Multi-module cooperative control system and method for spandex spinning channel

Abstract

The invention provides a multi-module cooperative control system and method for a spandex spinning channel, and the system comprises a multi-mode sensor array which is used for measuring the spinning viscosity of an inlet of the spandex spinning channel in real time, and measuring a distributed temperature field in the spandex spinning channel in real time; the mixed prediction model is used for predicting the basic viscosity based on the distributed temperature field and calculating the compensation amount of the basic viscosity; the fault diagnosis module is used for carrying out space gradient analysis and fault positioning on the distributed temperature field and outputting a correction instruction; the three-level controller is used for acquiring an optimization control increment based on a discrete state space model of the hybrid prediction model, and performing multi-source disturbance grading processing based on the spinning viscosity change rate of the spandex spinning channel entrance and a self-adaptive grading threshold value to acquire a grading control quantity; and obtaining a final heating control quantity based on the optimization control increment, the hierarchical control quantity or the correction instruction. And the system has an accurate modeling capability, a real-time sensing function, perspectiveness and a self-adaptive control capability.

Description

Technical Field

[0001] This invention relates to the technical field of spandex spinning production, and in particular to a control system and method for a spandex spinning tunnel. Background Technology

[0002] The spandex spinning process involves many crucial pieces of equipment, with the spinning tunnel being one of them. In its operation, the polyurethane spinning solution enters the spinning tunnel through a spinneret during spandex spinning, while simultaneously, heated gas is introduced into the tunnel. This heated gas flows with the fine stream of the spinning solution towards the bottom of the tunnel. Because the gas in the spinning tunnel remains at a high temperature throughout, the solvent quickly evaporates from the polyurethane spinning solution stream and naturally moves towards the bottom of the tunnel. Precise heating and material delivery within the spinning tunnel are essential to ensure the spinning solution reaches optimal and uniform viscosity, thereby guaranteeing the stability of subsequent spinning and the superior performance of the final fiber. Therefore, the core objective of spinning tunnel control is to maintain a uniform and constant viscosity of the spinning solution at the tunnel outlet under various disturbances; however, this control objective is extremely challenging.

[0003] Traditional control methods, such as fixed-parameter PID control, when faced with a combination of disturbances including raw material fluctuations and localized yarn blockages, primarily rely on simple feedback regulation to attempt to maintain stable temperature and raw material viscosity within the spinning tunnel, thereby ensuring spinning quality. These traditional control methods suffer from the following problems:

[0004] First, the spinning tunnel is a typical distributed parameter system. Its internal temperature and viscosity fields are distributed along the length of the tunnel, with strong thermal and fluid conduction coupling between the upstream and downstream sections. Simultaneously, the flow of the spinning solution from the inlet to the outlet exhibits a significant time lag of several minutes, posing substantial challenges to traditional feedback control. Second, the system faces complex and diverse disturbances, including high-frequency inlet viscosity fluctuations caused by upstream raw material batch changes, and slow-changing localized blockages on the tunnel wall due to long-term coking or wear. Traditional single control strategies cannot effectively address these two drastically different disturbances simultaneously. Finally, during long-term operation, localized, minute blockages (i.e., "wire blockage") may occur within the tunnel. While seemingly insignificant, these localized, minute blockages can slow the flow rate and reduce heat exchange efficiency in the affected area. Unfortunately, because traditional sensors are limited to the inlet and outlet, it is difficult to effectively detect and accurately locate this highly concealed fault. Summary of the Invention

[0005] To address the technical problems of existing technologies, such as strong coupling, large time delay, slow response, difficulty in dealing with multi-source disturbances, and difficulty in effectively sensing local anomalies, this invention proposes a multi-module collaborative control system and method for spandex spinning tunnels. It is a new generation of intelligent control system solution for spinning tunnels with accurate modeling capabilities, real-time sensing functions, and forward-looking and adaptive control characteristics.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A multi-module collaborative control system for spandex spinning tunnels includes:

[0008] A multimodal sensor array, including an online viscometer and a fiber Bragg grating sensor array, wherein the online viscometer is used to measure the spinning viscosity at the entrance of the spandex spinning channel in real time, and the fiber Bragg grating sensor array is used to measure the distributed temperature field inside the spandex spinning channel in real time;

[0009] The hybrid prediction model includes a parallel physical mechanism model and a neural network model as a compensator. The physical mechanism model is used to predict the base viscosity based on the distributed temperature field, and the neural network model is used to calculate the compensation amount for the base viscosity based on the spinning viscosity at the inlet of the spandex spinning channel, the spatial gradient of the distributed temperature field, and the number of historical blockages.

[0010] The fault diagnosis module is used to perform spatial gradient analysis and fault location on the distributed temperature field based on the second-order central difference in space and output correction instructions.

[0011] The three-level controller includes an upper-level model predictive controller, a middle-level hierarchical adaptive feedback controller, and a lower-level PID controller. The upper-level model predictive controller is used to perform quadratic programming optimization based on the discrete state-space model of the hybrid predictive model to obtain the optimized control increment. The middle-level hierarchical adaptive feedback controller is used to perform multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold to obtain the hierarchical control quantity. The lower-level PID controller obtains the final heating control quantity based on the optimized control increment, the hierarchical control quantity, or the correction command.

[0012] Furthermore, the physical mechanism model is a set of partial differential equations coupling the Carreau-Yasuda constitutive equation and the heat transfer equation:

[0013] Simplified equation of mass conservation:

[0014] Momentum conservation equation:

[0015] Energy conservation equation:

[0016] Carreau-Yasuda constitutive equations:

[0017]

[0018]

[0019] in, For the velocity vector, For gradient operators, For the Laplace operator, The density of the spinning solution inside the spinning channel. For time, The pressure of the spinning solution inside the channel. For the deviatoric stress tensor, The vector of gravitational acceleration. For isobaric specific heat capacity, Temperature inside the tunnel. The thermal conductivity of the spinning solution, For external heating power density, Heat dissipation power density Shear viscosity, For shear rate, For infinite shear viscosity, Zero shear viscosity For characteristic time, For shape parameters, Mobility behavior index This is the temperature sensitivity coefficient. For reference temperature;

[0020] The neural network model employs a multilayer perceptron, a long short-term memory network, or a gated recurrent unit network.

