A method for optimizing enameled wire drawing forming process parameters based on a PID algorithm

Abstract

The present application relates to a kind of enameled wire wire drawing forming process parameter optimization method based on PID algorithm, comprising the following steps, data acquisition and preprocessing, neural network decoupler decouples, preset PID basic control, BP neural network adaptive setting, RBF neural network disturbance prediction, superimposed feedforward compensation, form final control signal, actuator action.The enameled wire wire drawing forming process parameter optimization method based on PID algorithm effectively strips speed, tension, swing pole strong coupling by using neural network decoupler, decouples loss function quantization residual index, deviation change rate threshold security triggers BP network self-setting PID, according to the dynamic parameter adjustment of roll diameter, die length, swing pole deviation triggers RBF network one cycle in advance accurate prediction disturbance amplitude and direction, generate feedforward compensation and amplitude adaptation, completely solve the fixed feedforward and variable gain PID collaborative problem, realize the active inhibition of foreseeable disturbance such as traction wheel pulsation, speed fluctuation.

Description

Technical Field

[0001] This invention relates to the field of enameled wire drawing technology, specifically to a method for optimizing process parameters of enameled wire drawing based on PID algorithm. Background Technology

[0002] Enamelled wire drawing is a fundamental and core process in enamelled wire production. Through multiple consecutive drawing passes, the diameter of the rod material is gradually reduced to the target size while ensuring wire diameter accuracy, surface quality, and mechanical properties. The principle is to utilize the plastic deformation characteristics of metals. By using a drawing die, the wire undergoes plastic deformation under tension, resulting in a decrease in cross-sectional area and an increase in length. Enamelled wire drawing is a precisely controlled system engineering process that requires balancing multiple factors such as deformation amount, speed, and tension to produce high-quality bare wire that meets the requirements, providing a perfect base for subsequent coating processes.

[0003] In the current technology, PID control is a commonly used method in the process of drawing and forming enameled wire. It collects the actual position of the rod, the real-time tension of the wire, and the real-time speed, then calculates the position deviation, tension deviation, and speed deviation of the rod, outputs the adjustment amount based on the position deviation, tension deviation, and speed deviation, and then executes the adjustment action. In the actual enameled wire drawing and forming process, PID control performs closed-loop calculations for a single deviation, lacking multi-variable decoupling logic. It cannot handle the strong coupling relationship between speed, tension, and position within the system, easily leading to chain fluctuations where adjusting one variable disturbs others, failing to achieve multi-objective coordinated stability. Furthermore, PID control only adjusts after an actual deviation occurs, unable to proactively suppress foreseeable disturbances such as traction wheel pulsation, speed fluctuations, and coil diameter changes. The adjustment action always lags behind process disturbances. Moreover, as processing time increases, if time-varying factors such as mold wear, changes in wire characteristics, or load variations occur, PID control cannot adaptively adjust, leading to problems such as adjustment failure, slow response, or over-adjustment oscillation. Therefore, a PID algorithm-based optimization method for enameled wire drawing and forming process parameters is proposed to address these issues. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for optimizing process parameters in enameled wire drawing based on a PID algorithm. This method offers advantages such as decoupling and prediction, solving the problems inherent in PID control during actual enameled wire drawing. PID control performs closed-loop calculations for a single deviation, lacking multi-variable decoupling logic. It cannot handle the strong coupling relationships between speed, tension, and position within the system, easily leading to chain fluctuations where adjusting one variable disturbs others, hindering multi-objective coordinated stability. Furthermore, PID control only adjusts after an actual deviation occurs, failing to proactively suppress foreseeable disturbances such as traction wheel pulsation, speed fluctuations, and coil diameter changes. The adjustment action always lags behind process disturbances. Moreover, with increasing processing time, if time-varying factors such as mold wear, changes in wire characteristics, or load variations occur, PID control cannot adaptively adjust, leading to problems such as adjustment failure, slow response, or over-adjustment oscillation.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing enameled wire drawing process parameters based on PID algorithm, comprising the following steps: S1. Collect the given speed of the wire drawing machine Given tension Given the position of the pendulum rod actual speed Actual tension Actual position of the swing arm This forms the basic dataset, which is then preprocessed to obtain a clean basic dataset. S2. Based on a clean base dataset, a decoupler is constructed using a three-layer feedforward neural network. The decoupler is then used to decouple multiple variables, and the actual speed after decoupling is obtained. Actual tension after decoupling Actual position of the pendulum after decoupling And calculate the deviation and the rate of change of deviation; S3. Based on decoupling , , The system calculates the basic adjustment amounts of speed, tension, and swing arm position using preset PID parameters and deviations, and then converts these basic adjustment amounts into control signals for the actuator.

[0006] Furthermore, the steps in S2 for decoupling multiple variables using a decoupler and calculating the deviation and the rate of change of deviation include the following: The preprocessed part in S1 , , , , , A total of six parameters are input to the neural network decoupler; Decoupling based on , , , , , The network weights are adjusted online with six parameters in a lightweight manner, and then dynamically fitted. , , A nonlinear coupling mathematical model among the three; The tension on the velocity and the disturbance of the pendulum position are calculated in reverse based on the nonlinear coupled mathematical model. The amount of disturbance caused by the velocity and position of the pendulum during tension. The velocity and tension disturbances experienced by the pendulum position ,based on , , Decoupling compensation correction is applied to the actual value to obtain , , ; use , , and , , Perform difference calculations, and then use differential operations to calculate the real-time tension deviation after decoupling. Real-time pendulum position deviation after decoupling , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling And constitute a real dataset; actual speed after output decoupling Actual tension after decoupling Actual position of the pendulum after decoupling And real datasets.

