Research On Low Voltage Electrical Apparatus Clamping Assembly Technology Based On Deep Q Learning

The assembly process of small electrical apparatus,represented by low voltage electrical apparatus,has great uncertainty in assembly process because of its small size,many kinds,complex shape and different models,and complex relationship between electronic and electrical appliances.The assembly process is very uncertain,and the robot's operation space is relatively small,which puts forward higher intelligent requirements for flexible assembly process.Deep reinforcement learning has a good ability of perception and decision.It is an important theoretical basis for realizing the intellectualization of industrial robots,and it has become a hot topic in the current research of robot technology.In this paper,based on the theory of robot technology and deep reinforcement learning,an active force perception model based on deep Q learning is established,and the algorithm is verified in the Gazebo simulation environment.The main works of this paper are reflected in the following aspects:1.This paper analyses the compliant assembly system of low voltage electrical apparatus.Firstly,the kinematics modeling of the LBR-iiwa robot with seven degrees of freedom is given.The kinematics model provides the premise for the motion control of the manipulator;the assembly process of the low voltage electrical apparatus,represented by the small circuit breaker,is briefly introduced,and the assembly contact state is analyzed in detail,which lays the foundation for the later design model and the algorithm research.2.A deep Q learning algorithm based active force perception model is established.Based on the analysis and elaboration of the theory of deep Q learning algorithm,the characteristics of clamping and assembling of small circuit breaker's upper cover and pedestal are analyzed,the following research is carried out:(1)A Markov model for assembly process.The force information feedback of the manipulator's terminal posture is determined by the contact between the upper cover and the pedestal cover of the breaker.In view of this characteristic,the environmental model of compliant assembly system is designed,the action space of manipulator is discretized,and the continuous motion of mechanical arm is decomposed into four actions;Six axis force sensor collects manipulator assembly state information to get six dimensional assembly state model;The evaluation model based on SVM is studied.On this basis,the reward function of assembly algorithm is designed to guide the training of active perception model.(2)In this paper,we study the deep Q learning assembly algorithm,analyze and process the feedback state data and return value,and make decisions about this.The strategy network of assembly action is established.We use the experience pool replay mechanism to train it,and reduce the correlation between samples collected;The network parameters are updated by minimizing the mean square error between the current Q value and the target Q value.After introducing the target value network,the Q value of the target is kept unchanged for a period of time,and the stability of the algorithm is improved.Finally,the algorithm pseudo code is given,which lays the foundation for algorithm validation.3.Taking the assembly of HYB1-63 circuit breaker in the assembly process as the research object,the Gazebo flexible assembly simulation environment under the ROS environment is built,and the deep Q learning assembly algorithm is verified.The experimental results show that the built simulation environment can effectively simulate the real assembly process,and the established active force perception model can feed back the assembly state online and realize the clamping assembly.Finally,we summarize the work done in this paper and get experience and results,and analyze the shortcomings of this article and the problems that need further solution.