Forecasting the mountability level of a robotized assembly station

Abstract

This article presents the problem of determining the mountability level of the assembly station using an artificial neural network (ANN). The results of ANN modelling were compared with the results of experimental research and classical mathematical modelling. It was found that the error in predicting the mountability level using the artificial neural network is about two-fold lower than in the case of the error determined by classical mathematical modelling. Although the neural network ensures a lower prediction error, to obtain a good prediction it is necessary to conduct many experiments in the whole workspace of the robots to build a training set. Despite the worst prediction, a mathematical model of the mountability level only requires an analytical description of the kinematic structure of the assembly robot, so in industrial applications this is preferred due to the lower labour requirement.