[0007] Furthermore, it also includes online adaptive tuning of PID parameters based on a BP neural network, with the following specific steps: Obtain the decoupling tension deviation change rate output in claim 2. Rate of change of pendulum position deviation and respectively with engineering threshold , Compare; like ,or If the adaptive tuning is skipped, the current PID parameters are kept unchanged, and step S3 is executed directly. like ,and Then, the PID parameters are adaptively tuned online based on the BP neural network; The PID parameters are adaptively tuned online based on a BP neural network, including real-time values ​​of the wire drawing machine's roll diameter. Mold usage time , to use real datasets, , All data are normalized before being input into the BP neural network. First, a set of initial PID parameters is output using a BP neural network, which is then assigned to the velocity PID, tension PID, and swing arm position PID loops, respectively. Simultaneously, based on... , , Calculate the actual tension fluctuation variance Actual swing amplitude of the pendulum Actual adjustment overshoot and will , , As a practical performance evaluation indicator; The actual performance evaluation index and the preset tension fluctuation variance are combined. Preset swing amplitude of the pendulum Preset adjustment of overshoot Perform difference calculations to obtain the performance error value. Then As a back feedback signal for the BP network, it is backpropagated based on gradient descent to correct the weights of each layer of the network. The BP neural network synchronously outputs the optimal PID parameters for speed, tension, and swing arm position, which are adapted to the current wire drawing conditions. The preset PID parameters used in step S3 of claim 1 are updated to the optimal PID parameters for speed, tension, and pendulum position, and real-time operating condition feature values ​​are extracted. .

[0008] Furthermore, it also includes predicting disturbances in advance and generating feedforward compensation based on RBF neural networks, specifically including the following steps: Obtain the decoupled real-time pendulum position deviation output by claim 2. and the deviation threshold of the preset swing arm position Compare; like If so, the feedforward compensation is zero; like Then, based on the RBF neural network, the disturbance is predicted in advance and the feedforward compensation is generated; Based on the RBF neural network, disturbances are predicted in advance and feedforward compensation is generated, including the acquisition of the traction wheel speed of the wire drawing machine. Wire feed speed Calculate the position deviation rate of the pendulum. Normalized input is given to the RBF neural network. , , , Real-time operating condition characteristic values ; Based on the online lightweight fitting of the time-varying law of perturbation using RBF neural network, the mapping relationship between traction wheel pulsation, coil diameter change, wire feed speed fluctuation, tension perturbation, and speed fluctuation is obtained. The predicted tension disturbance amplitude for the next control cycle is obtained based on the fitted time-varying pattern of the disturbance. Predicting speed fluctuations Predicting the direction of tension disturbance Predicting the direction of speed fluctuations ; Based on the next control cycle , , , Generate tension feedforward compensation amount Velocity feedforward compensation Meanwhile, based on the optimal speed PID parameters and optimal tension PID parameters output in claim 3, the compensation amount is adjusted by amplitude adaptation. Assign the optimal parameters of the speed PID, tension PID, and swing arm position PID outputs in claim 3, and connect them to the decoupled outputs in claim 2. , , It forms an independent closed loop with the real dataset, calculates the basic adjustment amount and superimposes the feedforward compensation amount after amplitude adaptation correction, and finally converts it into the control signal of the actuator.

[0009] Furthermore, the dynamic fitting in claim 2 , , A nonlinear coupling mathematical model among the three; Let the connection weight matrix from the input layer to the hidden layer of the decoupler be... The bias vector of the hidden layer is The activation function of the hidden layer is The connection weight matrix from the hidden layer to the output layer is: The bias vector of the output layer is The input vector of the decoupler is The decoupler output is a coupling interference vector. ; The general formula for the nonlinear coupling mathematical model is as follows: ; Its expansion is: ; in, , , The weight matrix of the output layer corresponding to the hidden layer The three row vectors, , , For the output layer bias vector The three elements; The formula for subtracting the interference from the actual value is as follows: .

[0010] Furthermore, in claim 5 , The weight adjustment is based on the decoupling loss function constraint, the function formula of which is as follows: ; in, , , This represents the true value without coupling under the enameled wire drawing process. The normalized coefficients are always adjusted along the loss function. Update the gradient in the opposite direction, i.e. towards Correction of the decreasing direction; The online lightweight correction formula for the connection weights and bias vectors is as follows: , ; in, For the learning rate, and , , Let be the gradient, and let the gradient be subject to a cutoff constraint. Its calculation formula is as follows: ,and , , All are truncated according to the same rules as above.

[0011] Furthermore, the performance error value in claim 3 The calculation formula is as follows: ; in, This indicates that the actual performance evaluation indicators completely match the preset performance evaluation indicators; The formula for the loss function of a BP neural network is as follows: ; Among them, if the performance error value If so, the weights of the BP neural network are not modified; If performance error value Then, the weights of the BP neural network are slightly adjusted using the gradient descent method; like Then, the weights of the BP neural network are modified to a limited extent using the gradient descent method, and an alarm signal is triggered simultaneously.

[0012] Furthermore, the optimal speed PID parameter in claim 3 includes the optimal speed PID proportional coefficient. Optimal integral coefficient for speed PID Speed ​​PID optimal differential coefficients ; The optimal parameters of the tension PID include the optimal proportional coefficient of the tension PID. Optimal integral coefficient of tension PID Optimal differential coefficients of tension PID ; The optimal parameter for the PID controller at the swing arm position is the optimal proportional coefficient for the PID controller at the swing arm position. Optimal integral coefficients for PID control of rocker arm position PID optimal differential coefficients for lever position ; Real-time operating condition characteristic values This represents the comprehensive characteristic quantity of the current wire drawing process, derived from the real-time value of the roll diameter. Mold usage time Tension deviation change rate The optimal proportional gain for PID control is extracted using the following expression: ; in, Maximum value of roll diameter process Rated service life of the mold The maximum value of the tension PID proportional coefficient is given, and all components are mapped to... , , ; The optimal process constraints for PID parameters are: .