[0021] Furthermore, when the hybrid prediction model is applied to the upper-level model predictive controller, the grid cells are divided and time-discretely processed using the finite volume method, and linearized at the current operating point using the Jacobian matrix to obtain the discrete state-space model of the hybrid prediction model. Based on the discrete state-space model of the hybrid prediction model, the optimization objective function and constraints are set, and the control increment is optimized by solving the sequential quadratic programming algorithm. :

[0022] The discrete state-space model of the hybrid prediction model is as follows:

[0023] The objective function to be optimized is:

[0024] ;

[0025] The constraints include:

[0026] System dynamic constraints:

[0027] ;

[0028] Output equation constraints: ;

[0029] Input hard constraints: ;

[0030] Input rate of change constraint: ;

[0031] Temperature state constraints: ;

[0032] Viscosity state constraints: ;

[0033] in, To optimize the objective function, To predict the time domain, To control the time domain, j is the time index. Predicted at time k The outlet viscosity value at time [time]. Set the export viscosity value. To output the tracking error weight matrix, To control the increment, To control the incremental weight moments, Predicted at time k Time-based control input, The optimal control input value is the one that represents the economically optimal control. For the control quantity weight matrix, The square of the weighted L2 norm, Let k be the system state vector predicted at time k+j. The control input at time k+j is predicted at time k. The perturbation input at time k+j is predicted at time k. , To control the upper and lower limits of the input, , To control the upper and lower limits of the increment, Let k be the temperature of the nth discrete unit predicted at time k. , These are the safe upper and lower limits for temperature. Let be the viscosity of the nth discrete unit predicted at time k. , These are the safe upper and lower limits for temperature.

[0034] Furthermore, the fault diagnosis module calculates an approximate temperature Laplace operator by performing a second-order central difference in space on the temperature data collected by the distributed fiber Bragg grating sensor. When the temperature Laplace operator If the value exceeds the preset safety threshold Th_grad, it is determined that a partial blockage has occurred, and the location of the blockage is determined. Generate setpoint correction commands based on the severity of wire blockage. , is a coefficient.

[0035] Furthermore, the middle-layer hierarchical adaptive feedback controller is used to obtain hierarchical control quantities by performing multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold, including:

[0036] The adaptive grading threshold, including the first grading threshold, is updated online using a reinforcement learning module based on a deep Q-network algorithm. and the first grade threshold ;

[0037] Based on the first classification threshold and the first grade threshold Multi-source disturbances are classified according to the rate of change of the spinning viscosity at the inlet of the spandex spinning channel, including low-frequency disturbances, high-frequency disturbances, and extreme disturbances.

[0038] The feedforward compensation amount is obtained by classifying and processing low-frequency disturbances, high-frequency disturbances, and extreme disturbances. Or trigger H∞ robust control to calculate robust control quantity .

[0039] Furthermore, the state space of the deep Q-network includes ,in, Indicates the rate of change of inlet viscosity. For export viscosity deviation, The outlet viscosity is the measured value at time t. Set the export viscosity value. Historical congestion count; action space is The reward function is: , , , These are the weighting coefficients. This is an indicator function; it is 1 if H∞ robust control is triggered, and 0 otherwise. Penalize export viscosity deviation, Penalty for the number of times the wire is blocked. Punishment mode switch.

[0040] Furthermore, the classification and processing based on low-frequency disturbances, high-frequency disturbances, and extreme disturbances includes:

[0041] like The disturbance is a low-frequency disturbance, and the optimized control increment is calculated only based on the upper-level model predictive controller. ;

[0042] like The disturbance is a high-frequency disturbance, and the feedforward compensation amount is calculated based on the linear relationship between the historical viscosity change rate and the feedforward compensation amount. ;

[0043] like The disturbance is an extreme disturbance, and the feedforward compensation amount is... =0, triggering H∞ robust control calculation of robust control quantity. .

[0044] Furthermore, the H∞ robust control method involves using a balanced truncation method to reduce the discrete state-space model of the hybrid prediction model to a low-order approximate model. Based on low-order approximation model Constructing a generalized plant model Preset performance threshold Based on a generalized plant model Solve the state feedback algebraic Riccati equations and output injection algebraic Riccati equations for the H∞ control. Solve the positive definite solution matrix X on the state feedback side and the positive definite solution matrix Y on the output injection side, respectively. Calculate the spectral radius of the product of the positive definite solution matrices X and Y. If satisfied The state-space parameters are calculated according to the standard formula of the H∞ controller, and the state-space model of the H∞ controller K(s) is obtained. The real-time error signal is then processed. The robust control input is obtained from the state-space model of input K(s). .

[0045] Furthermore, the underlying PID controller obtains the final control quantity based on the optimized control increment, the hierarchical control quantity, or the correction instruction, including:

[0046] For normal operating conditions with both low-frequency and high-frequency disturbances, the final control quantity is calculated using an incremental superposition method:

[0047] ;

[0048] For operating conditions with extreme disturbances, only robust control variables are used. Calculate the final control value:

[0049] ;

[0050] For fault conditions, correction commands are based on setpoint instructions. Change the temperature setpoint: The final control quantity is calculated using a positional algorithm:

[0051]

[0052] in, , , These are proportional gain, integral gain, and derivative gain, respectively. The temperature deviation at time t, Location of the plug The actual temperature at time t, The sampling period is This is the sum of temperature deviations from the initial time to time t.

[0053] Furthermore, a multi-module collaborative control method for spandex spinning tunnels includes the following steps:

[0054] S1: The viscosity at the inlet of the spandex spinning channel is measured in real time by an online viscometer, and the rate of change of the inlet viscosity data is calculated. The distributed temperature field inside the spandex spinning channel is measured in real time by a fiber optic grating sensor array.

[0055] S2: Viscosity data, distributed temperature field data, and the rate of change of viscosity data are sent to the three-layer controller; distributed temperature field data are sent to the hybrid prediction model and fault diagnosis module; data transmission is performed using a shared memory communication mechanism of a multi-producer-single-consumer lock-free circular queue;

[0056] S3: The hybrid prediction model updates and predicts the state based on the distributed temperature field data; the fault diagnosis module performs spatial gradient analysis and fault location based on the distributed temperature field data and outputs correction instructions;

[0057] S4: The upper-layer model predictive controller performs quadratic programming optimization based on the discrete state space model of the hybrid predictive model to obtain the optimized control increment; the middle-layer hierarchical adaptive feedback controller performs multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold to obtain the hierarchical control quantity; and the bottom-layer PID controller obtains the final heating control quantity based on the optimized control increment, the hierarchical control quantity, or the correction instruction.

[0058] The beneficial effects of this invention are as follows:

[0059] 1. By setting up a multi-modal sensor array including an online viscometer and a fiber optic grating sensor, real-time data acquisition of the tunnel is achieved, enabling real-time perception of the tunnel environment, and then using the sensed data as the basis for subsequent decision-making;

[0060] 2. By setting up a hybrid predictive model that includes both physical mechanism models and neural network models, the input data is processed through the hybrid predictive model to obtain decision instructions, thereby achieving forward-looking and adaptive control characteristics; by combining with other modules, fast time-scale nested logic is realized, demonstrating excellent robustness and outstanding control performance in the face of complex, highly coupled, and multi-scale disturbances.

