A sheet metal positioning method for reducing the reversal of industrial robot joints

Abstract

The present invention discloses a plate positioning method for reducing the reversal of the joints of an industrial robot. The method obtains the change of the joint angles of the axes of the industrial robot during material transportation through the planning and analysis of the initial path, and then reduces the number of joint reversals of the axes according to the reinforcement learning algorithm. The optimization goal is to reach the target point with the minimum number of joint reversals required for the movement of each axis. The method reduces the absolute positioning error caused by the backlash of the reducer without changing the relevant hardware structure and accuracy, thereby improving the absolute positioning accuracy of the industrial robot and reducing the positioning error when the plate is placed on the embossing table. The method avoids subsequent multiple manual interventions while also improving the accuracy of positioning, promotes the smooth progress of the next step of embossing and subsequent production links, and improves production efficiency.

Description

Technical Field

[0001] The invention relates to the field of industrial robots, and in particular to a plate positioning method for reducing the reversal of joints of industrial robots. Background Art

[0002] In the current door manufacturing process, there are different offset patterns in the sheet metal embossing process. The effect of the so-called offset pattern process is achieved by the position offset when the sheet metal is placed before embossing. Therefore, it involves the positioning accuracy of the sheet metal, and the positioning error needs to be strictly controlled.

[0003] Currently in the industrial field, there are two main ways to position sheet materials. One is manual placement and positioning. In addition, industrial robots are also used to transport sheets by suction cups at the end operating points, and then move to the corresponding position according to the preset program to place and position the sheets.

[0004] The errors caused by purely manual loading and positioning are determined by manual operation, which requires extremely high technical level of personnel and has extremely low efficiency, resulting in a large waste of human and material resources.

[0005] When using industrial robots, absolute positioning errors will occur during movement. One of the sources of such errors is joint reversal, that is, when the joints alternate between forward and reverse rotations, there will be a backlash in the reducer, which will affect the absolute positioning accuracy. If the joints reverse too many times, it will cause a large absolute positioning error. This will cause the actual position of the sheet to deviate too much from the expected position when the deflection process is changed, resulting in the inability to perform the next embossing operation or the embossed sheet will not meet the accuracy requirements and be scrapped.

[0006] Therefore, the current industrial sheet positioning process still requires manual intervention, that is, after changing the deflection process, professional staff are still required to perform manual measurement and calibration when the sheet is placed. If the error is too large, multiple fine-tuning will be required. When purely manual loading is performed, the sheet position needs to be manually fine-tuned, and when using an industrial robot for loading, the relevant setting parameters of the industrial robot need to be manually fine-tuned and then verified through multiple actual placements, which will increase manpower and material resources and reduce production efficiency. Summary of the invention

[0007] The purpose of the present invention is to propose a sheet metal positioning method that reduces the reversal of the joints of industrial robots in response to the deficiencies in the prior art. The method of the present invention can effectively reduce the occurrence of joint reversals by analyzing the motion path of the end operating point of the industrial robot when the embossing of the sheet metal needs to change the eccentric pattern process, that is, reduce the number of positive and reverse rotations of the joints, reduce the positioning error when the sheet metal is placed on the embossing table, avoid subsequent multiple manual interventions, and also improve the accuracy of positioning, promote the smooth progress of the next step of embossing and subsequent production links, and improve production efficiency.

