A joint following compliant control method for an electronic component assembly robot

Abstract

The present invention discloses a joint following compliant control method for an electronic component assembly robot. The joint following control method comprises: establishing a multi-joint following model of an assembly robot mechanical arm; improving the Elman neural network, placing the receiving layer between the hidden layer and the output layer, simplifying the connection weight calculation process, and improving the calculation ability; designing an Elman neural network PID controller, and proposing a forward algorithm for the working signal and a reverse algorithm for the error signal. Compared with the existing PID controller, the present invention establishes a mechanical arm following model for a three-joint assembly robot, designs a K-based control algorithm based on an improved Elman neural network, and improves the accuracy of the control. p , K i , K d A three-parameter PID controller is proposed, and a control algorithm is given, which effectively makes up for the problems of insufficient error convergence accuracy and insufficient stability after convergence of the PID controller. The angular displacement error, angular velocity error, and angular acceleration error of the joint are reduced to the minimum, achieving stable tracking of the three joints.

Description

Technical Field

[0001] The invention relates to a robot joint following compliant control method, and more particularly to a robot joint following compliant control method for assembling an electronic component. Background Art

[0002] The electronics industry plays an indispensable role in the development of the national economy. The emergence of electronic component assembly robots has promoted the progress of the electronics industry. When enterprises use electronic component assembly robots for production, they have the following advantages: (1) improve assembly efficiency; (2) reduce assembly costs; (3) improve the working environment of workers.

[0003] The robot's ability to comply with the contact environment is called compliance. The robot's compliant control is the premise for ensuring the safe and efficient completion of physical human-machine interaction tasks. By designing a reasonable compliant control strategy for collaborative robots, it will help to better implement physical human-machine interaction technology in many fields. In the process of interaction between robots, humans and the environment, compliant motion can ensure the robot's stable contact and operational safety, and can also effectively enhance the robot's motion performance. Therefore, if the assembly robot's arm joint following control system does not have a stable output, it will cause unpredictable economic losses.

[0004] At present, the improvement of the joint following control technology of the robotic arm is mainly focused on establishing models of different joint following objects. The patent application with application number 202011472821.7 discloses a surgical robot control device and control method to control the real joints to change with the image model; the patent application with application number 201710636650.9 discloses a fuzzy adaptive control method for a five-degree-of-freedom hybrid robot, which adopts a "feedback correction + feedforward correction" composite control strategy, and takes the errors and change rates of each driving joint of the hybrid robot as input, and constructs feedback and feedforward fuzzy algorithms respectively, and realizes the online hierarchical fuzzy tuning of the feedback and feedforward controller parameters through threshold judgment; the patent application with application number 202110073439.7 discloses a master-slave tracking control method for a minimally invasive surgical robot, which combines a simplified kinematic modeling method of heterogeneous mapping mode with a homogeneous mapping mode. The master-slave manipulator positioning part adopts a heterogeneous mapping mode, and the master-slave manipulator posture part adopts a homogeneous mapping mode.

[0005] At present, the existing PID controller used by electronic component assembly robots has a relatively simple working environment and improves assembly efficiency to a certain extent when only repetitive labor is performed. However, once faced with an emergency or special task, the joint end will have a large error in following the trajectory. In addition, when the electronic component assembly robot moves and drives in a complex environment, it is disturbed by the environment and needs to perform a large amount of real-time calculations to reduce the deviation of the end trajectory of the robot arm, making the flexibility of the robot joint control system extremely challenging. Summary of the invention

[0006] In order to solve the problem of large deviation in the terminal following trajectory of the assembly robot in a complex environment in the prior art, the invention proposes a joint following compliant control method for the electronic component assembly robot based on the characteristics of automatic following of the assembly robot's mechanical arm joints. This method has high anti-interference performance and significantly improves the compliance of the mechanical arm joint control system.

[0007] The technical solution of this patent is as follows: A joint following compliance control method of an electronic component assembly robot comprises the following steps:

[0008] (1) Firstly, a multi-joint following model of an assembly robot is proposed, and the improved angular displacement error expression and the state space expression of the joint following model are obtained;

[0009] (2 Improve the existing Elman neural network, change the position of its receiving layer, simplify the calculation process of its connection weights, and design a 4-5-3 three-layer Elman neural network structure;

[0010] (3) Design the Elman neural network PID controller structure, use the strong adaptive ability of the neural network to achieve the best combination of PID control parameters, reduce the error between the expected output value and the actual output value, and improve the smooth control performance of the PID controller; propose a new forward algorithm for the working signal and a reverse algorithm for the error signal;

[0011] (4) By comparing the joint angular displacement errors under the existing PID controller and the Elman neural network PID controller, the joint following compliant control method proposed in this patent is verified.

