Assembly butt joint attitude adjustment motion control method based on TS-FNN-PID model

Abstract

The invention relates to a TS-FNN-PID-based attitude adjustment motion control method for assembly and butt joint, which aims to improve the motion precision and response speed of an attitude adjustment mechanism in the assembly and butt joint of complex products, and is characterized by comprising the following steps of: (1) establishing a TS-FNN-PID-based attitude adjustment motion control model, and initializing model parameters; (2) generating a track data set of assembly butt joint attitude adjustment motion by using simulation software, and training the TS-FNN-PID model; and (3) evaluating the model by using a trajectory tracking error and response time, and performing reasonable parameter adjustment on the TS-FNN model to improve the performance of the model so as to realize assembly butt joint attitude adjustment motion control. According to the method, the PID control parameters are dynamically optimized based on the TS-FNN, so that the assembly, butt joint and attitude adjustment motion control of complex products is realized, and the motion precision and the response speed of an attitude adjustment mechanism are improved.

Description

Technical Field

[0001] The present invention relates to a complex product assembly docking posture adjustment motion control technology, in particular to a posture adjustment motion control method for improving the motion accuracy and response speed of a posture adjustment mechanism during assembly docking, specifically, an assembly docking posture adjustment motion control method based on a TS-FNN-PID model. Background Art

[0002] In the industrial and aerospace fields, attitude control systems are used to accurately adjust the attitude and position of equipment, especially in nonlinear systems with multiple inputs and multiple outputs. For example, in the assembly process of spacecraft, attitude adjustment mechanisms are used to accurately control the position and angle of components to ensure the smooth completion of the assembly process. The attitude adjustment mechanism usually adopts a parallel structure. The nonlinear characteristics of this structure and the existence of uncertain parameters in the system make it difficult for traditional control methods to meet the requirements of high precision and fast response.

[0003] PID control is a widely used classical control method that reduces system errors by adjusting proportional, integral and differential coefficients to stabilize the system near the expected value. However, traditional PID control cannot provide sufficient control accuracy and dynamic response in complex nonlinear systems, especially systems with multiple degrees of freedom and large uncertainties. To address the limitations of PID control, fuzzy control enhances the system's ability to cope with uncertainty and complex environments by fuzzifying the accuracy information of input variables. However, traditional fuzzy control relies on manually designed rules, lacks adaptive capabilities, and is difficult to cope with dynamic changes in the system.

[0004] Fuzzy neural network (FNN) combines the advantages of fuzzy control and neural network, and can optimize control parameters through self-learning, thereby improving the system's ability to handle complex nonlinear problems. However, in some high-dynamic and strong nonlinear control tasks, there are still certain limitations. In order to further improve the performance, the present invention proposes a control method based on TS-FNN, which combines fuzzy control and deterministic functions, so that the control system can more accurately approximate complex nonlinear relationships. This method optimizes the parameters of the PID controller in real time through the self-learning ability of the neural network, thereby improving the control accuracy and response speed. Summary of the invention

[0005] The purpose of the present invention is to address the problem of insufficient motion accuracy and response speed of the assembly docking attitude adjustment mechanism of complex products due to its nonlinear and strong coupling characteristics. A TS-FNN-PID-based assembly docking attitude adjustment motion control method is proposed. By combining TS-FNN with the traditional PID control algorithm, the parameters of the PID controller are optimized in real time, thereby improving the accuracy and response speed of the attitude adjustment motion.

[0006] The technical solution of the present invention is:

[0007] A TS-FNN-PID-based assembly docking posture adjustment motion control method is used to improve the motion accuracy and response speed of a posture adjustment mechanism in complex product assembly docking. The method is characterized in that it comprises the following steps: firstly, a TS-FNN-PID-based posture adjustment motion control model is established and model parameters are initialized; a trajectory data set of assembly docking posture adjustment motion is generated by simulation software and the TS-FNN-PID model is trained; finally, the model is evaluated by using trajectory tracking error and response time, and the TS-FNN model is reasonably adjusted to improve model performance and realize assembly docking posture adjustment motion control.

[0008] The specific steps include:

[0009] Step 1: Establish a posture adjustment motion control model based on TS-FNN-PID;

[0010] Step 2: Generate a trajectory data set of assembly docking posture adjustment motion using simulation software and preprocess the data set;

[0011] Step 3: Use the dataset to train and evaluate TS-FNN-PID, reasonably adjust the parameters of the TS-FNN model to improve model performance, and realize assembly docking posture adjustment motion control.