[0008] The object of the present invention is achieved through the following technical solution: a sheet metal positioning method for reducing the reversal of the joints of an industrial robot, comprising the following steps:

[0009] Step 1: Establish the DH model of the industrial robot;

[0010] Step 2: When changing the offset process, adjust the end point of the end operation point in the workspace of the industrial robot according to the new offset distance, which is recorded as o x The remaining points can be used as before the deflection process is changed. The starting point is o1, and the i-th point among the intermediate points is o i , where i∈(1,x), a total of x points;

[0011] Step 3: Using the reinforcement learning algorithm, the optimization goal is to reduce the number of joint reversals of each axis during the process of the industrial robot transporting the sheet material, that is, the minimum number of joint reversals required for the movement of each axis to reach the target point. During the reinforcement learning process, the state is the posture of the end operation point of the industrial robot, and the action is the rotation angle of the joint forward and reverse rotation of each axis. The reward of a single axis joint rotation minus the penalty when the axis joint reverses or when the axis rotation angle exceeds the limit is used as the instantaneous reward for reinforcement learning to obtain a planning path that meets the optimization goal.

[0012] Furthermore, step 1 includes the following sub-steps:

[0013] (1.1) According to the link coordinate system convention of the DH modeling process, establish the link coordinate system of each axis of the industrial robot;

[0014] (1.2) Use the four parameters of joint angle, rod length, torsion angle and horizontal distance to describe the motion of the industrial robot rod and create a DH parameter table for subsequent path analysis and planning;

[0015] (1.3) The homogeneous transformation matrix from the end operation point coordinate system to the robot base coordinate system is calculated from the DH parameter table to obtain the DH model of the industrial robot.

[0016] Furthermore, step 3 includes the following sub-steps:

[0017] (3.1) The reinforcement learning algorithm uses the Sarsa algorithm. The Sarsa algorithm uses the Q table as a reference for selecting action strategies. Therefore, the Q table needs to be initialized. The table has states as rows and actions as columns. The Q value in the Q table represents the value of an action under a specific state.

[0018] (3.2) At a certain time t, the initial state s t ,s t is the starting position of the end operation point of the industrial robot, and then according to the ε-greedy strategy, from the current state s tSelect an action a from all possible joint positive rotation ω and joint reverse rotation -ω actions t ,ω is the minimum rotation angle that can be achieved by each axis of the current industrial robot;

[0019] (3.3) Action a selected by the ε-greedy strategy t , a t is the forward rotation ω or reverse rotation -ω of a joint of a certain axis at the current moment. In the current state s t Next, perform action a t , get the state s t+1 ,s t+1 is the next position of the end operation point of the industrial robot, and also has an instantaneous feedback r t , r t It consists of two parts, t =R t -P t , R t is the reward for a single axis joint rotation, and P t Penalty for when the axis has joint reversal or when the axis rotation angle exceeds the limit;

[0020] (3.4) In the current state s t+1 Select an action a from all possible joint forward rotation ω and joint reverse rotation -ω actions of each axis joint according to the ε-greedy strategy t+1 ;

[0021] (3.5) According to the new Q value Q(s t+1 ,a t+1 ), update the value function as well as the state and action, learn and find the strategy of forward and reverse rotation of each axis joint to obtain the maximum reward. If the current state is not the final state, repeat the three steps (3.3) to (3.5) to obtain the maximum cumulative reward in the process of sheet material transportation and positioning. T is the total material transportation time;

[0022] (3.6) until s is the final state, which is the final position of the end operation point of the industrial robot. At this time, Q(s,a) converges and the iteration is completed. When Q(s,a) converges, the ε-greedy strategy also converges and the optimal strategy is obtained. The planned motion path is obtained by analysis.

[0023] Furthermore, in step (3.2), the ε-greedy strategy is as follows:

[0024]

[0025] Where Q(s,a) is the value function; ε is 0.1, which indicates the exploration rate. There is a probability of 1-ε to select the action with the greatest value from the forward and reverse rotation of the joints of each axis, and there is a probability of ε to act randomly; A is the action set of the forward and reverse rotation of the joints of each axis; m is the number of all optional actions for the forward and reverse rotation of the joints of each axis; represents the probability of selecting action a in the forward and reverse behaviors of each axis with a probability of ε; a * It is the action with the greatest value at present, that is, the joint of a certain axis rotates forward or reversely. It may be selected in two cases. One is that it is selected when the action that is currently considered to have the greatest value is selected with a probability of 1-ε, and the other is that it is selected when it is randomly selected.