[0061] 3. The mechanism-data hybrid model combines the universality of physical laws with the fitting ability of neural networks, enabling accurate prediction of viscosity evolution. Upper-level model predictive control (MPC) leverages this precise prediction to implement forward-looking optimization strategies, thereby fundamentally solving the hysteresis problem inherent in traditional control.

[0062] 4. Deep perception and rapid diagnosis: The distributed optical grating sensor array provides an unprecedented holographic view of the tunnel's internal state, making it possible to diagnose and locate spatial anomalies such as local wire blockages, with small positioning errors and short response times.

[0063] 5. Hierarchical control, economical and efficient: The three-level controller can take measures with different strengths according to the disturbance level, ensuring control effect while taking into account computational efficiency and energy consumption. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is an overall structural block diagram of the system in Embodiment 1 of the present invention.

[0066] Figure 2 This is a detailed structural block diagram of the system in Embodiment 1 of the present invention.

[0067] Figure 3 This is a flowchart of the method in Embodiment 2 of the present invention. Detailed Implementation

[0068] 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.

[0069] Example 1

[0070] A multi-module collaborative control system for spandex spinning tunnels, such as Figure 1 , 2 As shown, it includes:

[0071] The multimodal sensor array includes an online viscometer and a fiber Bragg grating sensor array. The online viscometer is used to measure the spinning viscosity at the entrance of the spandex spinning channel in real time, and the fiber Bragg grating sensor array is used to measure the distributed temperature field inside the spandex spinning channel in real time.

[0072] In this embodiment, an online viscometer 110 deployed at the inlet samples the inlet viscosity at a frequency of at least 100 Hz, while a fiber optic grating sensor array deployed in the tunnel simultaneously acquires the temperature field of the entire tunnel at a frequency of at least 200 Hz using wavelength division multiplexing technology. The application of this technology, such as in dynamic measurement and vibration monitoring, can significantly improve the efficiency and accuracy of data acquisition.

[0073] The hybrid prediction model includes a parallel physical mechanism model and a neural network model as a compensator. The physical mechanism model is used to predict the base viscosity based on the distributed temperature field, and the neural network model is used to calculate the compensation amount for the base viscosity based on the spinning viscosity at the inlet of the spandex spinning channel, the spatial gradient of the distributed temperature field, and the number of historical blockages.

[0074] In this embodiment of the application, the physical mechanism model includes a set of partial differential equations (PDEs) coupling the Carreau-Yasuda constitutive equations and the heat transfer equations. This set of PDEs includes:

[0075] mass conservation equation: Since the spinning process is usually an incompressible fluid, this equation is often simplified to Used for solving the velocity field; among which... The density of the spinning solution in the spinning channel is given by t, where t is time. For the velocity vector, It is a divergence operator.

[0076] Momentum conservation equation: ;in, Local acceleration represents the rate of change of flow velocity at a fixed location with time. Convection acceleration represents the rate of change of fluid velocity due to changes in spatial position. Pressure describes the pressure distribution of the spinning solution within the channel, driving fluid flow; The pressure gradient is the rate of change of pressure in space and is one of the driving forces of fluid flow. For the deviatoric stress tensor, it describes the viscous forces of a fluid. The spatial rate of change of viscous force. This is the vector of gravitational acceleration.

[0077] Energy conservation equation: in, The specific heat capacity at constant pressure of spandex spinning solution is approximately 2000~3000 J / (kg・K); Temperature describes the temperature distribution of the spinning solution within the channel and is a core factor affecting viscosity. Local temperature change rate describes the rate of temperature change over time at a fixed location. This represents the rate of change of convective temperature. The thermal conductivity of the spinning solution; is the Laplace operator for temperature, describing the second-order rate of change of temperature in space; This is a viscous dissipation term, describing the heat generated by the fluid due to viscous internal friction; For external heating power density, The power density is determined by heat dissipation loss. This equation explicitly specifies the control input P(x,t) (heating power) as the source term. In the direct injection system, x represents the axial position of the spinning channel.

[0078] Constitutive equations (Carreau-Yasuda non-Newtonian fluid model):

[0079]

[0080]

[0081] in, Shear rate describes the intensity of relative shear motion between fluid micro-elements; For shear viscosity, the "internal frictional resistance" of the spinning solution is a function of shear rate and temperature; Zero shear viscosity is the viscosity at which the shear rate approaches zero. This is the viscosity at which the shear rate approaches infinity. The characteristic time describes how fast a fluid transitions from zero shear to infinite shear. These are shape parameters that control the shape of the shear thinning / thickening curve; As a liquidity behavior index, <1 indicates shear thinning (the faster the flow rate, the lower the viscosity; spandex spinning solution belongs to this category). >1 indicates shear thickening; The temperature sensitivity coefficient describes how quickly viscosity changes with temperature; the larger the coefficient, the more significant the effect of temperature on viscosity. Reference temperature. This is the viscosity. The core expressions for the state variables temperature T and rheological shear rate are established. viscosity at state Mapping. Parameters , , , , , It needs to be determined by fitting rheological experimental data.

[0082] The stress tensor τ is related to the viscosity η, which is defined by the classical Carreau-Yasuda model. This set of partial differential equations (PDEs) was successfully solved numerically using the finite element method (FEM).

[0083] The neural network model employs a multilayer perceptron (MLP), a long short-term memory network (LSTM), or a gated recurrent unit (GRU) network. Taking the multilayer perceptron (MLP) as an example, the MLP has two hidden layers. The first hidden layer uses 64 neurons (ReLU activation), the second hidden layer uses 32 neurons (ReLU), and the output layer is 1-dimensional.

[0084] During training, an offline supervised learning approach was adopted. The training dataset was derived from historical normal operation data. The input was a feature vector containing historical information, specifically including inlet viscosity, viscosity prediction error, spatial gradient of the distributed temperature field, and historical blockage count. The output was the compensation amount for the base viscosity predicted by the physical mechanism model. Training loss is the compensation amount Error with the true model The mean square error (MSE) between them.

[0085]

[0086] in, The inlet viscosity is collected by a multimodal sensor array. For time lag step size, This is the actual viscosity. To predict the basic viscosity using a physical mechanism model, For the temperature spatial gradient, The number of times the wire was blocked in the past within a preset time period of time k (e.g., within 5 minutes).

[0087] During inference, the inputs are the inlet viscosity, the spatial gradient of the Bucher temperature field, and the number of historical blockages, and the output is the real-time predicted compensation amount for the base viscosity predicted by the physical mechanism model.

