Application of neural networks augmenting Taguchi's orthogonal arrays for simulating kinematics problems of robotics

Abstract

This research focuses on integrating back propagation with the concepts of Taguchi's orthogonal arrays to map both robot forward kinematics and robot inverse kinematics. Traditionally, end users were required to have some experience with robot kinematics training programs in order to analyse the kinematics and train the robots. When the type of robot changes, end users must again analyse the robot kinematics and solve extensive mathematical equations in order to control the new robot. An alternative approach for developing robot manipulator models is to use the concepts of neural networks to approximate robot kinematics to overcome the drawbacks of the existing methods. Moreover, an additional benefit is derived from the use of Taguchi's orthogonal arrays. This study shows that using orthogonal arrays for data collection can reduce the amount of data required to map robot kinematics, although some accuracy is lost with solving the inverse kinematics problems. The simulations used to verify the model developed have been conducted on the Jumbo Drilling robot.