1) Artificial neural networks can be used for fault diagnosis of power transformers based on dissolved gas analysis. Different ANN models are trained to classify fault types based on inputs of different dissolved gas ratios from techniques like Rogers ratio and Duval triangle. 2) A smart fault diagnostic approach uses the outputs of each ANN model to make a normalized decision on fault type. This includes classifications like no fault, thermal fault, arcing, partial discharge, and undetermined fault. 3) The smart fault diagnostic approach can be enhanced by adding another ANN that is trained directly on raw gas concentration data, and integrating its output with the existing approach to improve accuracy of fault classification.