[0088] In this embodiment of the application, in order to apply the above-mentioned PDE model to the model predictive controller described later, spatial discretization is required. The hybrid prediction model is linearized at the current operating point using the Jacobian matrix to obtain the discrete-time spatial state-space model required for model predictive control.

[0089] Specifically, firstly, the finite volume method (FVM) is used to divide the tunnel into grid cells and discretize them in time, discretizing the continuous x domain into N control volumes, resulting in N=50 cells, each denoted by n.

[0090] Furthermore, for each element, the predicted fundamental viscosity value is calculated using the finite element method (FEM) based on the aforementioned system of differential equations (PDE). ;

[0091] Furthermore, based on the inlet viscosity, the spatial gradient of the Brownian temperature field, and the historical number of blockages, the neural network model is used to predict and obtain the compensation amount for the predicted base viscosity of each unit. .

[0092] Furthermore, based on the compensation amount and predicted basic viscosity value Calculate the final viscosity prediction:

[0093] .

[0094] Furthermore, define the system state vector: ,in, This represents the average temperature of the i-th unit at time k. This represents the predicted viscosity value of the i-th unit at time k; define the control input vector: , among which This represents the control input for the m-th independent heating zone, where M=5 is the number of independent heating zones; Measurable disturbance input is defined as follows: Define the Nth unit as the exit: At this point, a high-dimensional nonlinear difference equation can be obtained: ,in For process noise, This represents the nonlinear function vector corresponding to the hybrid prediction model.

[0095] Furthermore, when using it, at the current work point Jacobi linearization is performed to obtain a linear parameter time-varying (LPV) state-space model for prediction. :

[0096]

[0097] in, The state matrix, ,element Describe the effect of the variable in unit q (e.g., temperature) on the next variable in unit p (e.g., temperature); For the control matrix, ,element , describing the Zone power for the first The effect of the next variable (such as temperature) in the unit; Here is the perturbation matrix; ,element Describe the effect of inlet viscosity on each state; For the output matrix, Only the 2Nth element is 1 (corresponding to the outlet viscosity).

[0098] The fault diagnosis module is used to perform spatial gradient analysis and fault location on the distributed temperature field based on the second-order central difference in space, and output correction instructions.

[0099] In this embodiment, the fault diagnosis module performs a second-order central difference in space on the T(x,t) data collected by the distributed fiber Bragg grating sensor to calculate the approximate temperature Laplace operator (i.e., the divergence of the gradient), thereby reflecting sudden temperature rises / falls.

[0100]

[0101] Furthermore, when the temperature Laplace operator If the value exceeds the preset safety threshold Th_grad, it can be determined that a partial blockage has occurred, and the location of the blockage can be determined. Generate setpoint correction commands based on the severity of wire blockage. ,coefficient The value is 0.01, and the negative sign indicates a decrease in temperature. Compared to simple first-order gradient (temperature difference) detection, distributed fiber optic grating sensors, due to their high sensitivity, can more accurately detect sharp temperature peaks / valleys caused by local blockages.

[0102] The three-level controller includes an upper-level model predictive controller, a middle-level hierarchical adaptive feedback controller, and a lower-level PID controller. The upper-level model predictive controller is used to perform quadratic programming optimization based on the discrete state-space model of the hybrid predictive model to obtain the optimized control increment. The middle-level hierarchical adaptive feedback controller is used to perform multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold to obtain the hierarchical control quantity. The lower-level PID controller obtains the final heating control quantity based on the optimized control increment, the hierarchical control quantity, or the correction command.

[0103] In this embodiment of the application, the upper-level model predictive controller solves the following constrained quadratic programming (QP) problem at each sampling time k to obtain the optimized control increment.

[0104] The objective function is:

[0105]

[0106] in, To optimize the objective function, it needs to be minimized, comprehensively balancing control effectiveness and execution cost; To predict the time domain, To control the time domain, j is the time index. Predicted at time k The outlet viscosity value at time [time]. Set the export viscosity value. To output the tracking error weight matrix, increasing the weight can preferentially reduce viscosity bias. To control the increment, that is, the difference in control input between adjacent time steps, To control the incremental weighting torque, increasing it can make the control action smoother and reduce wear on the actuator. for Time-based control input, This is the economically optimal control input value (the energy-saving reference value under undisturbed conditions). The control input weight matrix drives the control input to approach... To save energy, This indicates taking the square of the weighted L2 norm.

[0107] The constraints include:

[0108] System dynamic constraints:

[0109]

[0110] Output equation constraints:

[0111] Input hard constraints - actuator physical limits:

[0112] Input rate of change constraint - protect heating element:

[0113] Temperature constraints - to prevent thermal degradation / curing of materials:

[0114] Viscosity state constraints - ensuring reliability:

[0115] in, Let k be the system state vector predicted at time k+j. The control input at time k+j is predicted at time k. The perturbation input at time k+j is predicted at time k. , To control the upper and lower limits of the input, , To control the upper and lower limits of the increment, The temperature of the nth discrete unit predicted at time k (obtained by solving the physical mechanism model). , These are the safe upper and lower limits for temperature. Let be the viscosity of the nth discrete unit predicted at time k. , These are the safe upper and lower limits for temperature.

[0116] In this embodiment, a sequential quadratic programming algorithm is used to solve for and output the final optimized control increment sequence. The control cycle is preferentially set to 5s-10s, and the first increment is taken as the final optimized control increment for the current operating point. .

[0117] In this embodiment, the middle-layer hierarchical adaptive feedback controller is used to obtain hierarchical control quantities by performing multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold.

[0118] Specifically, firstly, the adaptive grading threshold is updated online using a reinforcement learning module based on the Deep Q-Network (DQN) algorithm:

[0119] Define the state space ,in Indicates the rate of change of inlet viscosity. This refers to the viscosity deviation at the outlet, i.e. , The outlet viscosity is the measured value at time t. Set the outlet viscosity value; define the action space. , An adaptive grading threshold is used, and a discrete action space is employed; the reward function is set as follows. , =10、 =50、 =20 is the weighting coefficient. This is an indicator function; it is 1 if H∞ robust control is triggered, and 0 otherwise. Penalize export viscosity deviation (the larger the deviation, the smaller the reward). Penalty for each blocked wire (the more blocked wires, the smaller the reward). The penalty mode is switched (the more frequently it is switched, the smaller the reward).

[0120] Extract complete operating condition data from historical production logs, including state sequences. The adaptive grading threshold used for action recording at time t and the corresponding reward function The deep Q-network is trained using the Q-value; during online updates, the current state is collected in real time and input into the trained deep Q-network, selecting the value with the highest Q-value. The combined data serves as the current adaptive grading threshold. New samples are collected at preset fixed intervals (e.g., one hour) to update the deep Q-network.