[0012] Among them, for the established assembly robot manipulator joint following model, the angular displacement error expression and the joint following model state space expression are obtained according to the following steps:

[0013] The relative following error of the assembly robot angular displacement can be defined as

[0014] δ d1 =Δθ i -d s

[0015] In order to further improve the accuracy of the model and reduce the influence of the angular velocity of the joint on the angular displacement error, the above formula is optimized by angular acceleration, and the improved angular displacement error expression is as follows:

[0016] δ d =θ i+1 -θ i -d s +λω i+1 =δ d1 +λω i+1

[0017] The relative angular velocity error expression of the i+1th joint is as follows:

[0018] δ ω =ω i+1 -ω i

[0019] Taking the above three formulas as the core formulas, a joint following model based on a three-joint electronic component assembly robot is established. The state space expression is as follows:

[0020]

[0021] The proposed improved Elman neural network is different from the standard Elman neural network in that the receiving layer is moved from between the input layer and the hidden layer to between the hidden layer and the output layer, and there is no need to calculate the connection weights from the receiving layer to the intermediate layer. The output weights of the hidden layer are used to memorize the output of the previous hidden layer in the receiving layer, and the connection weights of the receiving layer are the same as the weight coefficients of the hidden layer. Therefore, the Elman neural network PID controller structure designed in this patent is divided into three layers: input layer, hidden layer (the receiving layer is included in the hidden layer and counted as one layer) and output layer. The specific neural network structure is as follows: Figure 4 The input layer contains 4 input signals, which are input rin, output yout, deviation error and 1; the 3 output signals of the output layer are the three parameters K of the PID controller. p , K i , K d , overall it is a 4-5-3 neural network structure.

[0022] For the forward algorithm of the working signal, in order to better embed the receiving layer between the hidden layer and the output layer, the input layer and the hidden layer of the 3-layer Elman neural network are analyzed, and the output of the hidden layer is obtained as follows:

[0023]

[0024] Where Kl is the hidden layer output weight, Kr is the hidden layer output weight of the previous step, and Kl = 0.2, Kr = 0.8. is the weight coefficient of the hidden layer; Is the hidden layer input; is the hidden layer neuron activation function, is the previous output of the hidden layer.

[0025] The input and output expressions of the output layer are as follows:

[0026]

[0027] In the formula is the weight coefficient of the output layer; is the hidden layer output; is the input of the output layer; O (3) (k) is the output layer output; is the activation function of the output layer neurons.

[0028] For the reverse algorithm of the error signal, the weight update of the Elman neural network is realized by the gradient descent method. By adding the inertia term, the algorithm's ability to quickly find the optimal solution and reach the global minimum is improved.

[0029] According to the relationship between the system performance index functions and the change in the output layer weight coefficient, it is deduced that:

[0030]

[0031] Where E() is the system performance index function; y() is the output of the system performance index function; u() is the control function.

[0032] Derived from the output layer input and output expressions and the incremental PID control expression:

[0033]

[0034] Where e() is the error function.

[0035] In summary, the algorithm expression of the output layer weight coefficient is:

[0036]

[0037] In the formula, is the local gradient of the output layer; f′(x)=f(x)[1-f(x)]; sgn() is the step function, η is the learning rate; α is the inertia coefficient.

[0038] Similarly, the algorithm expression of the hidden layer weight coefficient is:

[0039]

[0040] In the formula, is the local gradient of the hidden layer; X(k) is the output of the receiving layer, which is the input of the hidden layer; is the hidden layer weight coefficient; g′(x)=[1-g 2 (k)] / 2.

[0041] Beneficial effects:

[0042] (1) The present invention provides a joint following compliant control method for an electronic component assembly robot. The joint following model of the assembly robot mechanical arm is constructed for the Elman neural network PID controller to improve the joint following performance. The Elman neural network is improved by placing the receiving layer between the hidden layer and the output layer to simplify the calculation process. A working signal forward algorithm and an error signal reverse algorithm are proposed to control the three parameters K of the PID controller. p , K i , K d Rapid optimization, using the strong adaptive ability of neural networks, to achieve the best combination of PID control parameters, reduce the error between the expected output value and the actual output value, and improve the smooth control performance of the PID controller.

[0043] (2) The present invention provides a joint following compliant control method for an electronic component assembly robot. In order to further improve the accuracy of the joint following model of the assembly robot's mechanical arm and reduce the influence of the joint's angular velocity on the angular displacement error, the angular acceleration is optimized, and the existing PID controller is effectively compensated for the problems of insufficient error convergence accuracy and insufficient stability after convergence. The angular displacement error, angular velocity error, and angular acceleration error of the joint are reduced to the minimum, thereby achieving stable following of the three joints. At the same time, in order to realize the memory function and enable the PID control system to have the ability to adapt to real-time changes and further enhance the global stability of the neural network, the hidden layer output of the previous layer is added to the hidden layer output.