[0013] Furthermore, the real-time operating condition feature value in claim 8 of For dynamic adjustment, let the influence degree be . ,and ; This includes the influence of roll diameter as The influence of mold wear is The influence of tension fluctuation is The influence of PID regulation is Then, normalization is performed and mapped to... Its expression is as follows: ; in, For the real-time change rate of roll diameter, This represents the maximum value of the tension deviation change rate in the process. The weighting factor is the rate of change of roll diameter; Based on dynamic coefficient base value With influence Positive correlation, and calculate the basic value of the dynamic coefficient. Its expression is as follows: , and ; For the basic value of dynamic coefficient Normalization is performed to obtain the normalized coefficients. Its expression is as follows: ; And on Apply industrial hard constraints to obtain constraint coefficients Its expression is as follows: ; Then to Perform a first-order low-pass smoothing to obtain Dynamically adjusted coefficients Its expression is as follows: ; in, This represents the dynamic coefficient of the previous control cycle. For the control cycle of the wire drawing machine, As the smoothing coefficient, the final dynamic weighting coefficient satisfies , .

[0014] Furthermore, it also includes unified scheduling of multi-source data and adaptive input based on operating conditions, with the specific steps as follows: Construct a multi-source data pool, which contains three types of data: primary data, secondary data, and tertiary data. One type of data includes a given speed Given tension Given the position of the pendulum rod actual speed Actual tension Actual position of the swing arm ; The second type of data includes real-time volume diameter values. Mold usage time ; The three types of data include traction wheel speed. Wire feed speed ; Obtain the decoupled real-time pendulum position deviation output by claim 2. , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling ; Based on the current operating condition control mode, execute the data scheduling strategy; If only the basic control steps described in claim 1 are performed, then one type of data is selected from the multi-source data pool and input into the neural network decoupler described in claim 2; like ,and Then, two types of data are selected from the multi-source data pool, and after being normalized together with the real dataset described in claim 2, they are input into the BP neural network for PID parameter self-tuning. like Then, a third type of data is selected from the multi-source data pool, and combined with the real-time volume diameter value D and real-time operating condition characteristic values ​​collected in claim 3. And the pendulum position deviation rate in claim 4 After being normalized, the data are input into the RBF neural network for perturbation prediction and feedforward compensation generation.

[0015] Compared with the prior art, the present invention provides a method for optimizing the process parameters of enameled wire drawing based on PID algorithm, which has the following beneficial effects: 1. The method for optimizing the process parameters of enameled wire drawing based on PID algorithm uses a three-layer feedforward neural network to build a decoupler, learns online and compensates for nonlinear coupling interference between the three channels in real time, separates the intertwined actual values ​​into independent and controllable decoupled actual values, breaks the strong coupling chain of speed, tension and swing arm, and achieves multi-objective cooperative stable control. 2. The method for optimizing the process parameters of enameled wire drawing based on PID algorithm transforms the degree of coupling residue into a specific numerical index through the decoupling loss function, making the direction of decoupling weight correction clear and convergent controllable. The introduction of gradient cutoff constraint effectively suppresses parameter divergence caused by sensor spike noise and start-up and shutdown impact, ensuring the long-term stable operation of the decoupling module in continuous production. 3. This method for optimizing the process parameters of enameled wire drawing based on PID algorithm uses the deviation change rate threshold as a safety trigger condition. It only starts BP neural network tuning when the working conditions are stable. The real-time value of the roll diameter and the mold usage time are the core working condition features input. It drives the PID parameters to dynamically self-tune with the winding and unwinding inertia and friction characteristics. It also extracts the real-time working condition feature values ​​as the comprehensive criterion for the whole working conditions. The dynamic weight adjustment mechanism ensures that the feature values ​​always focus on the current dominant factors, so that the control quality remains optimal throughout the entire life cycle. 4. This PID algorithm-based optimization method for enameled wire drawing process parameters uses the decoupled swing arm position deviation as the trigger criterion, integrates multi-source information such as traction wheel speed, wire feed speed, and working condition characteristics, and uses an RBF neural network to accurately predict the amplitude and direction of disturbances one cycle ahead, generating feedforward compensation. The amplitude adaptation correction mechanism makes the feedforward intensity automatically scale with the PID proportional coefficient, completely solving the coordination problem between fixed feedforward and variable gain PID, and realizing the active suppression of predictable disturbances such as traction wheel pulsation, speed fluctuation, and sudden changes in wire diameter. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for optimizing enameled wire drawing process parameters based on a PID algorithm, as proposed in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: Please refer to Figure 1 The method for optimizing enameled wire drawing process parameters based on PID algorithm in this embodiment includes the following steps: S1. Collect the given speed of the wire drawing machine Given tension Given the position of the pendulum rod actual speed Actual tension Actual position of the swing arm This forms the basic dataset, which is then preprocessed to obtain a clean basic dataset. S2. Based on a clean base dataset, a decoupler is constructed using a three-layer feedforward neural network. The decoupler is then used to decouple multiple variables, and the actual speed after decoupling is obtained. Actual tension after decoupling Actual position of the pendulum after decoupling And calculate the deviation and the rate of change of deviation; S3. Based on decoupling , , The system calculates the basic adjustment amounts of speed, tension, and swing arm position using preset PID parameters and deviations, and then converts these basic adjustment amounts into control signals for the actuator.

