High-precision coating die thickness adjustment method based on intelligent feedback control

Abstract

The present invention discloses a high-precision coating die head coating thickness adjustment method based on intelligent feedback control, comprising the following steps: S1, constructing a data set and transmitting it to a controller module; S2, calculating an error and inputting the error into a fuzzy controller for preliminary control, generating a preliminary control instruction; S3, inputting the generated preliminary control instruction into a PID controller; S4, applying a model prediction controller to predict the future trend of coating thickness changes, and sending an adjustment signal in advance; S5, dynamically adjusting the parameters of the controller module based on a dual adaptive neural network control mechanism; S6, continuously optimizing the control strategy through a "trial-feedback" mechanism; S7, transmitting the control signal generated by the controller module to an actuator module; S8, through a data processing and self-learning module, recording and storing all data in the coating process, and continuously optimizing the control strategy. The present invention combines fuzzy control, PID and dual adaptive neural networks to achieve precise control of coating thickness.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent control of industrial coating processes, and in particular to a high-precision coating die coating thickness adjustment method based on intelligent feedback control. Background Art

[0002] In modern industrial coating processes, precise control of coating thickness is one of the key factors to ensure product quality and performance. With the continuous expansion of the application of coating technology in many high-precision fields, such as coating manufacturing of electronic components, surface treatment of optical materials, production of films and medical films, etc., the requirements for the control accuracy of coating thickness are increasing. However, in the actual coating process, the coating thickness is affected by many factors, such as coating liquid flow, coating die position, coating speed, and external environmental conditions (such as temperature and humidity). Fluctuations in these factors may lead to uneven coating thickness, thus affecting the quality and performance of the product.

[0003] Most existing coating processes use traditional feedback control methods, such as PID control and fuzzy control. These control strategies can, to a certain extent, cope with errors that occur during the coating process and adjust the coating thickness. However, due to the complexity of the coating process, traditional single control strategies are difficult to adapt to complex working conditions under different environments. For example, although PID control can make fine corrections to real-time errors, its ability to handle the nonlinear dynamic response of the system and external environmental disturbances is limited. Fuzzy control is good at handling complex and nonlinear systems, but the design of its rule set relies more on experience, and it is difficult to make adaptive adjustments in real time when facing a changing production environment. In addition, the control systems based on PID or fuzzy control in the prior art often lack the ability to predict future changes in working conditions when making adjustments, and can only make passive adjustments based on current errors, resulting in the system reacting more slowly to sudden fluctuations.

[0004] As industrial production develops towards automation and intelligence, the limitations of existing technologies are becoming increasingly apparent. The instability of coating thickness in the coating process has become a major challenge, especially when the environment changes greatly or the production conditions are complex. Traditional control methods are difficult to adapt to dynamic environmental changes, especially when environmental parameters such as temperature and humidity fluctuate greatly, the system is difficult to adjust quickly. Existing control systems also lack long-term adaptive capabilities. Although some control systems are capable of parameter adjustment, most still rely on fixed control strategies and fail to achieve self-learning and adjustment based on real-time and historical data.

[0005] In addition, the coating control system in the prior art shows obvious lack of control accuracy when facing complex nonlinear dynamic changes, especially when the thickness error is large or the environmental conditions change drastically, the system is difficult to maintain the stability of the coating thickness. At the same time, the existing control system is usually unable to predict the changes in future working conditions in real time and lacks effective prediction of future trends. Therefore, when dealing with sudden fluctuations or changes in working conditions over a long period of time, adjustments are not made in time, resulting in the coating thickness exceeding the allowable range.

[0006] In traditional control systems, although fuzzy control and PID control can correct errors, in complex, nonlinear and dynamic coating processes, a single control algorithm cannot solve the balance problem between global regulation and local fine adjustment, especially when facing complex coating environments, it lacks flexible adjustment capabilities and intelligent feedback mechanisms. In addition, existing technologies have not fully utilized adaptive neural networks and self-learning systems to dynamically adjust control parameters, resulting in the system's lack of adaptive capabilities to different working conditions. Most current coating control systems fail to fully utilize the large amount of data generated during the coating process, and fail to combine historical data for deep learning and optimization.

