Yarn winding back pressure control method based on radial basis function neural network

Abstract

The invention discloses a yarn winding back pressure control method based on a radial basis function neural network. The method comprises the following steps: 1, constructing a combined control model; 2, acquiring a control error; 3, outputting a control signal by the PID controller; and 4, inputting the input features into the radial basis function neural network. The radial basis function neural network outputs identification information and a Jacobian matrix; and 5, updating parameters of the PID controller. And 6, updating the network parameters of the radial basis nerve through a particle swarm algorithm. The back pressure is controlled by controlling the torque of the motor instead of a traditional mechanical device such as a spring, so that the back pressure is kept consistent, and the winding quality of yarns is improved. The PID controller is optimized through the radial basis function neural network; and the center, the width and the weight of the radial basis function neural network are corrected through a particle swarm optimization algorithm, so that the control requirement of nonlinear back pressure control is met, and a higher response speed and a better control effect are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of doubling machine production, and in particular to a yarn winding back pressure control method based on a radial basis function neural network. Background Art

[0002] The inventor of the present invention found that when the doubling machine is working, as the winding amount of the yarn increases, the volume and mass of the bobbin will also increase, resulting in the bobbin pressing on the winding roller to generate back pressure. In addition, during the winding process of the bobbin, the volume and mass of the bobbin change unevenly over time, causing the back pressure between the bobbin and the winding roller to change unevenly.

[0003] The back pressure control of traditional winding machines relies on counterweights, spring tension or compression adjustment control. The adjustment process is complicated and requires high professional skills. It is difficult for general operators without training to accurately adjust and control the back pressure, which greatly affects the factory's production efficiency. In addition, due to factors such as the type of yarn, the shape of the yarn cone, the winding diameter, and the hardness of the yarn winding, the stability of the cradle is poor. During the winding process, it is easy for the yarn winding to be not round, resulting in instability of the cradle, vibration, and noise, which affects the formation and packaging quality of the cone yarn. Summary of the invention

[0004] In view of the defects of the prior art, the purpose of the present invention is to provide a yarn winding back pressure control method based on a radial basis function neural network, which can optimize the existing back pressure control scheme and achieve precise control of the back pressure between the yarn bobbin and the winding roller during the yarn winding process by controlling the motor torque instead of traditional mechanical devices such as springs, thereby improving the winding quality of the bobbin yarn.

[0005] The present invention provides a yarn winding back pressure control method based on radial basis function neural network, wherein the back pressure balancing device used in the method comprises a cradle, a pressure sensor, a motor, a winding roller and a frame. The winding roller is rotatably connected to the frame, and a pressure sensor is arranged between the winding roller and the frame. The cradle is used to install the bobbin; the motor adjusts the back pressure between the bobbin on the cradle and the winding roller by adjusting the output torque.

[0006] The yarn winding back pressure control method based on radial basis function neural network comprises the following steps: Step 1: Construct a RBF-PID joint control model consisting of a radial basis function neural network and a PID controller.

[0007] Step 2: Obtain the actual back pressure at the current time k through the pressure sensor ; Calculate the control error at the current time k ; Where r is the back pressure setting value.

[0008] Step 3: Input the control error e(k) into the PID controller; the PID controller inputs the control signal increment of the motor And the control signal at the current moment .

[0009] Step 4: Construct the input features of the radial basis neural network X= [ ∆ u(k),y(k),y(k−1)] T ; Input features Input RBF neural network. RBF neural network outputs identification information and the Jacobian matrix .

[0010] Step 5: Update the PID controller parameters according to the Jacobian matrix output by the radial basis neural network.

[0011] Step 6: Update the network parameters of the radial basis neural network through the particle swarm algorithm. The fitness function used in the particle swarm algorithm is .

[0012] Step 7: cyclically execute steps 2 to 6 to dynamically adjust the yarn winding back pressure.

[0013] As a preferred embodiment, the specific process of step six is: Step 6-1. Combine the center, width and weight parameters of the radial basis neural network to form the particle position parameters and construct the particle swarm in the initial state.

[0014] Step 6-2. Use the fitness function to optimize the individual extreme value of the particle; iteratively determine the individual optimal extreme value of the particle.

[0015] Step 6-3. Use the fitness function to optimize the group extreme value; iteratively update the particle group extreme value.

