Comparison of power transformer sfra sweep frequency response analysis results with artificial intelligence applications

Electric energy has emerged as one of the most indispensable resources of the 21st century, playing a pivotal role in the economic, technological, and social development of nations. In the context of a rapidly evolving global landscape, the continuity and sustainability of modern life are fundamentally intertwined with the reliable generation, transmission, and utilization of electric power. Today, electricity is not merely a convenience but a critical enabler of progress, shaping the way people live, work, and interact with the world around them. From residential usage to large-scale industrial operations, from agricultural automation systems to modern transportation networks, electric energy permeates nearly every domain of human activity. The exponential growth of technology has resulted in an increasing dependency on electrical systems. The proliferation of electrically powered appliances, machines, and systems—such as smart home devices, HVAC (heating, ventilation, and air conditioning) systems, electrically driven manufacturing processes, and medical diagnostic and therapeutic equipment—has fundamentally transformed the infrastructure of contemporary society. One of the most notable developments in recent years has been the widespread adoption of electric vehicles (EVs), which has introduced a new dimension to energy demand and grid management. As EVs become increasingly common in urban and rural areas alike, they place additional demands on the electrical infrastructure. This shift toward electrified transportation systems not only contributes to reduced greenhouse gas emissions but also underscores the need for a resilient, stable, and intelligent energy grid capable of handling dynamic and high-load conditions. In the specific case of Turkey, a clear illustration of this transformation can be observed in the dramatic increase in per capita electricity consumption. While electricity usage per individual was limited to just a few kilowatt-hours at the beginning of the 20th century, it has now surged to approximately 3900 kWh per year, signifying a rapid trajectory of electrification and industrialization. This steep rise in energy consumption reflects not only increased access to electricity but also the growing technological sophistication and urbanization of the country. However, the growing demand for electricity comes with considerable technical and operational challenges, particularly concerning the reliability and stability of the power transmission and distribution network. At the heart of this network lie transformers, which serve as critical components for voltage regulation and efficient power delivery over long distances. These devices ensure that electricity generated at power plants can be safely and effectively transmitted to end users, regardless of geographic location. Despite their central role in the grid, transformers are susceptible to degradation and failure over time due to several contributing factors. Prolonged operation under heavy loads, exposure to adverse environmental conditions (such as moisture, dust, and temperature fluctuations), natural aging of insulation and other materials, and unforeseen external disturbances (including short circuits or lightning strikes) can all compromise the structural integrity and operational efficiency of transformers. If such issues are not identified and addressed in a timely manner, they can lead to unexpected outages, safety hazards, and significant economic losses. Therefore, the early detection and accurate diagnosis of transformer faults is of utmost importance for maintaining the overall health of the electrical grid. Traditional diagnostic techniques, such as dissolved gas analysis (DGA), insulation resistance testing, and other offline and online methods, have been widely adopted across the industry. However, while these conventional approaches are useful in identifying certain fault types, they may not always be effective in detecting incipient or internal faults, particularly during the early stages when symptoms are subtle and not easily discernible. This limitation underscores the necessity of complementing traditional test methods with advanced diagnostic tools and technologies. By doing so, utilities and energy providers can shift from reactive maintenance approaches toward more proactive and predictive maintenance strategies, enhancing both operational reliability and cost-effectiveness. At this point, the Sweep Frequency Response Analysis (SFRA) method plays a crucial role. SFRA testing involves applying low-amplitude signals between a transformer's input and output terminals to measure its frequency response characteristics. The resulting Bode diagrams reveal changes within the internal structure of the transformer. Issues such as core loosening, winding displacement, and short circuits can thus be identified. However, interpreting these diagrams generally requires experience and expertise. For this reason, the integration of Artificial Intelligence (AI) and Machine Learning (ML) algorithms with SFRA data marks a new era in fault diagnosis. Machine learning algorithms can successfully perform pattern recognition and classification tasks based on large datasets. This study addresses three primary conditions in transformers: core faults, winding displacement