[0026] Furthermore, in step (3.5), the value function, state, and action are updated as follows:

[0027] Q(s,a)=Q(s t ,a t )+α[r+γQ(s t+1 ,a t+1 )-Q(s t ,a t )]

[0028] s=s t+1 , a=a t+1

[0029] where r = r t , γ is the attenuation factor, α is the iteration step, and α should gradually decrease as the iteration proceeds to ensure that the value function can converge.

[0030] Furthermore, after obtaining the planned path that meets the optimization goal, the number of joint reversals of each axis is obtained through statistical analysis based on the changes in the movements of each axis in the reinforcement learning Q table. The positioning evaluation index of the planned path is obtained by summing up the number of joint reversals of each axis.

[0031] Beneficial effects of the present invention: The present invention reduces the occurrence of joint reversal in the material transport path of the industrial robot, that is, reduces the number of forward and reverse rotations of the joints, thereby effectively reducing the absolute positioning error caused by the reducer backlash generated thereby, and improves the absolute positioning accuracy of the sheet placement and positioning link without changing the relevant hardware structure and accuracy, thereby reducing the waste of manpower and material resources and improving the production efficiency of embossed plates. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A flow chart of a sheet metal positioning method for reducing the reversal of industrial robot joints provided by the present invention. DETAILED DESCRIPTION

[0034] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0035] like Figure 1 As shown, the application field of the present invention is the door panel embossing production line, which can improve the absolute positioning accuracy of the industrial robot when placing the sheet material without manual operation when changing the deflection process. The present invention provides a sheet material positioning method for reducing the reverse rotation of the industrial robot joint, comprising the following steps:

[0036] Step 1: Establish the DH model of the industrial robot.

[0037] (1.1) According to the link coordinate system convention of the DH modeling process, establish the link coordinate system of each axis of the industrial robot;

[0038] (1.2) Use the four parameters of joint angle, rod length, torsion angle and horizontal distance to describe the motion of the industrial robot rod and create a DH parameter table for subsequent path analysis and planning;

[0039] (1.3) The homogeneous transformation matrix from the end operation point coordinate system to the robot base coordinate system is calculated from the DH parameter table to obtain the DH model of the industrial robot.

[0040] Step 2: When changing the offset process, adjust the end point of the end operation point in the workspace of the industrial robot according to the new offset distance, which is recorded as o x The remaining points can be used as before the deflection process is changed. The starting point is o1, and the i-th point among the intermediate points is o i , where i∈(1,x), there are x points in total.

[0041] Step 3: Use the selected points to analyze and plan the path in the joint space to obtain the current motion path, which also needs to be verified.

[0042] (3.1) The joint angle is θ i First, based on the DH parameter table, the initial joint angle information θ of each axis is obtained i1 , and then use inverse kinematics to solve θi2 Similarly, if there are multiple target points, a strategy set will be formed, similar to [θ i1 ,θ i2 ,θ i3 ,θ i4 ,θ i5 …];

[0043] (3.2) By fitting the curve of the joint angle change of each axis of the industrial robot, the relationship between the joint angle change of each axis in the path and time can be obtained, that is, the angle change function curve, which can be used as the mapping of the coordinate system. This step may result in non-unique solutions. You can select any one to verify, which will not affect the subsequent update and optimization of the path;

[0044] (3.3) Use forward kinematics to verify, convert the joint angle state of each axis into the state of the end operation point, and then plan the path of the end operation point of the industrial robot to verify whether it passes through the x points initially selected. Because the solutions in steps (3.1) and (3.2) are calculated based on the trajectory of the end operation point, if yes, it means that the calculation is correct, if no, it means that the calculation is wrong and needs to be recalculated and verified.