[0121] Furthermore, based on the aforementioned adaptive classification threshold, the multi-source disturbances are classified according to the rate of change in the spinning viscosity at the inlet of the spandex spinning channel:

[0122] like The disturbance is a low-frequency disturbance, and the optimized control increment is calculated only based on the upper-level model predictive controller. Feedforward compensation amount =0;

[0123] like The disturbance is a high-frequency disturbance, and the feedforward compensation amount is calculated based on the linear relationship between the historical viscosity change rate and the feedforward compensation amount. The linear relationship between the historical viscosity change rate and the feedforward compensation obtained from the fitting is as follows:

[0124] ;

[0125] like The disturbance is an extreme disturbance, and the feedforward compensation amount is... =0, triggering H∞ robust control calculation of robust control quantity. .

[0126] The specific implementation method of H∞ robust control is as follows: The balanced truncation method is used to transform the linear parameter time-varying (LPV) state-space model. This reduces the order to a computationally feasible low-order approximation model. Based on low-order approximation model Constructing a generalized plant model : Where z is the controlled output, such as the weighted average of viscosity error. Here, d represents the noise input, and n represents the disturbance and sensor noise, respectively. For error signals, To control the input, The transitive relationship from w to z The transitive relationship from u to z The transitive relationship from w to e, Corresponding to the transitive relationship from u to e, the matrix elements are... The performance threshold is determined jointly with the performance weighting function (including error weighting, control quantity weighting, and disturbance weighting) and preset. Based on a generalized plant model Solve the state feedback algebraic Riccati equations and output injection algebraic Riccati equations for the H∞ control. Solve the positive definite solution matrix X on the state feedback side and the positive definite solution matrix Y on the output injection side, respectively. Calculate the spectral radius of the product of the positive definite solution matrices X and Y. If satisfied The state-space parameters are calculated according to the standard formula of the H∞ controller, and the state-space model of the H∞ controller K(s) is obtained. The real-time error signal is then processed. The robust control input is obtained from the state-space model of input K(s). .

[0127] In this embodiment of the application, the underlying PID controller obtains the final control quantity based on the optimized control increment, the hierarchical control quantity, or the correction instruction.

[0128] Specifically, for normal operating conditions with both low-frequency and high-frequency disturbances, the final control quantity is calculated using an incremental superposition method:

[0129] ;

[0130] For operating conditions with extreme disturbances, only robust control variables are used. Calculate the final control value:

[0131] ;

[0132] For fault conditions, correction commands are based on setpoint instructions. Change the temperature setpoint: The final control quantity is calculated using a positional algorithm:

[0133]

[0134] in, , , These are proportional gain, integral gain, and derivative gain, respectively. The temperature deviation at time t, Location of the plug The actual temperature at time t, The sampling period is This is the sum of temperature deviations from the initial time to time t.

[0135] Example 2

[0136] A multi-module collaborative control method for spandex spinning tunnels, such as... Figure 3 As shown, the steps include:

[0137] S1: The viscosity at the inlet of the spandex spinning channel is measured in real time by an online viscometer, and the rate of change of the inlet viscosity data is calculated. The distributed temperature field inside the spandex spinning channel is measured in real time by a fiber optic grating sensor array.

[0138] In step S1, the following sub-steps are also included in the specific implementation:

[0139] S111: Set the sampling frequency of the viscometer to be no less than 100Hz;

[0140] S112. Set the sampling frequency of the fiber optic grating sensor array to be no less than 200Hz, and synchronously use the temperature field data of the tunnel through wavelength division multiplexing technology.

[0141] In practical implementation, a torsional vibration type online viscometer is used, with a measurement range covering 1-1000 Pa•s, and also includes the following sub-steps:

[0142] S121. The inlet viscosity data is preprocessed by filtering using the Kalman wave algorithm;

[0143] S122. Calculate the rate of change of the inlet viscosity data after processing.

[0144] S123. Timestamp align the inlet viscosity data, the rate of change of inlet viscosity data, and the distributed temperature field data.

[0145] The setup of a distributed fiber Bragg grating sensor array includes the following sub-steps:

[0146] S131. No less than 50 fiber Bragg grating sensors are etched on a single-film fiber coated with polyimide by ultraviolet laser phase submersion mode.

[0147] S132. Set the spacing between the fiber Bragg grating sensors to be no greater than 10cm;

[0148] S133. By combining frequency-sweeping laser and high-speed photoelectric detection technology with wavelength division multiplexing to demodulate the signal wavelength, it is ensured that all temperature measurement data can be synchronously and accurately acquired at a frequency of at least 200Hz.

[0149] The distributed fiber Bragg grating sensor array is configured using the ultraviolet laser phase masking method, where at least 50 fiber Bragg grating sensors are etched onto a single-mode fiber coated with polyimide. The spacing between these sensors is no greater than 10 centimeters. The wavelength demodulation process employs advanced swept-frequency laser and high-speed photoelectric detection technology, combined with wavelength division multiplexing (WDM) principles, to ensure synchronous and accurate data acquisition from all temperature measurement points at a frequency of at least 200 Hz. This high-precision sensor array data acquisition system enhances overall sensing capabilities by integrating different types of sensors and employing a redundant design, and ensures high-precision synchronization of data acquisition through a synchronization system.

[0150] S2: Send viscosity data, distributed temperature field data, and the rate of change of viscosity data to the three-layer controller; send distributed temperature field data to the hybrid prediction model and fault diagnosis module.

[0151] In its specific implementation, S2 includes the following sub-steps:

[0152] S211. Employs a shared memory communication mechanism using a multi-producer-single-consumer (MPSC) lock-free ring buffer:

[0153] The edge computing node writes the pre-processed and timestamp-aligned data packets to the tail of the aforementioned circular column; the edge computing node is the producer.

[0154] The hybrid predictive model, the three-level controller, and the fault diagnosis module read their respective required data from the head of the queue at different rates, acting as multiple consumers.

[0155] In this step, to avoid data transmission and processing becoming a bottleneck, a shared memory communication mechanism using a multi-producer-single-consumer (MPSC) lock-free ring buffer is adopted. Edge computing nodes, acting as producers, write pre-processed and timestamped data packets to the tail of the ring buffer. The hybrid prediction model, the three-tier controller, and the fault diagnosis module, acting as multiple consumers, read their respective required data from the head of the queue at different rates. This lock-free queue design achieves synchronization through memory barriers and atomic operations, completely avoiding thread blocking and priority inversion problems that may arise from traditional locking mechanisms. This ensures that high-priority control and diagnostic data can be distributed instantly and reliably under millisecond-level latency requirements. This design guarantees that all subsequent modules can make decisions based on the same real-time data.