[0019] It should be noted that only data at a constant speed is collected. Given tension Given the position of the pendulum rod actual speed Actual tension Actual position of the swing arm Six core signals form a basic dataset that minimizes sensor dependence, significantly reducing hardware costs and field deployment difficulty, and is suitable for basic wire drawing equipment that does not have additional sensing capabilities; Based on this, a decoupler is constructed using a three-layer feedforward neural network, and the network weights are adjusted online in a lightweight manner for dynamic fitting. , , The nonlinear coupling mathematical model among the three effectively solves the multi-loop interference problem caused by the mutual coupling of speed, tension, and swing rod position in traditional wire drawing control.

[0020] In S2, the process of using a decoupler to decouple multiple variables and calculate the deviation and the rate of change of deviation includes the following steps: The preprocessed part in S1 , , , , , A total of six parameters are input to the neural network decoupler; Decoupling based on , , , , , The network weights are adjusted online with six parameters in a lightweight manner, and then dynamically fitted. , , A nonlinear coupling mathematical model among the three; The tension on the velocity and the disturbance of the pendulum position are calculated in reverse based on the nonlinear coupled mathematical model. The amount of disturbance caused by the velocity and position of the pendulum during tension. The velocity and tension disturbances experienced by the pendulum position ,based on , , Decoupling compensation correction is applied to the actual value to obtain , , ; use , , and , , Perform difference calculations, and then use differential operations to calculate the real-time tension deviation after decoupling. Real-time pendulum position deviation after decoupling , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling And constitute a real dataset; actual speed after output decoupling Actual tension after decoupling Actual position of the pendulum after decoupling And real datasets.

[0021] It should be noted that the three coupling interference quantities are calculated and compensated in reverse. , , The actual values ​​that are intertwined , , The actual values ​​after being separated into independent decoupled values , , This makes the subsequent PID controller deal with an object that is approximately a single input and single output, which greatly reduces the design difficulty of the control loop and the workload of parameter tuning. By subtracting and differentiating the decoupled actual value from the given value, this step generates the high signal-to-noise ratio deviation and deviation rate of change. Since the coupling interference has been removed, the calculated values ​​at this point... , , , It truly reflects the tracking error and control requirements of each loop, rather than the pseudo-deviation caused by fluctuations in other loops; This provides reliable operating condition feature inputs for the adaptive tuning of the BP neural network in Example 2, and also provides high-confidence deviation change trends for the disturbance prediction of the RBF neural network in Example 3, fundamentally ensuring the effectiveness and convergence stability of subsequent intelligent control algorithms.

[0022] Among them, dynamic fitting , , A nonlinear coupling mathematical model among the three; Let the connection weight matrix from the input layer to the hidden layer of the decoupler be... The bias vector of the hidden layer is The activation function of the hidden layer is The connection weight matrix from the hidden layer to the output layer is: The bias vector of the output layer is The input vector of the decoupler is The decoupler output is a coupling interference vector. ; The general formula for the nonlinear coupling mathematical model is as follows: ; Its expansion is: ; in, , , The weight matrix of the output layer corresponding to the hidden layer The three row vectors, , , For the output layer bias vector The three elements; The formula for subtracting the interference from the actual value is as follows: ; It should be noted that the model input vector also contains the given value. , , and actual value , , This enables the decoupler to learn both the forward control path and the reverse coupling path simultaneously. The forward control path reflects the driving relationship between the control quantity and the controlled quantity, while the reverse coupling path reflects the mutual interference channels between the loops. This bidirectional learning mechanism significantly improves the ability to fit complex coupling relationships. Hidden layer uses The activation function effectively alleviates the gradient vanishing problem while ensuring the nonlinear approximation capability of the neural network, and has extremely high computational efficiency, perfectly meeting the real-time requirements of the 10ms control cycle of the wire drawing machine.

[0023] In claim 5 , The weight adjustment is based on the decoupling loss function constraint, the function formula of which is as follows: ; in, , , This represents the true value without coupling under the enameled wire drawing process. The normalized coefficients are always adjusted along the loss function. Update the gradient in the opposite direction, i.e. towards Correction of the decreasing direction; The online lightweight correction formula for the connection weights and bias vectors is as follows: , ; in, For the learning rate, and , , Let be the gradient, and let the gradient be subject to a cutoff constraint. Its calculation formula is as follows: ,and , , All are truncated according to the same rules as above.

[0024] It should be noted that by defining the decoupling loss function This establishes a clear and quantifiable optimization objective for the neural network decoupler, and the loss function is based on the actual value after decoupling. , , Real value uncoupled from process , , The mean square error is used as a metric to transform the decoupling effect into a specific, calculable numerical index. When decoupling loss function When it approaches 0, it indicates , , It has approached the ideal uncoupled state; the mutual interference between velocity, tension, and pendulum position has been largely eliminated. Conversely, Increasing the weight indicates that the decoupler weights have deviated from the optimal region and need to be corrected in time. This solves the fundamental defects of traditional decoupling methods, such as the lack of a unified evaluation standard and the difficulty in quantifying optimization. The online lightweight correction formula uses the gradient descent method along... The weight matrix and bias vector are updated in the opposite direction of the gradient, so that the decoupler converges to a better decoupling state in each control cycle. The gradient is truncated, which solves the gradient explosion caused by instantaneous spike noise of the sensor and the risk of parameter divergence of the neural network under non-stationary conditions. Gradient truncation achieves a significant improvement in system robustness at extremely low computational cost.