[0045] Step 4: Since the joint reversal during the material transportation process of the industrial robot will cause the reducer backlash, thereby generating absolute positioning error and affecting the absolute positioning accuracy of the sheet metal, the number of joint reversals of each axis is considered as the positioning evaluation index. Let N i =A i +B i +C i +D i +E i +F i , where A i ,B i ,C i ,D i ,E i ,F i Respectively represent the number of times the joint angles of the six axes of the industrial robot reverse, min{N i} represents the completed optimized solution.

[0046] Step 5: Using the Sarsa reinforcement learning algorithm, the new path planning is carried out with the optimization goal of reducing the number of joint reversals of each axis during the process of the industrial robot transporting the sheet material, that is, the minimum number of joint reversals required for each axis to reach the target point. The Sarsa algorithm is a reinforcement learning algorithm belonging to the value-based category. It calculates and selects the action corresponding to the maximum value function, updates the value function and obtains the optimization strategy.

[0047] (5.1) Sarsa algorithm uses Q table as a reference for selecting action strategy, so it is necessary to initialize the Q table, which has states as rows and actions as columns. The state is the position of the end operation point of the industrial robot, and the action is the rotation angle of the joint forward and reverse rotation of each axis. The Q value in the Q table represents the action value under a specific state;

[0048] (5.2) At a certain time t, the initial state s t ,s t is the starting position of the end operation point of the industrial robot, and then according to the ε-greedy strategy, from the current state s t Select an action a from all possible joint positive rotation ω and joint reverse rotation -ω actions t ,ω is the minimum rotation angle that can be achieved by each axis of the current industrial robot.

[0049] ε-Greedy Strategy: Where Q(s,a) is the value function; ε is a small value, generally set to 0.1, indicating the exploration rate. There is a probability of 1-ε to select the action with the greatest value from the forward and reverse rotation of the joints of each axis, and there is a probability of ε to act randomly; A is the action set of the forward and reverse rotation of the joints of each axis; m is the number of all optional actions for the forward and reverse rotation of the joints of each axis. Each axis of the six-axis industrial robot has two possible actions, so m is 12; represents the probability of selecting action a in the forward and reverse behaviors of each axis with a probability of ε; a * The action with the greatest value at present, that is, the joint of a certain axis rotates forward or reversely. It may be selected in two cases: one is selected when the action with the greatest value is selected with a probability of 1-ε, and the other is selected when it is randomly selected;

[0050] (5.3) Action a selected by the ε-greedy strategy t , a t is the forward rotation ω or reverse rotation -ω of a joint of a certain axis at the current moment. In the current state s t Next, perform action a t , get the state s t+1 ,s t+1 is the next position of the end operation point of the industrial robot, and also has an instantaneous feedback r t , r t It consists of two parts, t =R t -P t , R t is the reward for a single axis joint rotation, and P t Penalty for when the axis has joint reversal or when the axis rotation angle exceeds the limit;

[0051] (5.4) In the current state s t+1Select an action a from all possible joint forward rotation ω and joint reverse rotation -ω actions of each axis joint according to the ε-greedy strategy t+1 ;

[0052] (5.5) According to the new Q value Q(s t+1 ,a t+1 ), update the value function as well as the state and action, learn and find the strategy of forward and reverse rotation of each axis joint to obtain the maximum reward. If the current state is not the final state, repeat the three steps (5.3) to (5.5). The essence is to expect the maximum cumulative reward in the process of sheet material transportation and positioning. T is the total transportation time.

[0053] Update value function: Q(s,a)=Q(s t ,a t )+α[r+γQ(s t+1 ,a t+1 )-Q(s t ,a t )], where r = r t , γ is the attenuation factor, α is the iteration step, and α should gradually decrease as the iteration proceeds to ensure that the value function can converge.

