Research On Technologies Of Collaborative Robot’s Collision Detection Based On Parameters Identification

Driven by the Made in China 2025 policy,various production fields have put forward higher requirements for intelligent manufacturing,which has led to a significant increase in the demand for collaborative robots.Conventional collaborative robots usually use position control mode,which is highly dependent on external sensors for control accuracy,and it is difficult to directly compensate for torque control.In this paper,we study a collaborative robot without external sensors,aiming to realize the human-robot collaboration function through control algorithms and collision detection based on a identification dynamics model.The details are as follows:(1)Set up the coordinate system and establish the dynamics model by applying the improved DH method for the experimental collaborative robot.The dynamic identification model is constructed and linearized to determine the minimum parameter set,and an improved genetic algorithm is proposed for excitation trajectory optimization to improve the convergence speed of trajectory parameters.With the moment sampling noise as the weight,the weighted least squares method is used to determine the value of the minimum parameter set.From the perspective of practical applications,the Cullen viscous friction model is used to construct the complete dynamics model,and the accuracy of the model is evaluated by the root mean square(RMS)of the moment residuals.(2)The full-parameter excitation trajectory identification method is used for load identification.Considering the high coupling of wrist joints,the first three joints are selected for identification to reduce the moment noise.The accuracy of the load recognition model is demonstrated by the load mass deviation and the root mean square of the moment residuals,and the experiment verifies that the method has high recognition accuracy,fast recognition speed and high application value.(3)The collision detection is implemented based on the recognition model of robot and load.A method is proposed to set dynamic thresholds for collision detection based on time series algorithm(TSA),which predicts external moments by an autoregressive(AR)model and updates the model using recursive least squares(FRLS)with forgetting factors,and sets upper and lower limits of the predicted values as collision thresholds,which is experimentally verified to have higher sensitivity than the traditional fixed threshold method.(4)A collision detection experimental platform is built,on which the kinetic parameter identification and collision detection algorithms are verified.In order to determine the actual effect of different collision response strategies,three strategies of stopping,retreating and zero force control are designed and analyzed in comparison with experiments,and different response strategies are selected according to different scenarios in practical applications to achieve the best effect.