[0007] Traditional coating control systems also lack a long-term feedback mechanism. Existing systems are mainly adjusted based on immediate feedback data, and fail to effectively use past historical feedback information to optimize the current control strategy, especially when dealing with long-term operation, the system performance is difficult to maintain the best state. Due to the lack of a long-term feedback adjustment mechanism, the system will gradually reduce the coating thickness control accuracy during long-term operation due to factors such as changes in working conditions, equipment aging or environmental fluctuations, affecting the quality and consistency of the final product.

[0008] Therefore, how to provide a high-precision coating die coating thickness adjustment method based on intelligent feedback control is a problem that technicians in this field need to solve urgently. Summary of the invention

[0009] One purpose of the present invention is to propose a high-precision coating die coating thickness adjustment method based on intelligent feedback control. The present invention combines a variety of intelligent control algorithms such as fuzzy control, PID control and dual adaptive neural network, and realizes precise control of coating thickness during the coating process through parallel learning of real-time feedback and historical data. Through model predictive control, future working condition changes are predicted in advance, and the control strategy is continuously optimized based on the "trial-feedback" mechanism to effectively cope with nonlinear and dynamic changes under complex working conditions. This method not only improves the response speed and control accuracy of the system, but also has long-term self-adaptation and self-learning capabilities, significantly reduces manual intervention, and ensures the long-term stability and automation level of the coating process.

[0010] A high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to an embodiment of the present invention comprises the following steps:

[0011] S1, using a laser thickness measurement sensor to collect coating thickness data during the coating process and environmental parameters collected by an environmental sensor in real time, constructing a data set and transmitting it to a controller module;

[0012] S2, calculating the error between the currently measured coating thickness and the preset target thickness, and inputting the error into the fuzzy controller for preliminary control. The fuzzy controller roughly adjusts the large-scale error according to the preset fuzzy rules and generates preliminary control instructions;

[0013] S3, inputting the generated preliminary control instruction into a PID controller, wherein the PID controller further fine-tunes the small-scale error based on proportional, integral and differential parameters;

[0014] S4. Use the model prediction controller to predict the future trend of coating thickness changes and send out adjustment signals in advance;

[0015] S5, based on the dual adaptive neural network control mechanism, the parameters of the controller module are dynamically adjusted through parallel learning of historical data and real-time data, including the parameters of the PID controller, the fuzzy control rules and the parameters of the model predictive control;

[0016] S6. Utilize feedback data during system operation to continuously optimize control strategies through a “trial-feedback” mechanism;

[0017] S7, transmitting the control signal generated by the controller module to the actuator module, wherein the actuator module includes a servo motor and a flow control valve, wherein the servo motor adjusts the position of the coating die head in real time according to the control signal, and the flow control valve adjusts the flow rate of the coating liquid;

[0018] S8. Through the data processing and self-learning module, all thickness data, error data and control signals in the coating process are recorded and stored.

[0019] Optionally, the S2 specifically includes:

[0020] S21, real-time acquisition of the currently measured coating thickness T actual (t) and the preset target coating thickness T target , and calculate the current error e(t):

[0021] e(t)=T target -T actual (t);

[0022] S22, normalize the obtained error e(t) to obtain the normalized error enorm (t);

[0023] S23. According to the actual process requirements, a fuzzy rule set is predefined, and the normalized error e is calculated based on the “if-then” rule structure. norm (t) performing preliminary control, wherein the fuzzy rule set is optimized and adjusted according to production conditions;

[0024] S24, normalized error e norm (t) Perform fuzzy processing and define the membership function:

[0025]

[0026] Among them, μ complex (e norm (t)) represents the membership degree after error fuzzification, exp represents the exponential function, c represents the center point of the membership function, α, β and ζ represent the shape adjustment parameters of the membership function, σ represents the standard deviation of the normalized error, η represents the amplitude of the oscillation term, ω represents the frequency of the oscillation term, and λ represents the adjustment coefficient;

