Using graph based knowledge representation approach in diagnostics of power transformer

Summary

Transformers are important components of the electrical supply system. A transformer failure can lead to heavy damage on the equipment and its environment and is therefore associated with high costs. Additionally a huge portion of the transformer fleet has been in operation for decades. To overcome this situation a reliable condition assessment is necessary. Focusing on the transformer diagnostic process with a system theoretical approach allows the development of novel diagnostic and condition assessment models. A diagnostic approach using graph based knowledge representation is an important part of this development process. This paper describes the creation of the knowledge base and its application in transformer diagnostics. Due to routine tests, a lot measurement information about the transformer condition is gathered. However, most of existing diagnostic methods are focused on particular measurement methods like dissolved gas analysis. This implies that available information remains unused in diagnostic processes. Solely health index methods tend to use as much information as possible. Unfortunately, health index methods cannot handle missing information, which makes this technique inflexible. A graph based knowledge representation allows a flexible and general applicable condition assessment. Flexibility is achieved by the determination of confidence values for the diagnostic statements. Each confidence value is determined by several aspects like the number and suitability of available measurement procedures for detecting a certain failure. The correlations referring to these aspects are provided by knowledge graphs. A graph based knowledge representation offers two main advantages. The first is an easy human readability and ascertainability. Which is the requirement for enhancing and improving the knowledge base and with it the diagnostic quality. A second advantage is the computer-based interpretability. The latter is important for getting efficient diagnostic statements by appropriate algorithms. Consequently, the presented approach is a necessary design step in development of capable diagnostic models for supporting utility experts.

Additional informations

Keywords

Transformer, Diagnostics, Graph Based Knowledge