[0025] Example 2: Please refer to Figure 1 Based on Example 1, a method for optimizing enameled wire drawing process parameters based on PID algorithm further includes online adaptive tuning of PID parameters based on BP neural network. The specific steps are as follows: Obtain the decoupling tension deviation change rate output in claim 2. Rate of change of pendulum position deviation and respectively with engineering threshold , Compare; like ,or If the adaptive tuning is skipped, the current PID parameters are kept unchanged, and step S3 is executed directly. like ,and Then, the PID parameters are adaptively tuned online based on the BP neural network; The PID parameters are adaptively tuned online based on a BP neural network, including real-time values ​​of the wire drawing machine's roll diameter. Mold usage time , to use real datasets, , All data are normalized before being input into the BP neural network. First, a set of initial PID parameters is output using a BP neural network, which is then assigned to the velocity PID, tension PID, and swing arm position PID loops, respectively. Simultaneously, based on... , , Calculate the actual tension fluctuation variance Actual swing amplitude of the pendulum Actual adjustment overshoot and will , , As a practical performance evaluation indicator; The actual performance evaluation index and the preset tension fluctuation variance are combined. Preset swing amplitude of the pendulum Preset adjustment of overshoot Perform difference calculations to obtain the performance error value. Then As a back feedback signal for the BP network, it is backpropagated based on gradient descent to correct the weights of each layer of the network. The BP neural network synchronously outputs the optimal PID parameters for speed, tension, and swing arm position, which are adapted to the current wire drawing conditions. The preset PID parameters used in step S3 are updated to the optimal PID parameters for speed, tension, and pendulum position, and real-time operating condition feature values ​​are extracted. .

[0026] It should be noted that this embodiment uses an online adaptive tuning method for PID parameters based on a BP neural network on the basis of decoupled control, which solves the problem that traditional PID parameters are fixed and cannot adapt to time-varying working conditions such as changes in roll diameter and die wear. BP network tuning is only initiated when both tension and the rate of change of pendulum deviation are below the threshold, thus avoiding mislearning under disturbed conditions and incorporating real-time values ​​of the roll diameter. Mold usage time As a network input, the PID parameters can be adaptively adjusted according to the winding and unwinding inertia and the friction characteristics of the die, organically linking the originally isolated PID tuning with disturbance suppression, forming a complete intelligent control link of perception, decision-making and compensation.

[0027] Among them, the performance error value in claim 3 The calculation formula is as follows: ; in, This indicates that the actual performance evaluation indicators completely match the preset performance evaluation indicators; The formula for the loss function of a BP neural network is as follows: ; Among them, if the performance error value If so, the weights of the BP neural network are not modified; If performance error value Then, the weights of the BP neural network are slightly adjusted using the gradient descent method; like Then, the weights of the BP neural network are modified to a limited extent using the gradient descent method, and an alarm signal is triggered simultaneously.

[0028] It should be noted that quantifying multi-dimensional process quality requirements into a unified performance error index This solves the problem of balancing tension fluctuation, lever oscillation, and overshoot in wire drawing control by using a weighted normalization formula. , , The relative deviation from the process preset target is compressed to the same dimension, and the weight coefficients can be flexibly adjusted according to the production line quality preference, so that the learning target of the BP network is always aligned with the process requirements. Judging the range of performance error values ​​can prevent parameters from going out of control under abnormal operating conditions and provide fault clues for process engineers.

[0029] Wherein, the optimal parameters of the speed PID in claim 3 include the optimal proportional coefficient of the speed PID. Optimal integral coefficient for speed PID Speed ​​PID optimal differential coefficients ; The optimal parameters for tension PID include the optimal proportional coefficient for tension PID. Optimal integral coefficient of tension PID Optimal differential coefficients of tension PID ; The optimal parameter for the PID controller at the rocker arm position is the optimal proportional coefficient for the PID controller at the rocker arm position. Optimal integral coefficients for PID control of rocker arm position PID optimal differential coefficients for lever position ; Real-time operating condition characteristic values This represents the comprehensive characteristic quantity of the current wire drawing process, derived from the real-time value of the roll diameter. Mold usage time Tension deviation change rate The optimal proportional gain for PID control is extracted using the following expression: ; in, Maximum value of roll diameter process Rated service life of the mold The maximum value of the tension PID proportional coefficient is given, and all components are mapped to... , , ; The optimal process constraints for PID parameters are: .

[0030] It should be noted that by using linear weighting, four high-dimensional operating condition information parameters—roll diameter, die duration, tension fluctuation, and current tension ratio—are compressed into a single-dimensional comprehensive index. This addresses the issue that downstream RBF neural networks cannot directly process raw operating data with multiple dimensions and varying units, and the need for a lightweight and standardized information interface between the two intelligent modules: adaptive PID and feedforward compensation. The transmission enables the feedforward module to sense the urgency of the overall working conditions and automatically enhance the disturbance suppression under heavy working conditions such as large roll diameter and the end of the die process.

[0031] Among them, the real-time operating condition feature value in claim 8 of For dynamic adjustment, let the influence degree be . ,and ; This includes the influence of roll diameter as The influence of mold wear is The influence of tension fluctuation is The influence of PID regulation is Then, normalization is performed and mapped to... Its expression is as follows: ; in, For the real-time change rate of roll diameter, This represents the maximum value of the tension deviation change rate in the process. The weighting factor is the rate of change of roll diameter; Based on dynamic coefficient base value With influence Positive correlation, and calculate the basic value of the dynamic coefficient. Its expression is as follows: , and ; For the basic value of dynamic coefficient Normalization is performed to obtain the normalized coefficients. Its expression is as follows: ; And on Apply industrial hard constraints to obtain constraint coefficients Its expression is as follows: ; Then to Perform a first-order low-pass smoothing to obtain Dynamically adjusted coefficients Its expression is as follows: ; in, This represents the dynamic coefficient of the previous control cycle. For the control cycle of the wire drawing machine, As the smoothing coefficient, the final dynamic weighting coefficient satisfies , .

