiority in dataset quality.
25 March 2024 in IEEE Transactions on Dielectrics and Electrical Insulation
Ensuring transformer health and accurate fault diagnosis is pivotal for the reliable operation of power system network. Traditional incipient fault diagnostic methods, including DP-1, DP-2, DT-1, DT-4, and IEC, have been employed extensively, but with varied levels of accuracy for different fault categories. In this research, authors introduce a four-gas diagnostic method, an innovative graphical approach that emphasizes analyzing only four gases, with satisfactory diagnostic accuracy while ensuring lower computational burden and higher interpretability. By adopting a set of two equations for graphically determining fault category, the method offers an intuitive understanding of incipient faults in an oil filled transformer. For this study, a total of 963 different transformer DGA data were utilized. A study was undertaken using the IEC TC 10 database results to assess the effectiveness of the proposed diagnostic method across diverse fault categories. The proposed 4GM method achieved an execution time of 1.46 seconds, demonstrating better performance in detecting discharge and high thermal faults, while its efficacy in identifying low thermal faults was slightly depleted. The proposed 4GM method stands out as it achieves a balance between accuracy and execution time. This research not only offers insights into the strengths and innovations of the 4GM method but also sets the stage for future advancements in transformer diagnostic techniques.
23 June 2025 in IEEE Transactions on Dielectrics and Electrical Insulation
Ensuring transformer health and accurate incipient fault diagnosis is crucial for the reliable operation of power systems. Limited real-world data is a challenge for developing data-driven techniques which generally have higher diagnostic accuracy when compared to traditional fault interpretation techniques. However, due to unavailability of ample data on different fault conditions, the diagnostic accuracy of these data-driven techniques varies significantly for different fault conditions. Utilizing a dataset from 1415 diverse transformers for machine learning (ML) model training, the study introduces a novel two-step iterative and optimized methodology for generating a synthetic database. The integration of real and synthetic data enhances the overall efficacy of fault identification using ML algorithms. To ensure robust evaluation and comparison of performance, the IEC TC 10 dataset is employed. The newly proposed approach achieved its highest accuracy of 94.01% with the deep learning model, surpassing the performance of the existing work. Additionally, the training run time before and after implementing Modified Synthetic Minority Over-Sampling Technique (M-SMOTE) is compared, showing a decrease from 3.28 seconds to 3.01 seconds, highlighting the efficiency gained using M-SMOTE.
2 February 2024 in International Journal of Emerging Electric Power Systems
Fostering green sustainable transportation in the Indian food supply chain with the adoption of electric trucks (ETs) is seemingly a promising alternative. However, before widescale adoption, a thorough cost-benefit analysis is essential for making an apt decision towards transitioning to ETs. Utilizing data of 13 ETs, a total cost of ownership (TCO) model has been developed that outlines the economic benefits of ETs by evaluating the impact of factors like energy consumption, charging strategy, annual driven distances, route optimization, etc. With acquisition costs of ETs constituting 50–55 % of TCO the study outlines that ETs have significantly lower energy costs of 6–12 % (50–58 % for IC-engine trucks) and maintenance costs of 6–8% (13–18 % for IC-engine trucks), thereby reducing the overall TCO. Subsequently, a fleet-level analysis of ET adoption in a food supply chain comprising of four heavy-duty trucks and 16 medium-duty trucks has been performed with emphasis on assessment of supply chain-specific parameters like battery-mass penalty, level of utilization, etc. Overall, this study provides insight for fleet managers, policymakers and industry leaders to promote sustainable through an extended use of ET in the present scenario for the food supply chain.
1 September 2025 in International Journal of Environmental Sciences
Study proposes a new method for predicting the remaining useful life (RUL) of lithium-ion batteries using the Extreme Learning Machine (ELM) algorithm by employing advanced feature extraction techniques to identify critical parameters influencing battery degradation. The model captures intricate patterns in battery health, enabling precise RUL prediction by analyzing voltage, current, temperature, and capacity data. A correlation investigation executed in this paper emphasizes the importance of designated variables, for example, discharge capacity dwindle and inner resistance, in identifying lifecycle of the battery. The ELM is an advanced machine learning algorithm, which has high speed for problem-solving capability, and is used to design a vigorous prognostic approach. Experiment-based results analysis validates that the suggested method accomplishes higher correctness and computational speed compared to conventional approaches. The algorithm's enactment is authenticated via several performance matrices (i.e., AE, MAPE and R2), validating model’s reliable enactment in concrete actual circumstances. Presented study delivers an appreciated algorithm for BLI health diagnosis, augmenting the consistency and reliability of BLI in EVs, RES, and further uses. The outcomes emphasize the impending of AI/ML methods in evolving analytical maintenance for ESS.