A novel optimization robust design of artificial neural networks to solve the inverse kinematics of a manipulator of 6 DOF

In the design of neural networks, generally the selection of the structural parameters is chosen through trial and error procedures, consuming large amounts of resources and unavailable time, without guaranteeing the optimal configuration of the parameters that allow obtaining the best performance of the network. In this paper, the robust design methodology of artificial neural networks based on the Taguchi philosophy was used to select the optimal parameters in a back-propagation network architecture to solve the inverse kinematics in a 6 degrees of freedom robotic manipulator. The parameters to optimize were the number of hidden layers, the number of neurons per layer, the learning rate, the momentum, the number of neurons per layer and the size of the training set versus the test set. Allowing to identify all the combinations possible in relation to the number of variables involved by performing a significant number of experiments compared to other methods where they usually run a huge number of experiments. The results obtained allowed to optimize the design parameters and substantially improve the precision of the results, achieving a prediction percentage of 90% with a margin of error less than 5% during the testing stage.