Comparative study of the neural and neuro-fuzzy networks for direct path generation of a new fully spherical parallel manipulator

Abstract

ABSTRACT In this paper, the direct kinematics problem of a new 3(RPSP)-S fully spherical parallel manipulator (SPM) is solved using three different models of the artificial intelligence networks, including: a back propagation neural network (BPNN), a radial basis function neural network (RBFNN) and an adaptive neuro-fuzzy inference system (ANFIS). For the proposed networks, we use a training data set which is made from solving the inverse kinematics problem of the robot. After making a database for training the network, different parameters of the networks are changed and finally the best ones for each of BPNN, RBFNN and ANFIS models are derived. Effectiveness of the proposed models is checked by comparing the results of these models with the results of elimination method. As the results show, BPNN has the greatest error and the greatest computational time equal to and 12.4 s, respectively. The next model is RBFNN which has much better precision and less computational time which are equal to and 4.9 s, respectively. Finally, the simulation results by the ANFIS model show that it is the best model for solving the FKP of the 3(RPSP)-S robot. The mean square error (MSE) and computation time of the ANFIS model are and 1.2 s, respectively. These results confirm the reliability of the designed networks.