faults, and healthy status. Six different algorithms were tested: Logistic Regression, Random Forest, k-Nearest Neighbors, Support Vector Machines, Naive Bayes, and Gradient Boosting Classifier (GBC). As a result of the study, the GBC algorithm demonstrated superior accuracy rates. This algorithm efficiently processes large volumes of data and minimizes the likelihood of false positives. Its performance was evaluated through confusion matrix analysis and various graphical methods. These include Learning Curve, Calibration Curve, Gain Chart, Cumulative Accuracy Profile, and KS statistics. Through these tools, the learning capacity, stability, and generalization capability of the algorithm were analyzed in depth. The Gradient Boosting Classifier (GBC) stood out among the tested algorithms for its remarkable ability to produce consistent and accurate results, even when trained on relatively small and imbalanced datasets. This capability is particularly valuable in real-world transformer fault diagnosis scenarios, where obtaining large quantities of labeled fault data can be both time-consuming and costly. The robustness of GBC against data limitations underscores its adaptability and efficiency in practical applications. During the training phase, hyperparameter tuning was meticulously performed to ensure optimal model performance. Techniques such as grid search and randomized search were employed to identify the most suitable values for parameters like learning rate, number of estimators, and tree depth. Additionally, k-fold cross-validation was used to evaluate the model's ability to generalize to unseen data, thus minimizing the risk of overfitting. As a result, the GBC model demonstrated both high precision and recall, contributing to a well-balanced and reliable classification system. One of the most significant advantages of incorporating such machine learning models into fault diagnosis processes is the reduction of subjective interpretation. Traditional diagnostic methods, particularly those relying on SFRA test results, often depend on the experience and judgment of expert engineers to interpret frequency response curves. This human-dependent process introduces variability and potential inaccuracies. In contrast, AI-based models can provide standardized, repeatable, and data-driven insights, reducing the room for error and improving the overall reliability of diagnostics. From a broader perspective, the integration of AI-driven diagnostic systems into routine transformer maintenance holds transformative potential. In particular, it paves the way for the adoption of predictive maintenance strategies, which focus on anticipating failures before they occur rather than responding after the fact. Predictive maintenance leverages real-time and historical data to monitor the health of equipment continuously, identify early signs of deterioration, and schedule maintenance activities at the most cost-effective and operationally convenient times. This proactive approach not only minimizes unplanned outages and maintenance costs but also helps extend the service life of critical assets like transformers. Furthermore, as energy transmission and distribution networks become increasingly digitalized, the incorporation of AI-supported diagnostic tools is rapidly becoming essential. These tools can be integrated into smart grid infrastructures, enabling automated, continuous monitoring of transformer conditions and real-time decision-making. By embedding machine learning models directly into the grid's control systems, utilities can optimize load balancing, reduce downtime, and enhance system resilience against unexpected disruptions. In conclusion, this study clearly illustrates the transformative impact that Artificial Intelligence and Machine Learning technologies can have on the field of power transformer diagnostics. The synergy between a robust testing methodology like Sweep Frequency Response Analysis (SFRA) and advanced machine learning algorithms enhances the accuracy, speed, and objectivity of fault detection processes. Among the evaluated models, the Gradient Boosting Classifier demonstrated exceptional performance, setting a benchmark for future research and industrial implementations. As the global demand for electric energy continues to grow, ensuring the stability, efficiency, and resilience of energy systems has never been more critical. Under these circumstances, reliance on conventional testing methods alone is insufficient. Instead, investing in intelligent, data-driven diagnostic systems has become an operational necessity. Such systems not only help utilities respond faster to emerging faults but also enable them to anticipate and prevent failures, thus ensuring uninterrupted energy delivery. Finally, it is important to highlight that academic research of this nature serves a dual purpose. On one hand, it contributes to the theoretical understanding of machine learning applications in electrical engineering. On the other hand, it offers practical, scalable solutions that can be directly applied in the field. For mission-critical infrastructure like transformers, the use of AI-enhanced fault detection methods promises greater operational efficiency, reduced maintenance costs, and improved asset longevity. In the future, such technologies will play an indispensable role in shaping a more intelligent, sustainable, and resilient energy infrastructure.