5 Fault Diagnosis Techniques for Large Power Transformers

Transformer Fault Diagnosis Methods

1. Ratio Method for Dissolved Gas Analysis

For most oil-immersed power transformers, certain combustible gases are produced in the transformer tank under thermal and electrical stress. The combustible gases dissolved in oil can be used to determine the thermal decomposition characteristics of the transformer oil-paper insulation system based on their specific gas content and ratios. This technology was first used for fault diagnosis in oil-immersed transformers. Later, Barraclough and others proposed a fault diagnosis method using four gas ratios: CH4/H2, C2H6/CH4, C2H4/C2H6, and C2H2/C2H4. In subsequent IEC standards, the C2H6/CH4 ratio was removed, and the modified three-ratio method became widely adopted. Rogers further provided detailed analysis and explanation of the gas component ratio coding and usage methods in IEEE and IEC standards. Long-term application of IEC 599 revealed that it doesn't match actual conditions in some cases and cannot diagnose certain fault scenarios. Consequently, both China and the Japan Electrical Association have made improvements to IEC coding, while other dissolved gas analysis methods have also gained widespread application.

2. Fuzzy Logic Diagnostic Method

American control theorist L.A. Zadeh first proposed fuzzy diagnosis methods, which have since gained broader application. Fuzzy logic is advantageous for expressing qualitative knowledge and experience with unclear boundaries. Using the concept of membership functions, it distinguishes fuzzy sets, processes fuzzy relationships, simulates human rule-based reasoning, and solves various uncertainty problems in practical applications. In practice, transformers often exhibit faults with unclear causes and mechanisms involving numerous uncertain and fuzzy relationships that traditional methods cannot explain or describe well. Fuzzy logic methods can effectively address these uncertain relationships in transformer faults, providing a new approach to power transformer fault diagnosis.

To address the limitation of critical ratio criteria deficiency in the commonly used Rogers ratio method for power transformer fault diagnosis, a method using fuzzy set theory has been proposed. This approach introduces fuzzy logic technology into traditional ratio methods by fuzzifying ratio boundaries. This method has shown good application effects in diagnosing multiple transformer faults and has evolved into a series of fault diagnosis methods, including coding combination methods, fuzzy clustering techniques, Petri networks, and grey systems. These models fully consider the inherent fuzziness of data, effectively improving performance with complex datasets and enhancing the accuracy of transformer fault diagnosis.

3. Expert System Diagnostic Method

Expert systems represent an important branch of artificial intelligence. They are computer program systems capable of simulating human expert experience and reasoning processes to a certain extent. Based on data provided by users, they apply stored expert knowledge or experience to make inferences and judgments, ultimately providing conclusions with confidence levels to assist user decision-making. Power transformer fault diagnosis is an extremely complex problem involving multiple factors.

Making accurate judgments based on various parameters requires solid theoretical foundations and rich operational maintenance experience. Additionally, due to variations in transformer capacity, voltage levels, and operating environments, the same fault may manifest differently across various transformers. Expert systems possess strong fault tolerance and adaptability, allowing them to modify their knowledge base based on acquired diagnostic knowledge to ensure completeness. Therefore, they can effectively diagnose different types of power transformers. Power transformer fault diagnosis expert systems can determine fault characteristics by synthesizing knowledge of fault causes and types, incorporating fault detection knowledge including dissolved gas analysis in oil. They can effectively handle fuzzy problems in fault diagnosis using fuzzy logic, address the bottleneck of difficulty in obtaining complete knowledge through rough set methods, and establish structures suitable for multi-expert collaborative diagnosis using blackboard model architecture.

4. Artificial Neural Network Diagnostic Method

Artificial neural networks mathematically model neuron activity and represent an information processing system based on mimicking the structure and function of brain neural networks. ANNs possess self-organizing, adaptive, self-learning, fault-tolerant capabilities, and strong nonlinear approximation abilities. They can implement prediction, simulation, and fuzzy control functions, making them powerful tools for processing nonlinear systems. Using artificial neural networks for transformer fault diagnosis based on dissolved gas components and concentrations in oil has been a research focus in recent years. This has led to the development of various fault diagnosis methods based on ANNs, such as the two-step ANN method, backpropagation artificial neural networks, decision tree neural network models, combined neural network hierarchical structure models, and radial basis function neural networks. These methods continuously improve the convergence speed, classification performance, and accuracy of neural network algorithms.

5. Other Diagnostic Methods

Beyond the four methods mentioned above, several other approaches are also used for transformer fault diagnosis. By organically combining neural networks and evidence theory to leverage their complementary advantages, a comprehensive transformer fault diagnosis method integrating multiple neural networks with evidence theory can be developed. Drawing inspiration from the efficient recognition and memory mechanisms of antibodies against antigens in biological immune systems, self-organizing antibody networks and antibody generation algorithms can be applied to solve power transformer fault diagnosis problems. Additionally, other transformer fault diagnosis methods include those based on information fusion, rough set theory, combined decision trees, Bayesian networks, artificial immune systems, novel radial basis function networks, and support vector machines.