[0027] S25, fuzzy reasoning generates a rough control signal based on the output of the membership function and the fuzzy rule set:

[0028]

[0029] Among them, u fuzzy (t) represents the rough control signal after fuzzy reasoning, z i represents the output value of the fuzzy rule set, ν represents the amplitude of the sinusoidal adjustment factor, ω1 represents the frequency of the sinusoidal adjustment factor, κ represents the weight factor of the squared error term, ζ1 represents the sensitivity of the control membership function output, and Ω represents the definition domain of the fuzzy membership function;

[0030] S26, defuzzifying the rough control signal generated by the fuzzy reasoning to generate a preliminary control instruction.

[0031] Optionally, the S3 specifically includes:

[0032] S31, input the preliminary control instruction generated by the fuzzy controller to the PID controller, the PID controller simultaneously obtains the currently measured coating thickness and the preset target coating thickness, and calculates the small range error e PID (t), as the input error signal of the PID controller;

[0033] S32, PID controller based on real-time error e PID (t), the derivative of the error and the integral of the error, the proportional gain K p (t), integral gain K i(t) and differential gain K d (t) Perform adaptive adjustment:

[0034]

[0035] Among them, K p0 Indicates the initial proportional gain value, K i0 Indicates the initial integral gain value, K d0 represents the initial differential gain value, α p , α i and α d represents the first-order adaptive adjustment coefficient, β p , β i and β d represents the second-order adaptive adjustment coefficient, γ p , γ i and γ d represents the coefficient of the adjusted error derivative, δ d represents the adjustment factor related to error accumulation, Indicates the rate of change of error, represents the acceleration of error change, Indicates the cumulative amount of error;

[0036] S33, introduce nonlinear dynamic compensation term u in PID controller NL (t), enhance the correction capability of nonlinear and complex dynamic errors:

[0037]

[0038] Among them, γ1, γ2 and γ3 represent nonlinear compensation coefficients, ω2 represents angular frequency, δ1 and δ2 represent adjustment factors of compensation terms;

[0039] S34, the adaptively adjusted PID control signal u PID (t) and the nonlinear dynamic compensation term u NL (t) Synthesize the final control signal u final (t):

[0040]

[0041] Among them, γ4, γ5, γ6, δ3 and ∈ represent adjustment coefficients;

[0042] S35, control signal u final (t) Perform amplitude limiting processing and eliminate high-frequency noise, filter the control signal after amplitude limiting processing, and obtain the smoothed control signal u filtered (t);

[0043] S36, the smoothed control signal u filtered(t) Output to the actuator module to accurately adjust the position of the coating die head and the flow rate of the coating liquid in real time, thereby achieving further fine correction of small-range errors.

[0044] Optionally, the S4 specifically includes:

[0045] S41, based on the multidimensional data set formed by the coating thickness, coating die position, coating liquid flow rate and ambient temperature and humidity collected in real time by the sensors deployed in the system, the model predictive controller establishes a dynamic mathematical model to describe the state changes and influencing factors of the coating process;

[0046] S42, the model predictive controller uses the current system status and historical data to predict the coating thickness change trend in multiple time steps in the future, performs multi-time domain prediction on the thickness in multiple time periods in the future, gradually estimates the coating thickness at each time point, and predicts the short-term thickness change trend and the long-term thickness change trend:

[0047]

[0048] Among them, T pred (t+k) represents the predicted coating thickness at the future time t+k, A k and B k represents the prediction matrix, x(t) represents the current system state vector, u(t) represents the current control signal, γ k represents the adjustment coefficient, δ k and η k represents the adjustment factor used to control the prediction accuracy, T env (t) represents the ambient temperature, Indicates the accumulated value of the control signal, Indicates the historical cumulative value of actual coating thickness;

