Extraction of features for any faulty signals is very difficult and critical in any type of fault detection technique. A recent & effective way for feature extraction is a Convolution Neural Network (CNN). This paper firstly collects the data for a faulty line as positive sequences of fault currents then creates scalograms using CWT (Continuous Wavelet Transform). Finally, features of the scalograms will be extracted by CNN. Two types of CNN architecture algorithms are used in this work for fault type detection, i.e., GoogLeNet and AlexNet. Accuracy of AlexNet architecture is better than GoogLeNet architecture for present faults classification problem.

convolution neural network (CNN), continuous wavelet transforms (CWT), GoogLeNet, AlexNet, Scalogram