Modified Training Method For Feedforward Neural Networks And Its Application in 4-Link Scara Robot Identification

Abstract

In this research the results of applying Artificial Neural Networks with modified activation function to perform the online and offline identification of four Degrees of Freedom (4-DOF) Selective Compliance Assembly Robot Arm (SCARA) manipulator robot will be described. The proposed model of identification strategy consists of a feed-forward neural network with a modified activation function that operates in parallel with the SCARA robot model. Feed-Forward Neural Networks (FFNN) which have been trained online and offline have been used, without requiring any previous knowledge about the system to be identified. The activation function that is used in the hidden layer in FFNN is a modified version of the wavelet function. This approach has been performed very successfully, with better results obtained with the FFNN with modified wavelet activation function (FFMW) when compared with classic FFNN with Sigmoid activation function (FFS) .One can notice from the simulation that the FFMW can be capable of identifying the 4-Links of SCARA robot more efficiently than the classic FFS