[0049] S43, the model prediction controller introduces a dynamic correction mechanism for prediction errors. After each prediction is completed, the system automatically calculates the error between the future predicted thickness and the target thickness, corrects the prediction parameters in the model according to the obtained error size, and performs error tolerance analysis to define the allowable error range according to different process requirements;

[0050] S44. Based on the prediction of the future coating thickness change trend, the model predictive controller automatically generates a control adjustment signal:

[0051]

[0052] Among them, u adj (t) represents the adjustment signal generated at the current moment, μ1, μ2 and μ3 represent the adjustment coefficients, and e pred (t+k) represents the prediction error at the future time t+k, P head(t) represents the current position of the coating die head, ν1 and ν2 represent dynamic adjustment factors, T actual (t) represents the actual measured coating thickness;

[0053] S45. When the adjustment signal is generated and sent, the model predictive controller monitors the thickness change during the coating process in real time, and corrects the adjustment signal according to the actual situation. It adopts a multi-level feedback mechanism to compare the gap between the actual change of coating thickness and the predicted value in real time, and uses the feedback signal to fine-tune the adjustment signal to adapt to the fluctuation of the current working conditions.

[0054] Optionally, the S5 specifically includes:

[0055] S51, constructing a dual adaptive neural network structure, wherein the dual adaptive neural network structure includes a real-time adaptive network and a historical adaptive network, wherein the real-time adaptive network is used to process real-time data acquired during the current coating process, and the historical adaptive network performs deep learning based on long-term accumulated data, and the real-time adaptive network and the historical adaptive network learn in parallel;

[0056] S52, real-time adaptive network continuously collects the error between coating thickness and target thickness real (t), dynamically adjust the parameters of the PID controller, the weights of the fuzzy control rules, and the prediction matrix of the model predictive controller;

[0057] S53, historical adaptive network analyzes coating performance and control strategies under different environments by deep learning of coating data accumulated over a long period of time, and identifies the long-term control mode of the system under different materials and environmental conditions by looking back on historical data;

[0058] S54, based on the outputs of the real-time adaptive network and the historical adaptive network, nonlinear feedback control is performed through the weight fusion mechanism, and the weights of the real-time adaptive network and the historical adaptive network are automatically adjusted according to the current working conditions and error fluctuations:

[0059]

[0060] Among them, W real (t) represents the dynamic weight of the real-time network, ξ, γ w and λ w represents the dynamic weight of the real-time network, f env (t) represents the influence function of environmental factors on the system state, exp represents the exponential function, Indicates the accumulated value of the control signal;

[0061] S55, based on the optimization results of the dual adaptive neural network output, the system proportional gain K of the PID controller p(t), integral gain K i (t) and differential gain K d (t) Dynamically adjust and adaptively adjust the fuzzy control rules:

[0062]

[0063] Among them, K p0 represents the initial proportional gain, β p ,θ p and p Represents the adaptive adjustment coefficient, W real (t) represents the weight of the real-time network, f hist (t) represents the impact function of historical data on the current system, Indicates the rate of change of real-time coating thickness;

[0064] S56, the control matrix of model predictive control is dynamically optimized through a dual adaptive network. The real-time network is responsible for quickly adjusting the control matrix based on the current error, and the historical network optimizes the control matrix according to the long-term trend:

[0065]

[0066] Among them, A(t) represents the optimized control matrix, A0 represents the initial control matrix, λ real and κ a Represents the adaptive adjustment coefficient.

[0067] Optionally, the S6 specifically includes:

[0068] S61. Based on the analysis of the current feedback data, a nonlinear tentative adjustment signal is generated to test the system's response to a small adjustment of the control parameters:

[0069]

[0070] Among them, u probe (t) represents the generated tentative adjustment signal, u0 represents the original control signal, α u and β u represents the adjustment coefficient, e real (t) represents the current error, P head (t) represents the current position of the coating die head, T actual (t represents the actual measured coating thickness, F flow (t) represents the current coating liquid flow rate, Indicates the accumulated value of the control signal;