[0016] Step 6-4. After the iteration is completed, the extreme value of the particle swarm obtained by the final optimization is the optimal value obtained by the particle swarm algorithm; then the particles corresponding to the optimal value are used to generate the updated center, width and weight of the corresponding radial basis neural network.

[0017] As a preference, The process of particle position and velocity iteration in step 6 is as follows: In the formula, , It is the movement speed of the particle at the current and next moment during the iteration process; , are the particle positions at the current and next moments respectively; is the individual optimal extreme value; is the optimal extreme value of the group; and is the learning factor, and for A random number between is the inertia factor of the particle.

[0018] Inertia Factor The expression is: In the formula, is the maximum value of the inertia factor; is the minimum value of the inertia factor; is the maximum number of iterations in the particle swarm algorithm, is the current iteration number of the algorithm.

[0019] As a preferred embodiment, the specific process of step three is as follows: Step 3-1. Calculate the three inputs of the PID controller at the current time k: ; ; ; Step 3-2. Calculate the control signal increment of the control quantity and the control signal at the current moment as follows: in, is the PID controller output at the previous moment; , , They represent the proportional coefficient, integral coefficient and differential coefficient of the PID controller at the current moment k respectively.

[0020] Preferably, the specific process in step 4 is: the radial basis neural network includes an input layer, a hidden layer and an output layer which are connected in sequence.

[0021] The identification information output by the output layer as follows: in, is the node weight, is the activation function of the hidden layer structure.

[0022] The expression of the Jacobian matrix output by the radial basis neural network is: in, is the center vector of the i-th hidden layer node; is the width of the i-th hidden layer node.

[0023] Preferably, the activation function of the hidden layer structure adopts the Gaussian basis function, which is expressed as follows: in, is the Euclidean norm.

[0024] As a preference, the specific process of updating the PID controller parameters in step 5 is as follows: Step 5-1. Calculate the parameter increments of the proportional coefficient, integral coefficient, and differential coefficient of the PID controller , , as follows: in, The learning rate of the proportion, integration and differentiation of PID control; , , They are the three inputs of the PID controller at the current time k.

[0025] Step 5-2. Calculate the proportional parameter, integral parameter and differential parameter of the PID controller at the next moment , , as follows: Preferably, the cradle is U-shaped, including a connecting beam and two side arms. The inner ends of the two side arms are respectively connected to the two ends of the connecting beam. The cradle is rotatably connected to the frame. A bobbin mounting assembly is installed between the two side arms of the cradle. The motor is installed on the frame.

[0026] Preferably, a torque sensor is provided between the cradle and the motor.

[0027] Preferably, the two ends of the winding roller are rotatably connected to two second supports respectively. The two second supports arranged at intervals are both installed on the frame through pressure sensors.

[0028] It can be seen from the above technical solution that the beneficial effects of the present invention are: 1. The back pressure control method adopted by the present invention is optimized compared with the traditional back pressure control method. By controlling the motor torque instead of traditional mechanical devices such as springs to control the back pressure, the back pressure of each spindle of yarn can be kept consistent, thereby improving the winding quality of the yarn.

[0029] 2. The present invention constructs a fitness function based on the back pressure measurement value and the identification information output by the radial basis neural network, and then uses the particle swarm algorithm to correct the center, width and weight of the radial basis neural network, so that the trained network has a higher convergence speed and convergence accuracy. The control algorithm is optimized. While retaining the characteristics of traditional PID, it can also use the self-learning and adaptive capabilities of RBF neural network to correct the three parameters of PID online in real time according to the actual operation of the system, so as to adapt to the changes in the back pressure control system model caused by various factors, meet the control needs of nonlinear back pressure control, and have faster response speed and better control effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A three-dimensional schematic diagram of a back pressure control device used in the present invention; Figure 2 Flow chart of optimizing RBF neural network parameters using the PSO algorithm provided by the present invention; Figure 3 The schematic diagram of the radial basis function neural network used in the present invention to adjust PID parameters; Figure 4 The present invention provides Control flow chart of the neural network PID controller. DETAILED DESCRIPTION

[0031] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with the accompanying drawings.

[0032] like Figure 1 As shown, a yarn winding back pressure control method based on radial basis function neural network, the back pressure balancing device used in the method includes a cradle 2, a pressure sensor 3, a coupling 4, a torque sensor 5, a motor 6, a winding roller 7, a controller 8, a first support, a second support 9 and a frame 10.

