Application of adaptive network-based fuzzy inference system with fast Fourier transform for waveform analysis and classification

This research presents electrical signal waveforms analysis and classification by applying the principle and theory of ANFIS. The input data for training and testing the network were processed by using Fast Fourier Transform. There are three input variables and one output for the network. From the experiment to determine the number of nodes in the 1st layer in order to obtain the optimal Mean Square Errors for analyze signal, the ANFIS learning, training function, genfis1 and learning function, Hybrid were used. The experimental result found that the best model consisted of the number of nodes to 3 models are 3-(6 6 6)-1, 3-(7 7 7)-1 and 3-(4 5 6)-1 input nodes, hidden nodes and output node, respectively. The transfer functions for output layer were linear function. The optimal MSE of training process were 6.62E-09, 3.32E-09 and 3.02E-08. The MSE of the test were 7.19E-09, 3.21E-09 and 2.46E-08, respectively. This provides the optimal percentage of Efficiency Index in the testing process. It showed that the proposed ANFIS can be used in signal pattern recognition in order to analyze and classify between good and bad signals.