[0071] S62. After the tentative adjustment signal is applied, the system immediately starts to monitor the response of various feedback data, using a nonlinear feedback mechanism to introduce the high-order derivatives of the error into the feedback process:

[0072]

[0073] Among them, e feedback (t) represents the corrected feedback error, γ e and e represents the regulating factor;

[0074] S63, by establishing a correlation model between feedback data and control signals, identifying which control parameters have the greatest impact on the current coating process, and obtaining feedback analysis results, wherein the feedback analysis results are processed by a weighted regression model to automatically adjust the contribution weight of each feedback signal to the controller;

[0075] S64. Based on the feedback analysis results, the system optimizes the weight of the trial adjustment signal through an adaptive mechanism:

[0076]

[0077] Among them, W opt (t) represents the optimized feedback weight, σ represents the activation function, f env (t) represents the influence function of environmental factors on the system state, u probe (t) represents a tentative adjustment signal;

[0078] S65. The system records the feedback data during the adjustment process to form an adaptive learning mechanism. In long-term operation, the system gradually reduces the frequency of trial adjustments and tends to the optimal control parameter combination.

[0079] The beneficial effects of the present invention are:

[0080] First, the present invention solves the problem of poor adaptability of the system to different working conditions and difficulty in simultaneously performing global adjustment and local fine control during the coating process by combining multiple intelligent control algorithms such as fuzzy control, PID control and model predictive control (MPC). Fuzzy control handles large-scale errors, PID control is responsible for fine adjustment in a small range, and MPC proactively sends adjustment signals by predicting the future coating thickness change trend in advance, ensuring that the system can adjust in advance when facing complex environmental changes, avoiding the loss of accuracy caused by delayed response. This multi-level, multi-algorithm combination improves the response speed and control accuracy of the system.

[0081] Secondly, the present invention makes full use of the parallel learning of real-time data and historical data by introducing a dual adaptive neural network mechanism. The real-time adaptive network can dynamically adjust the control parameters according to the current working conditions to ensure that the system can respond to process changes in a timely manner; the historical adaptive network analyzes the long-term accumulated data through deep learning, optimizes the control strategy, and enables the system to have the ability of self-learning and self-optimization during long-term operation. This mechanism effectively solves the problem of the lack of adaptive and self-learning capabilities of the control system in the prior art, allowing the system to maintain stable and efficient control under different environmental and material conditions.

[0082] In addition, through the "probe-feedback" mechanism, the present invention achieves continuous optimization of the control strategy. After the tentative adjustment signal is generated, the system monitors the adjustment results and further optimizes the control strategy based on the feedback data to ensure that the adjustment results meet the production requirements. This mechanism greatly enhances the adaptive ability of the system, enabling the system to respond flexibly under different production conditions, reducing the need for manual intervention and improving the level of automation. The dynamic monitoring of feedback data and the nonlinear feedback mechanism improve the perception and response capabilities of errors, especially when dealing with complex nonlinear and dynamically changing coating environments, the system exhibits excellent robustness and stability.

[0083] By real-time monitoring of multi-dimensional data such as coating thickness, die position, environmental parameters, etc. during the coating process, the present invention can accurately capture process changes and make dynamic adjustments, so that the coating process can maintain high precision and high stability during long-term operation. Compared with traditional control methods, the present invention can not only cope with sudden error changes, but also continuously optimize the control strategy through the accumulation of historical data and self-learning algorithms, reducing the risk of decreased control accuracy during long-term operation. In addition, the introduction of the model predictive controller enables the system to predict future operating condition changes in advance and actively adjust the control strategy, greatly improving the system's foresight and control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0085] Figure 1 A flow chart of a high-precision coating die head coating thickness adjustment method based on intelligent feedback control proposed by the present invention;

[0086] Figure 2 This is a schematic diagram of the control signal generation process of the model predictive controller of the high-precision coating die coating thickness adjustment method based on intelligent feedback control proposed in the present invention. DETAILED DESCRIPTION

[0087] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0088] refer to Figure 1 and Figure 2 A high-precision coating die head coating thickness adjustment method based on intelligent feedback control comprises the following steps:

[0089] S1, using a laser thickness measurement sensor to collect coating thickness data during the coating process and environmental parameters collected by an environmental sensor in real time, constructing a data set and transmitting it to a controller module;

[0090] S2, calculating the error between the currently measured coating thickness and the preset target thickness, and inputting the error into the fuzzy controller for preliminary control. The fuzzy controller roughly adjusts the large-scale error according to the preset fuzzy rules and generates preliminary control instructions;

[0091] S3, inputting the generated preliminary control instruction into a PID controller, wherein the PID controller further fine-tunes the small-scale error based on proportional, integral and differential parameters;

[0092] S4. Use the model prediction controller to predict the future trend of coating thickness changes and send out adjustment signals in advance;

[0093] S5, based on the dual adaptive neural network control mechanism, the parameters of the controller module are dynamically adjusted through parallel learning of historical data and real-time data, including the parameters of the PID controller, the fuzzy control rules and the parameters of the model predictive control;

[0094] S6. Utilize feedback data during system operation to continuously optimize control strategies through a “trial-feedback” mechanism;

[0095] S7, transmitting the control signal generated by the controller module to the actuator module, wherein the actuator module includes a servo motor and a flow control valve, wherein the servo motor adjusts the position of the coating die head in real time according to the control signal, and the flow control valve adjusts the flow rate of the coating liquid;

[0096] S8. Through the data processing and self-learning module, all thickness data, error data and control signals in the coating process are recorded and stored.

[0097] In this implementation, S2 specifically includes:

[0098] S21, real-time acquisition of the currently measured coating thickness T actual (t) and the preset target coating thickness T target , and calculate the current error e(t):

[0099] e(t)=T target -T actual (t);

[0100] S22, normalize the obtained error e(t) to obtain the normalized error e norm (t);

[0101] S23. According to the actual process requirements, a fuzzy rule set is predefined, and the normalized error e is calculated based on the “if-then” rule structure. norm (t) performing preliminary control, wherein the fuzzy rule set is optimized and adjusted according to production conditions;

[0102] S24, normalized error e norm (t) Perform fuzzy processing and define the membership function:

[0103]

[0104] Among them, μ complex (e norm (t)) represents the membership degree after error fuzzification, exp represents the exponential function, c represents the center point of the membership function, α, β and ζ represent the shape adjustment parameters of the membership function, σ represents the standard deviation of the normalized error, η represents the amplitude of the oscillation term, ω represents the frequency of the oscillation term, and λ represents the adjustment coefficient;

[0105] S25, fuzzy reasoning generates a rough control signal based on the output of the membership function and the fuzzy rule set:

[0106]

[0107] Among them, u fuzzy (t) represents the rough control signal after fuzzy reasoning, z i represents the output value of the fuzzy rule set, ν represents the amplitude of the sinusoidal adjustment factor, ω1 represents the frequency of the sinusoidal adjustment factor, κ represents the weight factor of the squared error term, ζ1 represents the sensitivity of the control membership function output, and Ω represents the definition domain of the fuzzy membership function;

[0108] S26, defuzzifying the rough control signal generated by the fuzzy reasoning to generate a preliminary control instruction.

[0109] In this implementation, S3 specifically includes:

[0110] S31, input the preliminary control instruction generated by the fuzzy controller to the PID controller, the PID controller simultaneously obtains the currently measured coating thickness and the preset target coating thickness, and calculates the small range error e PID (t), as the input error signal of the PID controller;

[0111] S32, PID controller based on real-time error e PID (t), the derivative of the error and the integral of the error, the proportional gain K p (t), integral gain K i (t) and differential gain K d (t) Perform adaptive adjustment:

[0112]

[0113] Among them, K p0 Indicates the initial proportional gain value, K i0 Indicates the initial integral gain value, K d0 represents the initial differential gain value, α p , α i and α d represents the first-order adaptive adjustment coefficient, β p , β i and β d represents the second-order adaptive adjustment coefficient, γ p , γ i and γ d represents the coefficient of the adjusted error derivative, δ d represents the adjustment factor related to error accumulation, Indicates the rate of change of error, represents the acceleration of error change, Indicates the cumulative amount of error;