[0033] The cradle 2 is U-shaped and includes a connecting beam and two side arms. The inner ends of the two side arms are respectively connected to the two ends of the connecting beam. The cradle 2 is rotatably connected to the frame 10 through a rotating shaft and two first supports. A bobbin mounting assembly is installed between the two side arms of the cradle 2. The bobbin mounting assembly is used to install the wound bobbin yarn 1. The axis of the bobbin yarn 1 is parallel to the rotation axis of the cradle 2. The motor 6 is installed on the frame 10. The rotating shaft of the cradle 2 is connected to the output shaft of the motor 6 through a coupling 4 and a torque sensor 5.

[0034] The two second supports 9 arranged at intervals are both mounted on the frame 10 through the pressure sensor 3. The two ends of the winding roller 7 are rotatably connected to the two second supports 9. The winding roller 7 is located below the bobbin 1 mounted on the cradle 2.

[0035] During operation, the winding roller 7 is driven to rotate by an external power element (generally a motor); the yarn is wound on the bobbin 1 through the winding roller 7. As more and more yarn is wound on the bobbin 1, the diameter of the bobbin 1 increases, pushing the cradle 2 to flip upward. As the diameter of the bobbin 1 increases, the weight of the bobbin 1 increases, and the back pressure between the bobbin 1 and the winding roller 7 increases; at this time, the motor 6 applies a torque to flip the cradle 2 upward through the coupling 4, so that the back pressure between the bobbin 1 and the winding roller 7 remains stable. As the weight of the bobbin 1 continues to increase, the torque applied by the motor 6 to the cradle 2 also continues to increase.

[0036] During the winding process, the torque sensor 5 measures the motor torque in real time; the pressure sensor 3 measures the change of the back pressure in real time and transmits the data to the controller 8; the controller 8 can control the motor 6 to output the corresponding torque according to the information input from the outside.

[0037] In this embodiment, the back pressure adjustment method is to change the cradle angle by controlling the motor torque, and give the bobbin a supporting force to offset the back pressure. When the detected back pressure does not meet the set value, the back pressure can be balanced by changing the motor torque, and a high-precision torque sensor is used to accurately measure the change of the motor torque.

[0038] like Figure 2 , Figure 3 and Figure 4 As shown, the present invention provides a yarn winding back pressure control method based on radial basis function neural network, which comprises the following steps: Step 1: Construct an RBF-PID joint control model consisting of an RBF neural network and a PID controller; and initialize the parameters of the RBF neural network (i.e., radial basis function neural network) and the PID controller. The input of the RBF-PID joint control model is the control error e(k), and the output is the parameters of the PID controller. During the yarn winding process, the PID controller dynamically adjusts the input current of the motor 6 to change the torque of the motor 6 on the cradle 2; at the same time, the RBF neural network periodically adjusts the parameters of the PID controller to improve the accuracy of the PID control. The parameters of the PID controller include the proportional parameter Kp, the integral parameter K i and the differential parameter K d .

[0039] Step 2: Set the back pressure setting value r, and obtain the actual back pressure at the current moment through the pressure sensor ; The actual back pressure value obtained by the pressure sensor The difference is compared with the back pressure setting value r. If the difference is not equal to 0, the difference is input into the RBF-PID joint control model as the control error e(k).

[0040] The expression of the control error e(k) at the current time k is: Among them, r is the back pressure setting value, and y(k) is the back pressure value at time k.

[0041] Step 3: Calculate the input and output of the PID controller based on the control error e(k).

[0042] Step 3-1. The three inputs of the PID controller at the current moment are: , , ; in, are the control errors at the current moment, the previous moment, and the two previous moments respectively.

[0043] Step 3-2. The PID controller uses the incremental PID algorithm to calculate the increment of the control quantity , and then obtain the output of the PID controller as follows: in, is the PID controller output at the previous moment; , , They represent the proportional coefficient, integral coefficient and differential coefficient of the PID controller at the current moment k respectively.

[0044] Step 3-3. According to the output of PID controller Get the current control current I of the motor at the current moment .

[0045] Step 3-4. The control current I obtained in step 3-3 Output the motor torque at the current moment.

[0046] Step 4: Calculate the identification information of the RBF neural network at the current time k And Jacobian matrix. The RBF neural network is divided into three layers, which are the input layer, hidden layer and output layer connected in sequence.