[0044] (3) Compared with the PID controller used in the existing assembly robot arm, the present invention can reduce the deviation of the end track of the electronic component assembly robot when facing a complex environment, and significantly improve the flexibility of the robot arm joint control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention provides a flow chart of a joint following compliance control method for an electronic component assembly robot.

[0046] Figure 2 This is a diagram of the joint following model of the assembly robot arm.

[0047] Figure 3 It is the structure diagram of the three-layer Elman neural network 4-5-3 in the embodiment.

[0048] Figure 4 1 is a structural diagram of the Elman neural network PID controller in the embodiment.

[0049] Figure 5 : is the joint angular displacement error diagram under the Elman neural network PID controller in the embodiment.

[0050] Figure 6 This is a diagram of the joint angular displacement error of the present embodiment under the existing PID controller. DETAILED DESCRIPTION

[0051] In order to deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with embodiments and drawings. The embodiments are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0052] like Figure 1 As shown, the method of the present invention comprises the following steps:

[0053] (1) Establishing the joint following model of the assembly robot, taking the relative angular displacement error expression and the relative angular velocity error expression as the core formula, establishing the state space expression of the joint following model of the electronic component assembly robot;

[0054] The relative following error of the assembly robot angular displacement can be defined as

[0055] δ d1 =Δθ i -d s

[0056] In order to further improve the accuracy of the model and reduce the influence of the angular velocity of the joint on the angular displacement error, the above formula is optimized by angular acceleration, and the improved angular displacement error expression is as follows:

[0057] δ d =θ i+1 -θ i -d s +λω i+1 =δ d1 +λω i+1

[0058] The relative angular velocity error expression of the i+1th joint is as follows:

[0059] δ ω =ω i+1 -ω i

[0060] Taking the above three formulas as the core formulas, a joint following model based on a three-joint electronic component assembly robot is established. The state space expression is as follows:

[0061]

[0062] In the formula, B=(λ,1) T , Γ=(0,-1) T , C=(0,1), are the system matrix, control matrix, interference matrix and output matrix respectively; X=(δ d ,δ ω ) T is the state variable of the joint following control system of the electronic component assembly robot; u is the control variable, that is, the angular acceleration control change of the i+1th joint (following joint); ψ is the interference variable, that is, the angular acceleration control change of the i-th joint (main joint); Y is the angular displacement error of the relative joint.

[0063] (2) Convert the state space expression of the joint following model into a transfer function form: G(s) = (0.1s + 1) / s 2 During the simulation, the angular velocity of the main joint is set to 5° / s = π / 36 rad / s, and the goal is to control the angular displacement error of the follower joint within 0.05 rad. Assume that the learning rate η = 0.6, the inertia coefficient α = 0.02, and the initial weight of the hidden layer ω ij , the initial weight of the output layer ω jl Take a random value on [0,1] and set the initial value of the PID controller parameter to K p =0.002; K i =2e-5; K d =0.1, simulation sampling time t s =0.002s.

[0064] (3) Figure 3 As shown in the figure, the Elman neural network is improved and a 4-5-3 three-layer Elman neural network structure is established: input layer, hidden layer (adjust the position of the receiving layer and place it between the hidden layer and the output layer. The connection weight of the receiving layer is consistent with the connection weight of the hidden layer and is counted as one layer) and output layer. The input layer contains 4 input signals, which are the input rin, the output yout, the deviation error and 1; the 3 output signals of the output layer are the three parameters K of the PID controller. p , K i , K d .

[0065] (4) Figure 4 As shown, an Elman neural network PID controller is designed, and a new working signal forward algorithm and error signal reverse algorithm are proposed;

[0066] The output of the hidden layer is:

[0067]

[0068] Where Kl is the hidden layer output weight, Kr is the hidden layer output weight of the previous step, and Kl = 0.2, Kr = 0.8. is the weight coefficient of the hidden layer; Is the hidden layer input; is the hidden layer neuron activation function, is the previous output of the hidden layer.

[0069] The input and output expressions of the output layer are as follows:

[0070]

[0071] In the formula is the weight coefficient of the output layer; is the hidden layer output; is the input of the output layer; O (3) (k) is the output layer output; is the activation function of the output layer neurons.

[0072] For the reverse algorithm of the error signal, the weight update of the Elman neural network is realized by the gradient descent method. By adding the inertia term, the algorithm's ability to quickly find the optimal solution and reach the global minimum is improved.

[0073] According to the relationship between the system performance index functions and the change in the output layer weight coefficient, it is deduced that:

[0074]

[0075] Where E() is the system performance index function; y() is the output of the system performance index function; u() is the control function.

[0076] Derived from the output layer input and output expressions and the incremental PID control expression:

[0077]

[0078] Where e() is the error function.