[0012] Further, in the step 1: establish a posture adjustment motion control model based on TS-FNN-PID; the model consists of a PID controller and a TS-type fuzzy neural network, and the TS-type fuzzy neural network dynamically adjusts the PID parameters through fuzzy reasoning and neural network learning to realize complex product assembly docking posture adjustment motion control. The input of the model is characteristic data such as trajectory error and error change rate, and the output is the optimized PID parameters. The Gaussian membership function is used in the fuzzy layer to initialize the center and width parameters of the membership function; the number of hidden layer nodes is set according to system requirements, and the activation function uses a positive and negative symmetric Sigmoid function; the number of output layer nodes is 3, corresponding to the three parameters of the PID controller, the initial weights and biases are set by random or preset values, and the learning rate and error tolerance of the model are determined to complete the initialization of the model parameters.

[0013] Furthermore, in step 2: the complex product assembly docking posture adjustment process is simulated by simulation software to generate the expected trajectory and simulation trajectory of each branch of the assembly posture adjustment mechanism. The expected trajectory is calculated by the path planning algorithm according to the assembly task requirements, representing the trajectory that the posture adjustment mechanism should execute under ideal conditions; and the simulation trajectory is the actual motion trajectory generated after the PID controller executes the task in the simulation environment. After obtaining the expected trajectory and the simulation trajectory, the trajectory error and error change rate between the two are calculated, and used as input data for training. In order to avoid oversaturation of the neural network activation function and improve the training efficiency, the trajectory error and error change rate are normalized to obtain a standardized data set.

[0014] Furthermore, in step 3: the data set generated in step 2 is divided into a training set and a test set. The TS-FNN-PID model is trained using the training set, with the input being the trajectory error and the error change rate, and the output being the optimization parameters of the PID controller. During the training process, the back propagation algorithm is used to adjust the network weights, minimize the error, and optimize the PID parameters.

[0015] After training, the model is evaluated using the test set, with the performance of the model mainly measured by two indicators: trajectory tracking error and response time. Based on the evaluation results, the model performance is improved by adjusting the membership function parameters, fuzzy rule weights and other key parameters of the TS-FNN model to achieve assembly docking posture adjustment motion control.

[0016] The beneficial effects of the present invention are:

[0017] 1. Aiming at the nonlinear characteristics of the posture adjustment mechanism in the process of complex product assembly and docking, the present invention proposes a TS-FNN-PID-based assembly and docking posture adjustment motion control method. This method optimizes the parameters of the PID controller through the adaptive adjustment and fuzzy reasoning mechanism of the TS-FNN model, thereby improving the motion accuracy of the posture adjustment mechanism.

[0018] 2. The present invention combines the adaptive learning ability of TS-FNN with the feedback mechanism of PID controller to improve the response speed of posture adjustment movement by dynamically adjusting the parameters of PID controller. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the present invention.

[0020] Figure 2 This is the principle diagram of the TS-FNN-PID model of the present invention.

[0021] Figure 3 This is a diagram showing the trajectory motion results of the branch chain of the posture adjustment mechanism of the present invention. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0023] like Figure 1-Figure 3 shown.

[0024] An assembly docking posture adjustment motion control method based on TS-FNN-PID is proposed. Firstly, a TS-FNN-PID-based posture adjustment motion control model is established and the model parameters are initialized. Then, a trajectory data set of the assembly docking posture adjustment motion is generated by simulation software, and the TS-FNN-PID model is trained. Finally, the trajectory tracking error and response time are used to evaluate the model, and the TS-FNN model is reasonably adjusted to improve the model performance and realize the assembly docking posture adjustment motion control. Figure 1 The present invention is a flow chart of posture adjustment motion control, which specifically includes the following steps:

[0025] Step 1: Establish a posture adjustment motion control model based on TS-FNN-PID;

[0026] The schematic diagram of TS-FNN-PID model is as follows: Figure 2 As shown, the control of each branch of the posture adjustment mechanism is realized by a PID controller, and the fuzzy neural network is responsible for online adjustment of its parameters. r(k) represents the expected output, which is the expected displacement Δl of the posture adjustment mechanism branch i calculated in the virtual space in the present invention. virtual , y(k) is used to represent the simulation output, which represents the simulation branch position Δl real , then the controlled deviation signal can be expressed as:

[0027] e(k)=r(k)-y(k) (1)

[0028] The control law is:

[0029] u(k)=k P e(k)+k I ∑ i＝0 e(i)+k D [e(k)-e(k-1)] (2)

[0030] Δu(k)=k P [e(k)-e(k-1)]+k I e(k)+k D [e(k)-2e(k-1)+e(k-2)] (3)

[0031] In equations (1), (2) and (3), u(k) is the output of the controller at the kth time point, k P ,k I ,k D is the adjustable parameter of the controller.