[0047] Input layer input The expression is as follows: X= [ ∆ u(k),y(k),y(k−1)] T Among them, y(k) and y(k-1) are the back pressure values ​​at the current time k and the previous time k-1 respectively.

[0048] The hidden layer structure is an activation function, which is composed of radial basis functions. In this embodiment, the radial basis functions are composed of Gaussian basis functions. for:

[0049] h= [ h 1 , h 2 , h 3 ··· h m ] T in, are the node radial basis functions of m hidden layer nodes respectively; m is the number of hidden layer nodes.

[0050] Gaussian basis function The expression is as follows: in, is the Euclidean norm; is the center vector of the i-th hidden layer node; is the width of the i-th hidden layer node.

[0051] The identification information output by the output layer as follows: In the formula, is the node weight, Gaussian basis function, The expression of the Jacobian matrix output by the RBF neural network is: Step 5: According to the Jacobian matrix output by the RBF neural network, use the gradient descent method to adjust the PID controller parameter increment.

[0052] Use the squared error function as a performance indicator: In the formula represents the performance index at time k, represents the back pressure control error at time k.

[0053] in, The learning rate of the proportion, integration and differentiation of PID control; Represents the Jacobian matrix of the controlled object, which is expressed as .

[0054] According to the PID controller parameter increment obtained in step 5, the PID parameters are updated as follows: in, , , They represent the proportional parameter, integral parameter and differential parameter input to the PID controller at the next moment k+1 after correction respectively; Step 6: Use the PSO algorithm to adjust the RBF neural network parameters, including the center, width, and weight of the RBF neural network. The specific process is as follows:

[0055] Step 6-1. Combine the center, width and weight parameters of the RBF neural network to form the particle position parameters. Then assign values ​​to the number of particles, particle positions and movement speeds in a random manner.

[0056] Step 6-2. First, optimize the individual extreme value of the particle. The fitness value is used to measure the individual extreme value. The fitness value is The objective function value of the RBF neural network is that if the fitness value of the current particle is less than the individual extreme value of the particle at the previous position , then the individual extreme value of the particle is updated, and the individual optimal extreme value of the particle is determined through continuous iteration.

[0057] Step 6-3. Optimize the group extreme value. The fitness value is also used to measure the group extreme value. By calculating the fitness value, the optimal individual extreme value in the current particle swarm will be obtained; if this optimal individual extreme value is smaller than the previous group extreme value , the group extreme value is updated, and the particle speed and position information are also updated.

[0058] The PSO algorithm continuously updates the speed and position according to the following algorithm: In the formula, , It is the movement speed of the particle at the current and next moment during the iteration process; , are the particle positions at the current and next moments respectively; is the individual optimal extreme value; is the optimal extreme value of the group; and is the learning factor, and for A random number between is the inertia factor of the particle.

[0059] Inertia Factor The calculation formula is as follows: In the formula, is the maximum value of the inertia factor; is the minimum value of the inertia factor; is the maximum number of iterations in the particle swarm algorithm, is the current iteration number of the algorithm; The fitness function is: In the formula, is the identification information of the neural network at time k, is the back pressure value at time k.

[0060] Step 6-4. If the maximum number of iterations of the particle swarm is reached or the convergence accuracy tends to be stable, the extreme value of the particle swarm finally found is the optimal value obtained by the particle swarm algorithm. Then the particles corresponding to the optimal value, that is, the one-dimensional array, are decoded into the corresponding network parameters, that is, the center, width and weight of the RBF neural network.

[0061] Step 6-5. Update the RBF neural network parameters according to the center, width and weight of the RBF neural network obtained in step 6-4.

[0062] Step 7: Execute steps 2 to 6 cyclically, and perform sampling with the preset sampling time as the cycle.