[0032] It should be noted that the real-time operating condition characteristic values The dynamic weight adjustment mechanism completely solves the problem that fixed weights cannot adapt to the drift of the dominant factors in the working condition, and introduces the influence degree. Real-time quantification of the combined contribution of roll diameter change rate, die wear degree, tension fluctuation intensity, and current PID control capability to the operating conditions, enabling... It can dynamically focus on the most prominent technological contradictions at present.

[0033] Example 3: Please refer to Figure 1 Based on Example 2, a method for optimizing enameled wire drawing process parameters based on PID algorithm further includes predicting disturbances in advance and generating feedforward compensation based on RBF neural network, specifically including the following steps: Obtain the decoupled real-time pendulum position deviation output by claim 2. and the deviation threshold of the preset swing arm position Compare; like If so, the feedforward compensation is zero; like Then, based on the RBF neural network, the disturbance is predicted in advance and the feedforward compensation is generated; Based on the RBF neural network, disturbances are predicted in advance and feedforward compensation is generated, including the acquisition of the traction wheel speed of the wire drawing machine. Wire feed speed Calculate the position deviation rate of the pendulum. Normalized input is given to the RBF neural network. , , , Real-time operating condition characteristic values ; Based on the online lightweight fitting of the time-varying law of perturbation using RBF neural network, the mapping relationship between traction wheel pulsation, coil diameter change, wire feed speed fluctuation, tension perturbation, and speed fluctuation is obtained. The predicted tension disturbance amplitude for the next control cycle is obtained based on the fitted time-varying pattern of the disturbance. Predicting speed fluctuations Predicting the direction of tension disturbance Predicting the direction of speed fluctuations ; Based on the next control cycle , , , Generate tension feedforward compensation amount Velocity feedforward compensation At the same time, based on the optimal parameters of the output speed PID and tension PID, the compensation amount is adjusted by amplitude adaptation. The optimal parameters for the speed PID, tension PID, and swing arm position PID of the allocated output are then decoupled and connected to the output. , , It forms an independent closed loop with the real dataset, calculates the basic adjustment amount and superimposes the feedforward compensation amount after amplitude adaptation correction, and finally converts it into the control signal of the actuator.

[0034] It should be noted that the introduction of RBF neural network feedforward compensation into the enameled wire drawing control solves the problems of delayed response of feedback control to measurable disturbances such as traction wheel pulsation and feed speed fluctuation, and the inability of conventional feedforward to adapt to time-varying working conditions. Based on real-time swing arm position deviation Deviation threshold from preset pendulum position By comparing the results, we can avoid invalid calculations for steady-state conditions, or perform multi-source disturbance prediction and advance compensation generation. Meanwhile, amplitude adaptation correction can ensure that feedback and feedforward work together to avoid overcompensation.

[0035] Among them, the weight correction rule and prediction error feedback mechanism of the RBF neural network are the same as those of the BP neural network. Pendulum position deviation rate The calculation formula is as follows: ; in, This represents the actual position of the pendulum after decoupling during the current cycle. The value of the previous period. To control the cycle; The formula for expressing the mapping relationship is as follows: ,and ; in, The weights from the hidden layer to the output layer. For the output layer bias of the RBF neural network, For the input vector, The parameters for the predicted next cycle of disturbance; Predicting the amplitude of tension disturbance Predicting speed fluctuations Predicting the direction of tension disturbance Predicting the direction of speed fluctuations The constraint rules are as follows: ; The calculation formula for generating the feedforward compensation is as follows: ; in, , Forward compensation coefficient, Direction factor; The formula for calculating amplitude adaptation correction is: ; in, , This is the baseline value for the PID proportional coefficient.

[0036] It should be noted that the pendulum position deviation rate The dynamic trend of the pendulum is plotted in real time in differential form, which is an early and sensitive indicator of tension and speed mismatch. The RBF network adopts Gaussian radial basis function, which has strong local approximation ability and can achieve millisecond-level online fitting of time-varying disturbances. It can adapt to continuous changes in roll diameter, mold and linear speed without offline training. Amplitude adaptation correction is adopted This allows the feedforward strength to adaptively decrease with the PID proportional coefficient. When the PID gain is high, the feedforward is automatically weakened to avoid overcompensation, and when the PID gain is low, the feedforward strength is maintained, thus always maintaining a harmonious combination of feedback and feedforward.

[0037] Example 4: A method for optimizing process parameters of enameled wire drawing based on PID algorithm, which also includes unified scheduling of multi-source data and adaptive input of working conditions. The specific steps are as follows: Construct a multi-source data pool, which contains three types of data: primary data, secondary data, and tertiary data. One type of data includes a given speed Given tension Given the position of the pendulum rod actual speed Actual tension Actual position of the swing arm ; The second type of data includes real-time volume diameter values. Mold usage time ; The three types of data include traction wheel speed. Wire feed speed ; Obtain the decoupled real-time pendulum position deviation output by claim 2. , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling ; Based on the current operating condition control mode, execute the data scheduling strategy; If only the basic control steps of claim 1 are executed, then one type of data is selected from the multi-source data pool and input into the neural network decoupler of claim 2; like ,and Then, two types of data are selected from the multi-source data pool, normalized together with the real dataset of claim 2, and then input into the BP neural network for PID parameter self-tuning; like Then, a third type of data is selected from the multi-source data pool, and combined with the real-time volume diameter value D and real-time operating condition characteristic values ​​collected in claim 3. And the pendulum position deviation rate in claim 4 After being normalized, the data are input into the RBF neural network for perturbation prediction and feedforward compensation generation.