[0156] In the above, by setting online viscometers and fiber optic grating sensors at the tunnel positions, and based on the settings and outputs in S100 and its sub-steps, a large amount of real-time relevant data was acquired for the tunnel. This data was then used as a basis for subsequent judgment and processing factors. Meanwhile, in S2, by setting the corresponding shared memory communication mechanism, the acquired large amount of real-time data was transmitted to the prediction model, the three-level controller, and the fault diagnosis module. This ensured data consistency, prevented deviations, reduced latency, and maintained high synchronization, thus guaranteeing the rationality and accuracy of subsequent decisions.

[0157] S3: The hybrid prediction model updates and predicts the state based on the distributed temperature field data; the fault diagnosis module performs spatial gradient analysis and fault location based on the distributed temperature field data and outputs correction instructions.

[0158] Specifically, the hybrid prediction model performs state updates and predictions based on the state-space model:

[0159] ;

[0160] The fault diagnosis module calculates the approximate temperature Laplace operator by performing a second-order central difference in space on the T(x,t) data collected by the distributed fiber Bragg grating sensor. Generate setpoint correction commands based on the severity of wire blockage. .

[0161] S3: The upper-layer model predictive controller performs quadratic programming optimization based on the discrete state space model of the hybrid predictive model to obtain the optimized control increment; the middle-layer hierarchical adaptive feedback controller performs multi-source disturbance hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold to obtain the hierarchical control quantity; and the bottom-layer PID controller obtains the final heating control quantity based on the optimized control increment, the hierarchical control quantity, or the correction instruction.

[0162] Specifically, the upper-layer model predictive controller performs quadratic programming optimization based on the discrete state-space model of the hybrid predictive model to obtain the optimal control increment, including solving a quadratic programming (QP) problem at each sampling time k to obtain the optimal control increment. ;

[0163] The intermediate-layer hierarchical adaptive feedback controller obtains hierarchical control quantities by performing multi-source perturbation hierarchical processing based on the change rate of the spinning viscosity at the entrance of the spandex spinning channel and the adaptive hierarchical threshold. This includes updating the adaptive hierarchical threshold online through a reinforcement learning module based on a deep Q-network (DQN) algorithm. Based on the adaptive classification threshold, multi-source disturbances are classified according to the rate of change of spinning viscosity at the inlet of the spandex spinning channel. The classification includes low-frequency disturbances, high-frequency disturbances, and extreme disturbances. Feedforward compensation is obtained based on the classification of multi-source disturbances. Or trigger H∞ robust control to calculate robust control quantity ;

[0164] The underlying PID controller obtains the final control quantity based on the optimized control increment, the hierarchical control quantity, or the correction instruction. This includes calculating the final control quantity using an incremental superposition method for normal operating conditions with both low-frequency and high-frequency disturbances; calculating the final control quantity using only robust control quantity for extreme disturbance conditions; and calculating the final control quantity based on setpoint correction instructions for fault conditions. Change the temperature setpoint and calculate the final control quantity using a positional algorithm.

[0165] In practical implementation, the above scheme uses a nested design of time scales to collaboratively handle disturbances of different frequencies. The measured control delay data for each level are as follows: MPC (5 seconds) > Diagnostics (approximately 200 milliseconds) > Feedforward (<10 milliseconds) > PID (<1 millisecond).

[0166] Mid-level feedforward (millisecond-level response, handling high-frequency disturbances): The hierarchical adaptive feedback controller in the middle layer of the three-level controller acts as a feedforward module to handle inlet viscosity. This is the first line of defense against high-frequency, rapid disturbances. It directly monitors the inlet viscosity on a millisecond-scale timescale (response delay less than 10 milliseconds). rate of change Once a mutation is detected, it skips complex model calculations and quickly calculates a feedforward compensation increment. .

[0167] The upper-level model predictive controller (second-level optimization to handle gradually changing trends): The hybrid predictive model continuously receives the latest temperature field T(x,t) to update the state of its internal PDE solver. The upper-level model predictive controller of the three-level controller then updates the state based on the long-term prediction results of the hybrid predictive model on a second-level timescale (control cycle 5-10 seconds). Perform global, forward-looking optimization to calculate the optimal temperature setpoint increment. .

[0168] The underlying PID controller (executed in milliseconds, ultimately synthesized): These two control variables, one handling high-frequency and the other low-frequency disturbances, are incrementally and dynamically superimposed within the underlying PID controller. The final output control variable is in the following form: This incremental superposition design can effectively avoid the problem of integral saturation of the control quantity caused by the introduction of feedforward compensation.

[0169] In one embodiment, high-precision temperature and viscosity control under normal operating conditions is set up. Compared with existing single-loop PID controllers, the results are compared in the following table:

[0170]

[0171] Performance Analysis: In industrial control systems, traditional PID controllers, due to their inherent large time delays and coupling effects, often lead to significant periodic fluctuations in outlet viscosity control. In contrast, the upper-level MPC (Model Predictive Control) strategy adopted in this invention draws on the advanced nature of MPC in industrial control. Through its ability to predict future states based on mathematical models and dynamically adjust control inputs through rolling optimization, it effectively overcomes the time delay and coupling problems, significantly reducing the fluctuation range of outlet viscosity.

[0172] In another embodiment, adaptive control is used to address significant viscosity fluctuations caused by raw material switching during production.

[0173] At time t=0, the upstream feedstock switches to a new batch with 15% higher viscosity. A comparison is made with a system using a conventional H∞ controller, which, without model order reduction, has a computational delay of 1.2 seconds.

[0174] The system response process is as follows:

[0175] The online viscometer deployed at the inlet at t=1s immediately detects the inlet viscosity. It began to rise rapidly, and its rate of change It quickly exceeded the first threshold Th1.

[0176] At t=1.01s, the mid-level hierarchical adaptive feedforward controller of this invention is activated (response delay <10ms). It quickly calculates the positive feedforward compensation amount based on a linear relationship and superimposes it onto the output of the bottom-level PID controller.

[0177] From t=1.01s to t=300s, thanks to the feedforward effect, the heating amount can be advanced and precisely adjusted when the high-viscosity spinning solution flows through the channel.

[0178] The feedforward control quantity is calculated, and the middle layer controller calculates the feedforward increment and applies it directly to the corresponding heating zone.

[0179] MPC performs rolling optimization, while the upper-level MPC runs continuously with a 5-second cycle. In each cycle k, it solves the quadratic programming (QP) problem described in the invention based on the latest state estimate X(k) and future disturbance prediction d(k+j|k). The optimizer outputs an optimal control sequence U(k), U(k+1),..., but only sends the first control increment ΔU(k) ​​= U(k) - U(k-1) to the lower layer.