[0114] S33, introduce nonlinear dynamic compensation term u in PID controller NL (t), enhance the correction capability of nonlinear and complex dynamic errors:

[0115]

[0116] Among them, γ1, γ2 and γ3 represent nonlinear compensation coefficients, ω2 represents angular frequency, δ1 and δ2 represent adjustment factors of compensation terms;

[0117] S34, the adaptively adjusted PID control signal u PID (t) and the nonlinear dynamic compensation term u NL (t) Synthesize the final control signal u final (t):

[0118]

[0119] Among them, γ4, γ5, γ6, δ3 and ∈ represent adjustment coefficients;

[0120] S35, control signal u final(t) Perform amplitude limiting processing and eliminate high-frequency noise, and filter the control signal after amplitude limiting processing to obtain a smoothed control signal u filtered (t);

[0121] S36, smoothing the control signal u filtered (t) Output to the actuator module to accurately adjust the position of the coating die head and the flow rate of the coating liquid in real time, thereby achieving further fine correction of small-range errors.

[0122] In this implementation manner, the S4 specifically includes:

[0123] S41, based on the multidimensional data set formed by the coating thickness, coating die position, coating liquid flow rate and ambient temperature and humidity collected in real time by the sensors deployed in the system, the model predictive controller establishes a dynamic mathematical model to describe the state changes and influencing factors of the coating process;

[0124] S42, the model predictive controller uses the current system status and historical data to predict the coating thickness change trend in multiple time steps in the future, performs multi-time domain prediction on the thickness in multiple time periods in the future, gradually estimates the coating thickness at each time point, and predicts the short-term thickness change trend and the long-term thickness change trend:

[0125]

[0126] Among them, T pred (t+k) represents the predicted coating thickness at the future time t+k, A k and B k represents the prediction matrix, x(t) represents the current system state vector, u(t) represents the current control signal, γ k represents the adjustment coefficient, δ k and η k represents the adjustment factor used to control the prediction accuracy, T env (t) represents the ambient temperature, Indicates the accumulated value of the control signal, Indicates the historical cumulative value of actual coating thickness;

[0127] S43, the model prediction controller introduces a dynamic correction mechanism for prediction errors. After each prediction is completed, the system automatically calculates the error between the future predicted thickness and the target thickness, corrects the prediction parameters in the model according to the obtained error size, and performs error tolerance analysis to define the allowable error range according to different process requirements;

[0128] S44. Based on the prediction of the future coating thickness change trend, the model predictive controller automatically generates a control adjustment signal:

[0129]

[0130] Among them, u adj (t) represents the adjustment signal generated at the current moment, μ1, μ2 and μ3 represent the adjustment coefficients, and e pred (t+k) represents the prediction error at the future time t+k, P head (t) represents the current position of the coating die, ν1 and ν2 represent dynamic adjustment factors, T actual (t) represents the actual measured coating thickness;

[0131] S45. When the adjustment signal is generated and sent, the model predictive controller monitors the thickness change during the coating process in real time, and corrects the adjustment signal according to the actual situation. It adopts a multi-level feedback mechanism to compare the gap between the actual change of coating thickness and the predicted value in real time, and uses the feedback signal to fine-tune the adjustment signal to adapt to the fluctuation of the current working conditions.