[0038] It should be noted that by constructing a multi-source data pool and implementing data classification and hierarchical management, the data dependency boundaries of each intelligent module are clarified. Furthermore, the adaptive scheduling based on operating conditions and the allocation of data paths on demand enable data routing on demand, reducing invalid I / O reads and writes within a single control cycle, thereby shortening control cycle jitter. In addition, data classification and isolation prevent the propagation of sensor anomalies across modules, enhancing its robustness.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing process parameters of enameled wire drawing based on PID algorithm, characterized in that, Includes the following steps: S1. Collect the given speed of the wire drawing machine Given tension Given the position of the pendulum rod actual speed Actual tension Actual swing arm position This forms the basic dataset, which is then preprocessed to obtain a clean basic dataset. S2. Based on a clean base dataset, a decoupler is constructed using a three-layer feedforward neural network. The decoupler is then used to decouple multiple variables, and the actual speed after decoupling is obtained. Actual tension after decoupling Actual position of the pendulum after decoupling And calculate the deviation and the rate of change of deviation; S3. Based on decoupling , , The system calculates the basic adjustment amounts of speed, tension, and swing arm position using preset PID parameters and deviations, and then converts these basic adjustment amounts into control signals for the actuator.

2. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 1, characterized in that: The steps in S2 for decoupling multiple variables and calculating the deviation and the rate of change of deviation using a decoupler include: The preprocessed part in S1 , , , , , A total of six parameters are input to the neural network decoupler; Decoupling based on , , , , , The network weights are adjusted online with six parameters in a lightweight manner, and then dynamically fitted. , , A nonlinear coupling mathematical model among the three; The tension on the velocity and the disturbance of the pendulum position are calculated in reverse based on the nonlinear coupled mathematical model. The amount of disturbance caused by the velocity and position of the pendulum during tension. The velocity and tension disturbances experienced by the pendulum position ,based on , , Decoupling compensation correction is applied to the actual value to obtain , , ; use , , and , , Perform difference calculations, and then use differential operations to calculate the real-time tension deviation after decoupling. Real-time pendulum position deviation after decoupling , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling And constitute a real dataset; Actual speed after output decoupling Actual tension after decoupling Actual position of the pendulum after decoupling And real datasets.

3. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 2, characterized in that, It also includes online adaptive tuning of PID parameters based on a BP neural network, the specific steps of which are as follows: Obtain the decoupling tension deviation change rate output in claim 2. Rate of change of pendulum position deviation and respectively with engineering threshold , Compare; like ,or If the adaptive tuning is skipped, the current PID parameters are kept unchanged, and step S3 is executed directly. like ,and Then, the PID parameters are adaptively tuned online based on the BP neural network; The PID parameters are adaptively tuned online based on a BP neural network, including real-time values ​​of the wire drawing machine's roll diameter. Mold usage time , to use real datasets, , All data are normalized before being input into the BP neural network. First, a set of initial PID parameters is output using a BP neural network, which is then assigned to the velocity PID, tension PID, and swing arm position PID loops, respectively. Simultaneously, based on... , , Calculate the actual tension fluctuation variance Actual swing amplitude of the pendulum Actual adjustment overshoot and will , , As a practical performance evaluation indicator; The actual performance evaluation index and the preset tension fluctuation variance are combined. Preset swing amplitude of the pendulum Preset adjustment of overshoot Perform difference calculations to obtain the performance error value. Then As a back feedback signal for the BP network, it is backpropagated based on gradient descent to correct the weights of each layer of the network. The BP neural network synchronously outputs the optimal PID parameters for speed, tension, and swing arm position, which are adapted to the current wire drawing conditions. The preset PID parameters used in step S3 of claim 1 are updated to the optimal PID parameters for speed, tension, and pendulum position, and real-time operating condition feature values ​​are extracted. .

4. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 3, characterized in that, It also includes predicting disturbances in advance and generating feedforward compensation based on RBF neural networks, specifically including the following steps: Obtain the decoupled real-time pendulum position deviation output by claim 2. and the deviation threshold of the preset swing arm position Compare; like If so, the feedforward compensation is zero; like Then, based on the RBF neural network, the disturbance is predicted in advance and the feedforward compensation is generated; Based on the RBF neural network, disturbances are predicted in advance and feedforward compensation is generated, including the acquisition of the traction wheel speed of the wire drawing machine. Wire feed speed Calculate the position deviation rate of the pendulum. Normalized input is given to the RBF neural network. , , , Real-time operating condition characteristic values ; Based on the online lightweight fitting of the time-varying law of perturbation using RBF neural network, the mapping relationship between traction wheel pulsation, coil diameter change, wire feed speed fluctuation, tension perturbation, and speed fluctuation is obtained. The predicted tension disturbance amplitude for the next control cycle is obtained based on the fitted time-varying pattern of the disturbance. Predicting speed fluctuations Predicting the direction of tension disturbance Predicting the direction of speed fluctuations ; Based on the next control cycle , , , Generate tension feedforward compensation amount Velocity feedforward compensation Meanwhile, based on the optimal speed PID parameters and optimal tension PID parameters output in claim 3, the compensation amount is adjusted by amplitude adaptation. Assign the optimal parameters of the speed PID, tension PID, and swing arm position PID outputs in claim 3, and connect them to the decoupled outputs in claim 2. , , It forms an independent closed loop with the real dataset, calculates the basic adjustment amount and superimposes the feedforward compensation amount after amplitude adaptation correction, and finally converts it into the control signal of the actuator.

5. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 2, characterized in that: Dynamic fitting in claim 2 , , A nonlinear coupling mathematical model among the three; Let the connection weight matrix from the input layer to the hidden layer of the decoupler be... The bias vector of the hidden layer is The activation function of the hidden layer is The connection weight matrix from the hidden layer to the output layer is: The bias vector of the output layer is The input vector of the decoupler is The decoupler output is a coupling interference vector. ; The general formula for the nonlinear coupling mathematical model is as follows: ； Its expansion is: ; in, , , The weight matrix of the output layer corresponding to the hidden layer The three row vectors, , , For the output layer bias vector The three elements; The formula for subtracting the interference from the actual value is as follows: 。 6. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 5, characterized in that: In claim 5 , The weight adjustment is based on the decoupling loss function constraint, the function formula of which is as follows: ； in, , , This represents the true value without coupling under the enameled wire drawing process. The normalized coefficients are always adjusted along the loss function. Update the gradient in the opposite direction, i.e. towards Correction of the decreasing direction; The online lightweight correction formula for the connection weights and bias vectors is as follows: ， ； in, For the learning rate, and , , Let be the gradient, and let the gradient be subject to a cutoff constraint. Its calculation formula is as follows: ,and , , All are truncated according to the same rules as above.

7. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 3, characterized in that: The performance error value in claim 3 The calculation formula is as follows: ； in, This indicates that the actual performance evaluation indicators completely match the preset performance evaluation indicators; The formula for the loss function of a BP neural network is as follows: ; Among them, if the performance error value If so, the weights of the BP neural network are not modified; If performance error value Then, the weights of the BP neural network are slightly adjusted using the gradient descent method; like Then, the weights of the BP neural network are modified to a limited extent using the gradient descent method, and an alarm signal is triggered simultaneously.

8. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 3, characterized in that: The optimal speed PID parameter in claim 3 includes the optimal speed PID proportional coefficient. Optimal integral coefficient for speed PID Speed ​​PID optimal differential coefficients ; The optimal parameters of the tension PID include the optimal proportional coefficient of the tension PID. Optimal integral coefficient of tension PID Optimal differential coefficients of tension PID ; The optimal parameter for the PID controller at the swing arm position is the optimal proportional coefficient for the PID controller at the swing arm position. Optimal integral coefficients for PID control of lever position PID optimal differential coefficients for lever position ; Real-time operating condition characteristic values This represents the comprehensive characteristic quantity of the current wire drawing process, derived from the real-time value of the roll diameter. Mold usage time Tension deviation change rate The optimal proportional gain for PID control is extracted using the following expression: ； in, Maximum value of roll diameter process Rated service life of the mold The maximum value of the tension PID proportional coefficient is given, and all components are mapped to... , , ; The optimal process constraints for PID parameters are: 。 9. The method for optimizing enameled wire drawing process parameters based on PID algorithm according to claim 8, characterized in that: The real-time operating condition feature value in claim 8 of For dynamic adjustment, let the influence degree be . ,and ; This includes the influence of roll diameter as The influence of mold wear is The influence of tension fluctuation is The influence of PID regulation is Then, normalization is performed and mapped to... Its expression is as follows: ； in, For the real-time change rate of roll diameter, This represents the maximum value of the tension deviation change rate in the process. The weighting factor is the rate of change of roll diameter; Based on dynamic coefficient base value With influence Positive correlation, and calculate the basic value of the dynamic coefficient. Its expression is as follows: , and ; For the basic value of dynamic coefficient Normalization is performed to obtain the normalized coefficients. Its expression is as follows: ； And on Apply industrial hard constraints to obtain constraint coefficients Its expression is as follows: ； Then to Perform a first-order low-pass smoothing to obtain Dynamically adjusted coefficients Its expression is as follows: ； in, This represents the dynamic coefficient of the previous control cycle. For the control cycle of the wire drawing machine, As the smoothing coefficient, the final dynamic weighting coefficient satisfies , .

10. A method for optimizing enameled wire drawing process parameters based on a PID algorithm according to any one of claims 1-9, characterized in that, It also includes unified scheduling of multi-source data and adaptive input based on operating conditions. The specific steps are as follows: Construct a multi-source data pool, which contains three types of data: primary data, secondary data, and tertiary data. One type of data includes a given speed Given tension Given the position of the pendulum rod actual speed Actual tension Actual swing arm position ; The second type of data includes real-time volume diameter values. Mold usage time ; The three types of data include traction wheel speed. Wire feed speed ; Obtain the decoupled real-time pendulum position deviation output by claim 2. , tension deviation change rate after decoupling Rate of change of pendulum position deviation after decoupling ; Based on the current operating condition control mode, execute the data scheduling strategy; If only the basic control steps described in claim 1 are performed, then one type of data is selected from the multi-source data pool and input into the neural network decoupler described in claim 2; like ,and Then, two types of data are selected from the multi-source data pool, and after being normalized together with the real dataset described in claim 2, they are input into the BP neural network for PID parameter self-tuning. like Then, a third type of data is selected from the multi-source data pool, and combined with the real-time volume diameter value D and real-time operating condition characteristic values ​​collected in claim 3. And the pendulum position deviation rate in claim 4 After being normalized, the data are input into the RBF neural network for perturbation prediction and feedforward compensation generation.

Images