[0032] The fuzzy rules of a fuzzy neural network are a set of rules that describe the relationship between input and output. These rules are used in the reasoning process in the fuzzy system to map fuzzy input to fuzzy output. The domain in a fuzzy neural network refers to the range of values ​​of a fuzzy variable. Each input or output variable has its domain, which defines all possible values ​​that this variable can take. In the present invention, the basic domain of the branch displacement deviation e(k) is defined, and the quantized fuzzy domain is e={-3,-2,-1,0,+1,+2,+3}. The input variables of the system are fuzzy processed, and the fuzzy subset is set to: e={NL,NM,NS,Z,PS,PM,PL}, that is, {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}. The system error is normalized and mapped to [0,1] for neural network normalization input, and the normalization method is e(k) / r(k) to avoid oversaturation of the activation function.

[0033] The number of nodes in the input layer is: x1(k) = e(k), x2(k) = e(k)-e(k-1). Among them, x1(k) is the error amount, and x2(k) is the error change rate. According to the above, for the motion control of the t-th branch of the parallel posture adjustment mechanism, the mechanism displacement error e t and displacement error change Δe t As the input of TS fuzzy neural network, the input vector x of the fuzzy neural network input layer is t It can be expressed by formula (4):

[0034] x t =[x y1 ,x t2 ]=[e t ,Δe t ] T ,(t=1,2,…6) (4)

[0035] The membership function of the fuzzy layer is a Gaussian function. The Gaussian membership function is similar to a bell curve and is controlled by the peak and standard deviation parameters, namely the membership function center c and the membership function width b. Since in the movement of the assembly docking posture adjustment mechanism, e t and Δe t The change range is small, the membership function center c∈[-1,1], and the membership function width b∈[1,2].

[0036] The reasoning layer uses the multiplication operator to calculate the credibility of each rule, sums the credibility of H rule nodes, and then normalizes the hth rule node by dividing the credibility product of each rule node.

[0037] According to the fuzzy rules in step 3, the number of hidden layer nodes is set to 6, and the positive and negative symmetric Sigmoid function is used as the activation function of the network. The expression of the Sigmoid function is:

[0038]

[0039] In the fuzzy neural network posture adjustment motion PID control of the present invention, the network output is the three parameters k of the PID controller P ,k I ,k D Therefore, the number of output nodes is set to 3. Since the parameter of the PID controller is a positive number, the activation function of the output layer is non-negative, and its expression is:

[0040]

[0041] The network uses the gradient descent method to update weights. By calculating the gradient of the indicator function E(k) with respect to the weights, the output layer and hidden layer weight coefficients in the network are updated along the opposite direction of the gradient. The expression of E(k) in this paper is shown in formula (7):

[0042]

[0043] The main steps of the PID control algorithm for posture adjustment based on TS-type fuzzy neural network are as follows: (1) determining the structure of the back propagation neural network, including the number of nodes i, j, k in the input layer, hidden layer and output layer; (2) initializing weight parameters, including the hidden layer weight and the output layer weight: w hi and w kj , and set the learning rate η at the same time, and set the iteration variable to k=1; (3) By sampling the system, the initial error amount e(k)=r(k)-y(k) is obtained; (4) The error state variable e(k) is fuzzified, and the fitness value of each rule is calculated, and then the fitness value is normalized, and the fuzzification result is used as the input of the neural network; (5) The input and output of each layer of the TS-type fuzzy neural network are calculated; (6) The output u(k) of the controller is calculated using the control law of the incremental PID; (7) The self-learning of the neural network is performed through the gradient descent method of the back propagation network, and the network weight coefficient w is dynamically adjusted. ji and w kj ; (8) Iteratively update k=k+1, return to (3), and continue PID control of the posture adjustment movement.

[0044] Step 2: Generate a trajectory data set of assembly docking posture adjustment motion using simulation software and preprocess the data set;

[0045] The simulation software is used to simulate the complex product assembly docking posture adjustment process, and the expected trajectory and simulation trajectory of each branch of the assembly posture adjustment mechanism are generated. Among them, the expected trajectory is calculated by the path planning algorithm according to the assembly task requirements, representing the trajectory that the posture adjustment mechanism should execute under ideal conditions; and the simulation trajectory is the actual motion trajectory generated after the PID controller executes the task in the simulation environment. After obtaining the expected trajectory and the simulation trajectory, the trajectory error and error change rate between the two are calculated and used as input data for training. In order to avoid oversaturation of the neural network activation function and improve training efficiency, the trajectory error and error change rate are normalized to obtain a standardized data set.