[0180] Control input synthesis; the final control instructions executed at the underlying level are determined by formulas. (Corresponding to the incremental superposition formula of PID in the invention content) Synthesis. This combines the speed of feedforward with the optimization capability of MPC.

[0181] Ultimately, during the entire switching process, the outlet viscosity under the control of this invention only experienced a slight fluctuation of ±1.5%, and then quickly returned to stability. In contrast, under conventional control, due to its significant 1.2-second delay, its compensation effect was severely lagging, resulting in an overshoot of up to +9% in the outlet viscosity.

[0182] In another embodiment, a coordinated approach is taken to address the combined failure of localized wire blockage and raw material switching.

[0183] During the raw material switching process ( =18%), and a slight, localized blockage occurred near the fifth heating zone of the tunnel (location x=10m).

[0184] The collaborative response process of this invention is as follows:

[0185] t=0.5s (feedforward response), the middle-layer feedforward controller first responds to the sudden change in inlet viscosity, injecting... =3.2% compensation to suppress global disturbances caused by raw material switching.

[0186] At t=0.8s (diagnostic response), almost simultaneously, the fault diagnosis module, through analysis of the temperature field, detected that the temperature gradient at x=10m reached 6.1℃ / 10cm, exceeding the alarm threshold. Immediately, the system sent a high-priority command to the underlying PID controller of the 5th heating zone, instantly lowering the setpoint by 3℃ to handle the abnormal situation.

[0187] t=0.8s (diagnostic response), the fault diagnosis module calculates the approximate second-order differential value of the temperature at location x=10m: ‖∇²T‖=610℃ / m²;

[0188] This value far exceeds the safety threshold Th_grad=300℃ / m², indicating a wire blockage at this location. A high-priority instruction is then generated.

[0189] At t=300s (optimized response), after the emergency disturbance is stabilized by the feedforward and diagnostic modules, the upper-level MPC performs a global optimization based on the updated hybrid model that includes information on wire blockage faults, and calculates a new optimal temperature setpoint sequence that can adapt to the current operating conditions in the long term.

[0190] At t=300s (optimized response), the upper-level MPC detects that the state constraint T(10m)>T_max is about to be violated. In the following optimization cycle, the MPC's QP problem will prioritize preventing the temperature from exceeding the limit at this point as a higher-priority optimization objective. By adjusting the weight matrix or using a constraint-first strategy, the MPC recalculates the global optimization setpoint sequence, which will automatically lower the temperature setpoint near the fault region while coordinating the work of other heating zones to minimize the impact on the outlet viscosity Y(k) while satisfying the state constraints.

[0191] Results: When faced with extremely complex compound faults, the collaborative control system of this invention employs a compound control principle. Through its nested time-scale logic of "millisecond-level feedforward + sub-second-level diagnosis + second-level optimization," it successfully suppressed the outlet viscosity fluctuation within ±2.1%, while the wire breakage rate only slightly increased from the normal 1.2% to 1.5%. In comparative experiments, the traditional system almost immediately collapsed when faced with such compound faults, with the wire breakage rate soaring to over 45%.

[0192] In yet another embodiment, the self-learning effect of DQN threshold is verified.

[0193] Training phase: Input 100,000 sets of historical operating condition data from the factory over the past year. The DQN module is trained offline to obtain a set of optimal threshold solutions Th1 and Th2 (i.e., the first and second thresholds mentioned above) for various operating conditions.

[0194] Online Phase and Comparison: In a production shift with a particularly high frequency of raw material switching (5 times / shift), the effects of using a fixed threshold and the adaptive threshold of the present invention (DQN) were compared.

[0195] Fixed threshold: Setting Th2 to a fixed value resulted in three unnecessary switches to H∞ mode during this shift due to frequent viscosity fluctuations, increasing computational overhead and potentially causing unnecessary control actions.

[0196] In this invention, the DQN module recognizes that it is currently in a high-frequency disturbance state and automatically increases the H∞ switching threshold Th_2 by 12% to increase the tolerance to moderate disturbances and avoid overreaction.

[0197] Results: After adopting this invention, the probability of incorrect switching due to improper thresholds decreased by 100% in this shift, without increasing the defect rate due to insufficient disturbance suppression. This proves that the DQN module can intelligently and adaptively optimize the system's cooperative strategy.

[0198] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-module coordinated control system for spandex spinning shafts, characterized in that, Comprise: A multi-modal sensor array, comprising an online viscometer for measuring the spinline viscosity at the entrance of the spinline duct in real time, and a fiber Bragg grating sensor array for measuring the distributed temperature field within the spinline duct in real time; A hybrid prediction model, comprising a parallel physical mechanism model for predicting the base viscosity based on the distributed temperature field, and a neural network model as a compensator for calculating the compensation amount of the base viscosity based on the spinline viscosity at the entrance of the spinline duct, the spatial gradient of the distributed temperature field, and the historical number of blockages; A fault diagnosis module for spatial gradient analysis and fault location of the distributed temperature field based on the second-order central difference on space, and outputting a correction instruction; A three-level controller, comprising an upper model predictive controller for quadratic programming optimization to obtain an optimized control increment based on the discrete state space model of the hybrid prediction model, a middle hierarchical adaptive feedback controller for multi-source disturbance hierarchical processing to obtain a hierarchical control amount based on the spinline viscosity variation rate at the entrance of the spinline duct and an adaptive hierarchical threshold, and a bottom PID controller for obtaining a final heating control amount based on the optimized control increment, the hierarchical control amount, or the correction instruction.

2. The multi-module coordinated control system for spandex spinning nozzle according to claim 1, wherein, The physical mechanism model is a partial differential equation set coupling the Carreau-Yasuda constitutive equation and the heat conduction equation: Mass-conserved simplified equation: Momentum conservation equation: Energy conservation equation: Carreau-Yasuda constitutive equation: wherein, is the flow velocity vector, is the gradient operator, is the Laplacian operator, is the dope density in the spin- duct, is time, is the pressure of the dope in the duct, is the deviatoric stress tensor, is the gravitational acceleration vector, is the constant pressure specific heat capacity, is the temperature in the duct, is the thermal conductivity of the dope, is the external heating power density, is the heat loss power density, is the shear viscosity, is the shear rate, is the infinite shear viscosity, is the zero shear viscosity, is the characteristic time, is the shape parameter, is the flow behavior index, is the temperature sensitivity coefficient, is the reference temperature; The neural network model adopts a multi-layer perception, a long short-term memory network, or a gated recurrent unit network.