Claims

1. A yarn winding back pressure control method based on radial basis function neural network, characterized in that: The back pressure balancing device used comprises a cradle (2), a pressure sensor (3), a motor (6), a winding roller (7) and a frame (10); the winding roller (7) is rotatably connected to the frame (10), and a pressure sensor (3) is provided between the winding roller (7) and the frame (10); the cradle (2) is used for mounting a bobbin; the motor (6) adjusts the back pressure between the bobbin on the cradle (2) and the winding roller (7) by adjusting the output torque; The yarn winding back pressure control method based on radial basis function neural network comprises the following steps: Step 1: Construct a RBF-PID joint control model consisting of a radial basis function neural network and a PID controller; Step 2: Obtain the actual back pressure at the current time k through the pressure sensor ; Calculate the control error at the current time k ; Where r is the back pressure setting value; Step 3: Input the control error e(k) into the PID controller; the PID controller inputs the control signal increment of the motor And the control signal at the current moment ; Step 4: Construct the input features of the radial basis neural network ; Input features Input radial basis neural network; radial basis neural network output identification information and the Jacobian matrix ; Step 5: Update the PID controller parameters according to the Jacobian matrix output by the radial basis neural network; Step 6: Update the network parameters of the radial basis neural network through the particle swarm algorithm; the fitness function used in the particle swarm algorithm is ; Step 7: cyclically execute steps 2 to 6 to dynamically adjust the yarn winding back pressure.

2. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: The specific process of step six is ​​as follows: Step 6-1. Combine the center, width and weight parameters of the radial basis neural network to form particle position parameters, and construct a particle swarm in the initial state; Step 6-2. Use the fitness function to optimize the individual extreme value of the particle; iteratively determine the individual optimal extreme value of the particle; Step 6-3. Use the fitness function to optimize the group extreme value; iteratively update the particle group extreme value; Step 6-4. After the iteration is completed, the extreme value of the particle swarm obtained by the final optimization is the optimal value obtained by the particle swarm algorithm; then the particles corresponding to the optimal value are used to generate the updated center, width and weight of the corresponding radial basis neural network.

3. The yarn winding back pressure control method based on radial basis function neural network according to claim 2, characterized in that: The process of particle position and velocity iteration in step 6 is as follows: In the formula, , It is the movement speed of the particle at the current and next moment during the iteration process; , are the particle positions at the current and next moments respectively; is the individual optimal extreme value; is the optimal extreme value of the group; and is the learning factor, and for A random number between is the inertia factor of the particle; Inertia Factor The expression is: In the formula, is the maximum value of the inertia factor; is the minimum value of the inertia factor; is the maximum number of iterations in the particle swarm algorithm, is the current iteration number of the algorithm.

4. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: The specific process of step three is as follows: Step 3-1. Calculate the three inputs of the PID controller at the current time k: ; ; ; Step 3-2. Calculate the control signal increment of the control quantity and the control signal at the current moment as follows: in, is the PID controller output at the previous moment; , , They represent the proportional coefficient, integral coefficient and differential coefficient of the PID controller at the current moment k respectively.

5. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: The specific process in step 4 is as follows: the radial basis neural network includes an input layer, a hidden layer and an output layer connected in sequence; The identification information output by the output layer as follows: in, is the node weight, is the activation function of the hidden layer structure; The expression of the Jacobian matrix output by the radial basis neural network is: in, is the center vector of the i-th hidden layer node; is the width of the i-th hidden layer node.

6. The yarn winding back pressure control method based on radial basis function neural network according to claim 5, characterized in that: The activation function of the hidden layer structure adopts the Gaussian basis function, which is expressed as follows: in, is the Euclidean norm.

7. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: Step 5: The specific process of updating the PID controller parameters is as follows: Step 5-1. Calculate the parameter increments of the proportional coefficient, integral coefficient, and differential coefficient of the PID controller , , as follows: in, The learning rate of the proportion, integration and differentiation of PID control; , , They are the three inputs of the PID controller at the current time k; Step 5-2. Calculate the proportional parameter, integral parameter and differential parameter of the PID controller at the next moment , , as follows: 。 8. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: The cradle (2) is U-shaped and comprises a connecting beam and two side arms; the inner ends of the two side arms are respectively connected to the two ends of the connecting beam; the cradle (2) is rotatably connected to the frame (10); a bobbin mounting assembly is installed between the two side arms of the cradle (2); and the motor (6) is installed on the frame (10).

9. The yarn winding back pressure control method based on radial basis function neural network according to claim 8, characterized in that: A torque sensor (5) is provided between the cradle (2) and the motor (6).

10. The yarn winding back pressure control method based on radial basis function neural network according to claim 1, characterized in that: The two ends of the winding roller (7) are rotatably connected to two second supports (9) respectively; the two second supports (9) arranged at intervals are both mounted on the frame (10) via pressure sensors (3).

Images