[0132] In this implementation manner, S5 specifically includes:

[0133] S51, constructing a dual adaptive neural network structure, wherein the dual adaptive neural network structure includes a real-time adaptive network and a historical adaptive network, wherein the real-time adaptive network is used to process real-time data acquired during the current coating process, and the historical adaptive network performs deep learning based on long-term accumulated data, and the real-time adaptive network and the historical adaptive network learn in parallel;

[0134] S52, real-time adaptive network continuously collects the error between coating thickness and target thickness real (t), dynamically adjust the parameters of the PID controller, the weights of the fuzzy control rules, and the prediction matrix of the model predictive controller;

[0135] S53, historical adaptive network analyzes coating performance and control strategies under different environments by deep learning of coating data accumulated over a long period of time, and identifies the long-term control mode of the system under different materials and environmental conditions by looking back on historical data;

[0136] S54, based on the outputs of the real-time adaptive network and the historical adaptive network, nonlinear feedback control is performed through the weight fusion mechanism, and the weights of the real-time adaptive network and the historical adaptive network are automatically adjusted according to the current working conditions and error fluctuations:

[0137]

[0138] Among them, W real (t) represents the dynamic weight of the real-time network, ξ, γ w and λ w represents the dynamic weight of the real-time network, f env(t) represents the influence function of environmental factors on the system state, exp represents the exponential function, Indicates the accumulated value of the control signal;

[0139] S55, based on the optimization results of the dual adaptive neural network output, the system proportional gain K of the PID controller p (t), integral gain K i (t) and differential gain K d (t) Dynamically adjust and adaptively adjust the fuzzy control rules:

[0140]

[0141] Among them, K p0 represents the initial proportional gain, β p ,θ p and p Represents the adaptive adjustment coefficient, W real (t) represents the weight of the real-time network, f hist (t) represents the impact function of historical data on the current system, Indicates the rate of change of real-time coating thickness;

[0142] S56, the control matrix of model predictive control is dynamically optimized through a dual adaptive network. The real-time network is responsible for quickly adjusting the control matrix based on the current error, and the historical network optimizes the control matrix according to the long-term trend:

[0143]

[0144] Among them, A(t) represents the optimized control matrix, A0 represents the initial control matrix, λ real and κ a Represents the adaptive adjustment coefficient.

[0145] In this implementation manner, S6 specifically includes:

[0146] S61. Based on the analysis of the current feedback data, a nonlinear tentative adjustment signal is generated to test the system's response to a small adjustment of the control parameters:

[0147]

[0148] Among them, u probe (t) represents the generated tentative adjustment signal, u0 represents the original control signal, α u and β u represents the adjustment coefficient, e real (t) represents the current error, P head (t) represents the current position of the coating die head, T actual(t) represents the actual measured coating thickness, F flow (t) represents the current coating liquid flow rate, Indicates the accumulated value of the control signal;

[0149] S62. After the tentative adjustment signal is applied, the system immediately starts to monitor the response of various feedback data, using a nonlinear feedback mechanism to introduce the high-order derivatives of the error into the feedback process:

[0150]

[0151] Among them, e feedback (t) represents the corrected feedback error, γ e and e represents the regulating factor;

[0152] S63, by establishing a correlation model between feedback data and control signals, identifying which control parameters have the greatest impact on the current coating process, and obtaining feedback analysis results, wherein the feedback analysis results are processed by a weighted regression model to automatically adjust the contribution weight of each feedback signal to the controller;

[0153] S64. Based on the feedback analysis results, the system optimizes the weight of the trial adjustment signal through an adaptive mechanism:

[0154]

[0155] Among them, W opt (t) represents the optimized feedback weight, σ represents the activation function, f env (t) represents the influence function of environmental factors on the system state, u probe (t) represents a tentative adjustment signal;

[0156] S65. The system records the feedback data during the adjustment process to form an adaptive learning mechanism. In long-term operation, the system gradually reduces the frequency of trial adjustments and tends to the optimal control parameter combination.

[0157] Embodiment 1:

[0158] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the production line of a precision optical device manufacturing company. In the production process of the company, precise control of coating thickness is the key to ensuring the performance of optical components, but due to complex factors such as coating liquid flow, die position, ambient temperature and humidity, it is difficult for traditional coating control systems to ensure the uniformity and stability of coating thickness, especially when production conditions change frequently, error accumulation leads to a high product defect rate.

[0159] On the production line of this enterprise, the coating process is used to apply an anti-reflective film on the surface of optical glass. