Correlations among three performance indices with decision about training parameters based on energy variation in a robot’s part micro-assembly methodologies

摘要

Correlations among three performance indices, including the probability of success, fuzzy entropy, and system energy, with a decision about training parameters based on the energy variation of a neural network system are newly introduced. Methodologies are used based on the robot’s part micro-assembly. The enhancement of efficiency in the performance of the robot’s part-in-assembly hole (target) task implies the maximization of the probability of success as well as the minimization of fuzzy entropy and system energy in the execution of the associated task. Two part micro-assembly algorithms are introduced that bring a part from an initial position to a target for the purposes of part-mating and storage. The two part micro-assembly algorithms are then compared through simulations and chosen specific criteria. Also, the novelty of the methodologies used in this paper is described. In the 1st algorithm, a grid based on a neural network categorizes an assembly type which is fed to a fuzzy coordinator that places the part at the selected position, where it is ready to mate successfully with a target. The energy variation of the neural network system is used as a new tool to find better values of training parameters, such as the learning rate, the number of nodes in a hidden layer, and the momentum. The 2nd algorithm is a fuzziness-minimizing learning algorithm that assists the robot to adapt to its unfamiliar workspace. The 2nd algorithm then finds a plan with the lowest degree of uncertainty among the generated feasible plans composed of a sequence of coordinates from the part’s arbitrary starting to the target positions. The results obtained by the two algorithms show that the three performance indices reach the desirable states such that the fuzzy entropy and the system energy are minimized and the probability of success is simultaneously maximized as trainings are successively reiterated. The results also show that the three performance indices can be a useful tool to estimate the performance results of a robot’s various types of part assemblies.