Background:
Power transformers are one of the most applicable electricity network devices
which transmit output power of the generator to the network through increasing voltage and
decreasing current. Due to high cost of such devices and cost of disconnecting device upon failure,
disconnection and failure of the transformer should be avoided as much as possible.
Objective:
In addition, in order to increase reliability and reduce maintenance costs, such devices
should be monitored constantly. Internal faults ionize and warm up oil and as a result, gases like
carbon dioxide, methane, ethane, ethylene and acetylene are produced. Various methods have been
proposed for diagnosing fault in power transformers where one of the most well-known methods is
dissolved gas analysis (DGA). DGA in oil is one of the effective tools for diagnosing initial faults in
transformers.
Methods:
Common fault detection methods using oil-dissolved gas analysis include Dornemburge,
Duval’s triangle, IEC/IEEE standard, key gases and Rogers. In recent years, artificial intelligence
like genetic algorithm, fuzzy logic and neural networks have been used to detect faults using DGA.
In this paper, support vector machine (SVM) and decision tree are used to detect internal faults in
power transformers.
Results:
By evaluation of the proposed methods, total accuracies of classifiers using SVM and decision
tree were 90% and 97.5%, respectively.
Conclusion:
Decision tree shows better performance and it is suggested as a proper method for obtaining
promising results.
Cognitive artificial-intelligence for doernenburg dissolved gas analysis interpretation