[0054] Update state and action: s=s t+1 , a=a t+1 ;

[0055] (5.6) until s is the final state, which is the final position of the end operation point of the industrial robot. At this time, Q(s,a) converges and the iteration is completed. When Q(s,a) converges, the ε-greedy strategy also converges at the same time, and the optimal strategy is obtained. The updated motion path is obtained by analysis.

[0056] Step 6: Based on the updated path, the updated evaluation index is calculated in combination with the positioning evaluation index proposed in step 4, and compared with the initial evaluation index to obtain an optimization solution.

[0057] (6.1) Analyze the function curve of the joint angle of each axis changing with time under the initial conditions in step 3. When the slope of the curve is 0, it means that the joint is reversed. That is, the number of coordinate points with a tangent slope of 0 is the number of joint reversals of each axis. Substitute it into the positioning evaluation index formula set in step 4 to calculate the positioning evaluation index N1 of the initial path.

[0058] N1=A1+B1+C1+D1+E1+F1

[0059] Among them, A1, B1, C1, D1, E1, and F1 are the joint reversal times of each axis in the initial path;

[0060] (6.2) According to the changes in the movements of each axis in the Q table of the reinforcement learning process in step 5, the number of joint reversals of each axis during the material transportation process is statistically analyzed, and the positioning evaluation index formula set in step 4 is used to calculate the positioning evaluation index N2 of the updated path;

[0061] (6.3) Compare N1 and N2. If N1>N2, then min{N}=N2, indicating that sheet metal positioning optimization is completed; and if N1=N2, then min{N}=N2=N1, indicating that sheet metal positioning of the initial path under this method no longer needs to be optimized.

[0062] N2=A2+B2+C2+D2+E2+F2

[0063] Among them, A2, B2, C2, D2, E2, and F2 are the number of joint reversals of each axis in the updated path.

[0064] After the eccentric pattern process is changed, the traditional sheet metal positioning requires multiple manual intervention adjustments regardless of whether it is manual or industrial robot loading. The method of the present invention can reduce the absolute positioning error caused by the reducer backlash without changing the relevant hardware structure and accuracy, thereby improving the absolute positioning accuracy of the industrial robot and reducing the positioning error when the sheet metal is placed on the embossing table. This avoids subsequent multiple manual interventions while also improving the accuracy of positioning, promoting the smooth progress of the next step of embossing and subsequent production links, and improving production efficiency.

[0065] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A sheet metal positioning method for reducing the reversal of industrial robot joints, characterized in that: The following steps are involved: Step 1: Establish the DH model of the industrial robot; Step 2: When changing the offset process, adjust the end point of the end operation point in the workspace of the industrial robot according to the new offset distance, which is recorded as o x The rest of the points are the same as before the deflection process was changed. The starting point is o1, and the i-th point among the intermediate points is o i , where i∈(1,x), a total of x points; Step 3: Using the reinforcement learning algorithm, the optimization goal is to reduce the number of joint reversals of each axis during the process of the industrial robot transporting the sheet material, that is, the minimum number of joint reversals required for each axis to reach the target point. During the reinforcement learning process, the state is the position of the end operation point of the industrial robot, and the action is the rotation angle of the joint forward and reverse rotation of each axis. The reward of a single axis joint rotation minus the penalty when the axis joint reverses or when the axis rotation angle exceeds the limit is used as the instantaneous reward for reinforcement learning to obtain a planning path that meets the optimization goal; It includes the following sub-steps: (3.1) The reinforcement learning algorithm uses the Sarsa algorithm. The Sarsa algorithm uses the Q table as a reference for selecting action strategies. Therefore, the Q table needs to be initialized. The table has states as rows and actions as columns. The Q value in the Q table represents the value of an action under a specific state. (3.2) At a certain time t, the initial state s t ,s t is the starting position of the end operation point of the industrial robot, and then according to the ε-greedy strategy, from the current state s t Select an action a from all possible joint positive rotation ω and joint reverse rotation -ω actions t ,ω is the minimum rotation angle that can be achieved by each axis of the current industrial robot; (3.3) Action a selected by the ε-greedy strategy t , a t is the forward rotation ω or reverse rotation -ω of a joint of a certain axis at the current moment. In the current state s t Next, perform action a t , get the state s t+1 ,s t+1 is the next position of the end operation point of the industrial robot, and also has an instantaneous feedback r t , r t It consists of two parts, t =R t -P t , R t is the reward for a single axis joint rotation, and P t Penalty for when the axis has joint reversal or when the axis rotation angle exceeds the limit; (3.4) In the current state s t+1 Select an action a from all possible joint forward rotation ω and joint reverse rotation -ω actions of each axis joint according to the ε-greedy strategy t+1 ; (3.5) According to the new Q value Q(s t+1 ,a t+1 ), update the value function as well as the state and action, learn and find the strategy of forward and reverse rotation of each axis joint to obtain the maximum reward; if the current state is not the final state, repeat the three steps (3.3) to (3.5), and expect to obtain the maximum cumulative reward in the process of sheet material transportation and positioning T is the total material transportation time; (3.6) until s is the final state, which is the final position of the end operation point of the industrial robot. At this time, Q(s,a) converges and the iteration is completed. When Q(s,a) converges, the ε-greedy strategy also converges and the optimal strategy is obtained. The planned motion path is obtained by analysis.