[0079] In summary, the algorithm expression of the output layer weight coefficient is:

[0080]

[0081] In the formula, is the local gradient of the output layer; f′(x)=f(x)[1-f(x)]; sgn() is the step function, η is the learning rate; α is the inertia coefficient.

[0082] Similarly, the algorithm expression of the hidden layer weight coefficient is:

[0083]

[0084] In the formula, is the local gradient of the hidden layer; X(k) is the output of the receiving layer, which is the input of the hidden layer; is the hidden layer weight coefficient; g′(x)=[1-g 2 (k)] / 2.

[0085] (5) The m file of the PID control algorithm based on the Elman neural network written in m language is imported into the module of the S function, and the simulation time is set to 60s;

[0086] (6) After the Elman neural network adjusts the weight coefficients of each layer, the output K of the output layer of the Elman neural network is obtained. p , K i , K d The change curve of can be used to obtain the relative angular displacement error result under the Elman neural network PID controller.

[0087] (7) Figure 5 and Figure 6 As shown in the figure, under the Elman neural network PID controller system, the angular displacement deviation is stable at about 0.044 rad, while under the existing PID controller following system, the joint angular displacement deviation is large.

[0088] As a further improvement, the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A joint following compliance control method for an electronic component assembly robot, characterized in that: The steps include: S1: Establish a multi-joint following model of the assembly robot manipulator, optimize the angular acceleration and joint following model, and obtain the improved angular displacement error expression and joint following model state space expression; S2: Improve the Elman neural network, place the receiving layer between the hidden layer and the output layer, simplify the connection weight calculation process, and design a 4-5-3 three-layer Elman neural network structure; S3: Design an Elman neural network PID controller and propose a new forward algorithm for the working signal and a reverse algorithm for the error signal; S4: Compare the joint angle displacement errors under the existing PID controller and the Elman neural network PID controller to verify the method; The input and output expressions of the hidden layer are as follows: Where Kl is the hidden layer output weight; Kr is the hidden layer output weight of the previous step; is the weight coefficient of the hidden layer; Is the hidden layer input; is the hidden layer neuron activation function, is the previous output of the hidden layer; Add the hidden layer weight coefficient algorithm expression to the receiving layer term, which is: In the above formula, is the local gradient of the hidden layer; X(k) is the output of the receiving layer, which is the input of the hidden layer; is the hidden layer weight coefficient; g′(x)=[1-g 2 (k)] / 2; For the reverse algorithm of the error signal, the weight update of the Elman neural network adopts the gradient descent method. According to the relationship between the system performance index function and the change of the output layer weight coefficient, it is derived that: Where, E() is the system performance index function; y() is the output of the system performance index function; u() is the control function; Derived from the output layer input and output expressions and the incremental PID control expression: Where e() is the error function; In summary, the algorithm expression of the output layer weight coefficient is: In the formula, is the local gradient of the output layer; f′(x)=f(x)[1-f(x)]; sgn() is the step function, η is the learning rate; α is the inertia coefficient.

2. The joint following compliance control method of an electronic component assembly robot according to claim 1, characterized in that: The angular acceleration is optimized in S1, and the improved angular displacement error expression is as follows: d d =θ i+1 -θ i -d s +lo i+1 =d d1 +lo i+1 In the above formula, δ d is the angular displacement error of the i+1th joint relative to the ith joint; ω i+1 is the angular velocity of the i+1th joint; λ is the compensation time required for the angular displacement error to converge during joint following.

3. The joint following compliance control method of an electronic component assembly robot according to claim 2, characterized in that: The state space expression of the improved joint following model in S1 is: In the formula, B=(λ,1) T , Γ=(0,-1) T , C=(0,1), which are the system matrix, control matrix, interference matrix and output matrix respectively; X=(δ d ,δ ω ) T State variables of the joint following control system for the electronic component assembly robot; u is the control variable, i.e., the angular acceleration control change of the i+1th joint, i.e., the follower joint; ψ is the disturbance variable, i.e., the controlled change of the angular acceleration of the i-th joint, i.e., the main joint; Y is the angular displacement error of the relative joint.

4. The joint following compliance control method of an electronic component assembly robot according to claim 1, characterized in that: The Elman neural network PID controller structure in S3 includes an input layer, a hidden layer, and an output layer. The input layer includes four input signals, namely, the input rin, the output yout, the deviation error, and 1; the three output signals of the output layer are the three parameters K of the PID controller. p , K i , K d .

5. The joint following compliance control method of an electronic component assembly robot according to claim 1, characterized in that: The input and output expressions of the output layer are as follows: In the formula is the weight coefficient of the output layer; is the hidden layer output; is the input of the output layer; O (3) (k) is the output layer output; is the activation function of the output layer neurons.

Images