[0046] Step 3: Use the data set to train and evaluate TS-FNN-PID, reasonably adjust the parameters of the TS-FNN model to improve model performance, and realize assembly docking posture adjustment motion control;

[0047] The dataset generated in step 2 is divided into a training set and a test set. The TS-FNN-PID model is trained using the training set, with the trajectory error and error change rate as input and the optimized parameters of the PID controller as output. During the training process, the back propagation algorithm is used to adjust the network weights, minimize the error and optimize the PID parameters.

[0048] After training, the model is evaluated using the test set, mainly using two indicators, trajectory tracking error and response time, to measure the performance of the model. Based on the evaluation results, the model performance can be improved by adjusting the membership function parameters, fuzzy rule weights and other key parameters of the TS-FNN model. Figure 3 The results show that for the expected trajectory of the branch chain motion, the assembly docking posture adjustment motion control based on TS-FNN-PID can obtain a relatively excellent branch chain motion trajectory. The assembly docking posture adjustment motion control using TS-FNN-PID can basically achieve the same tracking trajectory as the expected input trajectory of the branch chain for the docking branch chain motion, and compared with the traditional PID controller and the fuzzy PID controller, it can be seen that its trajectory tracking error is smaller.

[0049] The present invention does not involve parts that are the same as the prior art or can be implemented by using the prior art.

Claims

1. A TS-FNN-PID-based assembly docking posture adjustment motion control method, characterized in that: It includes the following steps: first, establish a posture adjustment motion control model based on TS-FNN-PID and initialize the model parameters; use simulation software to generate a trajectory data set of assembly docking posture adjustment motion, and train the TS-FNN-PID model; finally, use trajectory tracking error and response time to evaluate the model, and reasonably adjust the parameters of the TS-FNN model to improve the model performance, realize assembly docking posture adjustment motion control, and thus improve the motion accuracy and response speed of the posture adjustment mechanism in complex product assembly docking.

2. The method according to claim 1, characterized in that The steps include: Step 1: Establish a posture adjustment motion control model based on TS-FNN-PID; Step 2: Generate a trajectory data set of assembly docking posture adjustment motion using simulation software and preprocess the data set; Step 3: Use the dataset to train and evaluate TS-FNN-PID, reasonably adjust the parameters of the TS-FNN model to improve model performance, and realize assembly docking posture adjustment motion control.

3. The method according to claim 2, characterized in that In step 1: Establish a posture adjustment motion control model based on TS-FNN-PID; the model consists of a PID controller and a TS-type fuzzy neural network. The TS-type fuzzy neural network dynamically adjusts the PID parameters through fuzzy reasoning and neural network learning to realize the posture adjustment motion control of complex product assembly docking; the input of the model is the trajectory error and error change rate characteristic data, and the output is the optimized PID parameters; the Gaussian membership function is used in the fuzzy layer to initialize the center and width parameters of the membership function; the number of hidden layer nodes is set according to system requirements, and the activation function uses a positive and negative symmetric Sigmoid function; the number of output layer nodes is 3, corresponding to the three parameters of the PID controller, the initial weights and biases are set by random or preset values, and the learning rate and error tolerance of the model are determined to complete the initialization of the model parameters; the control of each branch of the posture adjustment mechanism is realized by the PID controller, and the fuzzy neural network is responsible for online tuning of its parameters; r(k) is used to represent the expected output, and the expected displacement Δl of the posture adjustment mechanism branch i calculated in the virtual space is virtual , y(k) is used to represent the simulation output, which represents the simulation branch position Δl real , then the controlled deviation signal can be expressed as: e(k)=r(k)-y(k) (1) The control law is: u(k)=k P e(k)+k I ∑ i＝0 e(i)+k D [e(k)-e(k-1)] (2) Δu(k)=k P [e(k)-e(k-1)]+k I e(k)+k D [e(k)-2e(k-1)+e(k-2)] (3) In equations (1), (2) and (3), u(k) is the output of the controller at the kth time point, k P ,k I ,k D is the adjustable parameter of the controller.

4. The method according to claim 2, characterized in that: In step 2: The complex product assembly docking and posture adjustment process is simulated by simulation software to generate the expected trajectory and simulation trajectory of each branch of the assembly posture adjustment mechanism. The expected trajectory is calculated by the path planning algorithm according to the assembly task requirements, and represents the trajectory that the posture adjustment mechanism should execute under ideal conditions. The simulation trajectory is the actual motion trajectory generated after the PID controller performs the task in the simulation environment. After obtaining the expected trajectory and the simulation trajectory, the trajectory error and error change rate between the two are calculated and used as input data for training. In order to avoid oversaturation of the neural network activation function and improve training efficiency, the trajectory error and error change rate are normalized to obtain a standardized data set.