2. A sheet metal positioning method for reducing the reversal of industrial robot joints according to claim 1, characterized in that: Step 1 includes the following sub-steps: (1.1) According to the link coordinate system convention of the DH modeling process, establish the link coordinate system of each axis of the industrial robot; (1.2) Use the four parameters of joint angle, rod length, torsion angle and horizontal distance to describe the motion of the industrial robot rod and create a DH parameter table for subsequent path analysis and planning; (1.3) The homogeneous transformation matrix from the end operation point coordinate system to the robot base coordinate system is calculated from the DH parameter table to obtain the DH model of the industrial robot.

3. A sheet metal positioning method for reducing the reversal of industrial robot joints according to claim 1, characterized in that: In step (3.2), the ε-greedy strategy is as follows: Where Q(s,a) is the value function; ε is 0.1, which indicates the exploration rate. There is a probability of 1-ε to select the action with the greatest value from the forward and reverse rotation of the joints of each axis, and there is a probability of ε to act randomly; A is the action set of the forward and reverse rotation of the joints of each axis; m is the number of all optional actions for the forward and reverse rotation of the joints of each axis; represents the probability of selecting action a in the forward and reverse behaviors of each axis with a probability of ε; a * It is the action with the greatest value at present, that is, the joint of a certain axis rotates forward or reversely. It may be selected in two cases. One is that it is selected when the action that is currently considered to have the greatest value is selected with a probability of 1-ε, and the other is that it is selected when it is randomly selected.

4. A sheet metal positioning method for reducing the reversal of industrial robot joints according to claim 1, characterized in that: In step (3.5), update the value function, state, and action as follows: Q(s,a)=Q(s t ,a t )+α[r+γQ(s t+1 ,a t+1 )-Q(s t ,a t )] s=s t+1 ,a=a t+1 where r = r t , γ is the attenuation factor, α is the iteration step, and α should gradually decrease as the iteration proceeds to ensure that the value function can converge.

5. A sheet metal positioning method for reducing the reversal of industrial robot joints according to claim 1, characterized in that: After obtaining the planned path that meets the optimization goal, the number of joint reversals of each axis is obtained through statistical analysis based on the changes in the movements of each axis in the reinforcement learning Q table. The positioning evaluation index of the planned path is obtained by summing up the number of joint reversals of each axis.

6. A sheet metal positioning method for reducing the reversal of industrial robot joints according to claim 1, characterized in that: In step 2, when changing the biasing process, if the biasing process is a biasing process for which a reinforcement learning planning path has been performed before, the previous corresponding path can be directly selected.

Images