3. The multi-module coordinated control system for spandex spinning nozzle according to claim 1, wherein, The hybrid prediction model, when applied to the model predictive controller of the upper layer, performs grid element division and time discretization by the finite volume method, linearization at a current working point by a Jacobian matrix, obtains a discrete state space model of the hybrid prediction model, sets an optimization objective function and a constraint condition based on the discrete state space model of the hybrid prediction model, and solves an optimization control increment by a sequential quadratic programming algorithm : The discrete state space model of the hybrid prediction model is: The optimization objective function is: ； The constraint conditions include: System dynamics constraints: ； Output equation constraints: ; Input hard constraints: ; Input rate of change constraint: ; Temperature state constraints: ; Viscosity state constraint: ; wherein, is the optimization objective function, is the prediction horizon, is the control horizon, j is the time index, is the predicted outlet viscosity value at time k, is the outlet viscosity setpoint, is the output tracking error weight matrix, is the control increment, is the control increment weight matrix, is the predicted control input at time k+j, is the economically optimal control input value, is the control weight matrix, is the square of the weighted L2 norm, is the predicted system state vector at time k+j given the prediction at time k, is the predicted control input at time k+j given the prediction at time k, is the predicted disturbance input at time k+j given the prediction at time k, are the control input upper and lower limits, are the control increment upper and lower limits, is the predicted temperature of the nth discrete element at time k, are the temperature safety upper and lower limits, is the predicted viscosity of the nth discrete element at time k, are the temperature safety upper and lower limits.​​​​ 4. The multi-module coordinated control system for spandex spinning ducts according to any one of claims 1-3, wherein, The fault diagnosis module calculates an approximate temperature Laplacian by performing a spatial second-order central difference on temperature data collected by the distributed fiber grating sensor: When the temperature Laplacian is greater than a preset safety threshold Th_grad, it is determined that a local blockage occurs, and the location of the blocked fiber is located According to the severity of the blocked fiber, a set point correction instruction is generated , is a coefficient.

5. The multi-module coordinated control system for spandex spinning ducts of claims 1-3, wherein, The middle hierarchical adaptive feedback controller is used for multi-source disturbance hierarchical processing to obtain a hierarchical control amount based on the spinline viscosity variation rate at the entrance of the spinline duct and an adaptive hierarchical threshold, including: The adaptive hierarchical threshold, including the first hierarchical threshold, is updated online by a reinforcement learning module based on a deep Q network algorithm and the first hierarchical threshold ; based on the first hierarchical threshold and the first hierarchical threshold , according to the rate of change of the spinning viscosity at the entrance of the spandex spinning nozzle, multi-source disturbances are classified into low-frequency disturbances, high-frequency disturbances and extreme disturbances; According to the low-frequency disturbance, the high-frequency disturbance and the extreme disturbance, the feedforward compensation quantity is obtained Or trigger H∞ robust control to calculate the robust control quantity .

6. The multi-module coordinated control system for spandex spinning ducts of claim 5, wherein, The state space of the deep Q network comprises wherein, represents an inlet viscosity change rate, is an outlet viscosity deviation, is an outlet viscosity measurement value at t, is an outlet viscosity set value, is a historical number of blockages; the action space is ; the reward function is: , , , is a weight coefficient, is an indicator function, 1 if H∞ robust control is triggered, otherwise 0, penalizes the outlet viscosity deviation, penalizes the number of blockages, penalizes the mode switching.

7. The multi-module coordinated control system for spandex spinning nozzle according to claim 5 or 6, characterized in that, Hierarchical processing according to low-frequency disturbance, high-frequency disturbance, and extreme disturbance includes: If , the disturbance is a low frequency disturbance, only the model predictive controller of the upper layer is used to calculate the optimization control increment ; If , the disturbance is a high-frequency disturbance, the feedforward compensation amount is calculated based on a linear relationship between a historical viscosity change rate and the feedforward compensation amount ; If , the disturbance is an extreme disturbance, the feedforward compensation amount = 0, triggering H∞ robust control to calculate a robust control amount .

8. The multi-module coordinated control system for spandex spinning tunnels of claim 7, wherein, The H∞ robust control method is: using the balanced truncation method to reduce the order of the discrete state space model of the mixed prediction model to a low-order approximate model , constructing a generalized plant model based on the low-order approximate model , presetting a performance threshold , solving the state feedback side algebraic Riccati equation and the output injection side algebraic Riccati equation of the H∞ control based on the generalized plant model , solving the positive definite solution matrix X of the state feedback side and the positive definite solution matrix Y of the output injection side, and calculating the spectral radius of the product of the positive definite solution matrix X and the positive definite solution matrix Y , if the following condition is met , calculating the state space parameters according to the standard formula of the H∞ controller to obtain the state space model of the H∞ controller K (s), inputting the real-time error signal into the state space model of K (s) to obtain the robust control amount .​ 9. The multi-module coordinated control system for spandex spinning tunnels of claim 8, wherein, The bottom PID controller obtains a final control amount based on the optimized control increment, the hierarchical control amount, or the correction instruction, including: For low-frequency disturbance and high-frequency disturbance in normal working conditions, the final control amount is calculated using an incremental superposition method: ； For extreme disturbance conditions, only robust control quantities are used Calculate final control quantities: ； For the fault condition, the setpoint correction command is based on Change the temperature setpoint: Calculate the final control variable using the positional algorithm. wherein, , , Kp, Ki and Kd are proportional gain, integral gain and derivative gain, respectively, is a temperature deviation at time t, is a plugged wire position is an actual temperature at time t, is a sampling period, is a temperature deviation cumulative sum from an initial time to t.

10. A multi-module cooperative control method for spandex spinning nozzle, using the multi-module cooperative control system for spandex spinning nozzle according to any one of claims 1 to 9, characterized in that, The steps are: S1: Real-time measurement of the spinline viscosity at the entrance of the spinline duct by the online viscometer, and calculation of the viscosity data variation rate, and real-time measurement of the distributed temperature field within the spinline duct by the fiber Bragg grating sensor array; S2: Send the viscosity data, distributed temperature field data, and viscosity data variation rate to the three-level controller; send the distributed temperature field data to the hybrid prediction model and the fault diagnosis module; use a shared memory communication mechanism of a multi-producer-single-consumer lock-free ring queue for data transmission; S3: The hybrid prediction model performs state update and prediction based on the distributed temperature field data; The fault diagnosis module performs spatial gradient analysis and fault positioning based on the distributed temperature field data and outputs a correction instruction; S4: the model predictive controller of the upper layer performs quadratic programming optimization based on the discrete state space model of the hybrid prediction model to obtain an optimized control increment, the hierarchical adaptive feedback controller of the middle layer performs hierarchical processing of multiple source disturbances based on the change rate of the spandex spinning nozzle inlet spinning viscosity and an adaptive hierarchical threshold to obtain a hierarchical control amount, and the PID controller of the bottom layer obtains a final heating control amount based on the optimized control increment, the hierarchical control amount, or the correction instruction.

Images