5. The method according to claim 2, characterized in that: In step 3: Divide the data set generated in step 2 into a training set and a test set; use the training set to train the TS-FNN-PID model, with the trajectory error and error change rate as input, and the PID controller optimization parameters as output; during the training process, use the back propagation algorithm to adjust the network weights, minimize the error and optimize the PID parameters; after training, use the test set to evaluate the model, mainly using trajectory tracking error and response time as the two indicators to measure the performance of the model; according to the evaluation results, adjust the membership function parameters and fuzzy rule weight key parameters of the TS-FNN model to improve the model performance and realize assembly docking posture adjustment motion control; the fuzzy rules of the fuzzy neural network are a set of rules that describe the relationship between input and output; these rules are used in the fuzzy system In the reasoning process, the fuzzy input is mapped to the fuzzy output; the domain in the fuzzy neural network refers to the value range of the fuzzy variable; each input or output variable has its domain, which defines all possible values ​​that this variable can take, so the basic domain of the branch displacement deviation e(k) is defined, and the quantized fuzzy domain is e={-3,-2,-1,0,+1,+2,+3}. The input variables of the system are fuzzy processed, and the fuzzy subset is set to: e={NL,NM,NS,Z,PS,PM,PL}, that is, {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}, and the system error is normalized and mapped to [0,1] for neural network normalization input. The normalization method is e(k) / r(k) to avoid oversaturation of the activation function. The number of nodes in the input layer is: x1(k) = e(k), x2(k) = e(k)-e(k-1). Among them, x1(k) is the error amount, x2(k) is the error change rate, and for the motion control of the t-th branch of the parallel posture adjustment mechanism, the mechanism displacement error amount e is t and displacement error change Δe t As the input of TS fuzzy neural network, the input vector x of the fuzzy neural network input layer is t It can be expressed by formula (4): x t =[x t1 ,x t2 ]=[e t ,Δe t ] T ,(t=1,2,…6) (4) The membership function of the fuzzy layer is a Gaussian function. The Gaussian membership function is similar to a bell curve and is controlled by the peak and standard deviation parameters, namely the membership function center c and the membership function width b. Since in the movement of the assembly docking posture adjustment mechanism, e t and Δe t The change range is small, the membership function center c∈[-1,1], and the membership function width b∈[1,2]; The reasoning layer uses the multiplication operator to calculate the credibility of each rule, and sums the credibility of H rule nodes, and then divides the credibility product of each rule node to normalize the hth rule node; according to the fuzzy rule, the hidden layer nodes are set to 6, and the positive and negative symmetric Sigmoid function is used as the activation function of the network. The expression of the Sigmoid function is: In the PID control of fuzzy neural network posture adjustment motion, the network output is the three parameters k of the PID controller P ,k I ,k D Therefore, the number of output nodes is set to 3. Since the parameter of the PID controller is a positive number, the activation function of the output layer is non-negative, and its expression is: The network uses the gradient descent method to update weights. By calculating the gradient of the indicator function E(k) with respect to the weights, the output layer and hidden layer weight coefficients in the network are updated along the opposite direction of the gradient. The expression of E(k) in this paper is shown in formula (7):

6. The method according to claim 2, characterized in that The main steps of the PID control algorithm for attitude adjustment based on TS-type fuzzy neural network are: (1) determine the structure of the back propagation neural network, including the number of nodes i, j, k in the input layer, hidden layer and output layer; (2) initialize the weight parameters, including the hidden layer weight and the output layer weight: w ji and w kj , and set the learning rate η at the same time, and set the iteration variable to k=1; (3) By sampling the system, the initial error amount e(k)=r(k)-y(k) is obtained; (4) The error state variable e(k) is fuzzified, and the fitness value of each rule is calculated, and then the fitness value is normalized, and the fuzzification result is used as the input of the neural network; (5) The input and output of each layer of the TS-type fuzzy neural network are calculated; (6) The output u(k) of the controller is calculated using the control law of the incremental PID; (7) The self-learning of the neural network is performed through the gradient descent method of the back propagation network, and the network weight coefficient w is dynamically adjusted. ji and w kj ; (8) Iteratively update k=k+1, return to (3), and continue PID control of the posture adjustment movement.

Images