Research On Robot Assembly Contact State Recognition Method Based On Machine Learning

In recent years,with the continuous expansion of robotic automated assembly in the industrial field,small and complex parts that are traditionally assembled by hand have presented more severe challenges to their compliance.The force-controlled robot has the advantages of high precision,high production efficiency and scalability,and has become a hot topic in scientific research and a hot spot of interest in industrial applications.The contact state is the most important information in the assembly task.Using the machine learning algorithm to learn the contact state from the force data collected during the assembly process becomes the key link of the flexible assembly intelligence.Therefore,the research on the robot assembly contact state recognition based on machine learning has Significance.Firstly,this paper analyzes the basic principles of machine learning classifiers for small sample size,establishes support vector machine and various parameter optimization algorithm models,and analyzes the advantages and disadvantages of support vector machines in detail,and obtains support vector machine parameters.The impact on its classification.Based on this,the working principle of support vector machine parameter optimization is introduced.Secondly,the contact state recognition scheme of rigid/non-rigid parts assembly is introduced and the experimental platform is established.The influence of the material of the assembly parts on the recognition accuracy is analyzed.On this basis,the Mitsubishi robot simulation model is built and designed by using MATLAB robot toolbox.Robotic assembly action for rigid shaft hole parts and non-rigid snap parts.The robot assembly correction system is established by using software such as MATLAB and RT ToolBox2.Based on the system,the offset angle of the parts under different contact conditions is identified,and the corrective pose transformation is completed by the inverse kinematics calculation of the robot.Finally,aiming at the problem of different assembly force data characteristics of different material parts,this paper deduces the process of genetic algorithm and whale optimization algorithm support vector machine parameter optimization,and proposes a support vector machine classifier for hybrid whale-genetic algorithm parameter optimization.The identification of the contact state of the rigid/non-rigid parts is realized,and the recognition accuracy of the rigid part assembly contact state reaches 99.985%.In the case that it is more difficult to identify the non-rigid contact state identification,this paper proposes that the accuracy of the classification of the contact state of the non-rigid parts is 98.850%,which has obvious accuracy and calculation compared with other contact state recognition classifiers.The speed advantage provides a stable and efficient theoretical basis and technical support for the force-controlled robot to assemble complex parts.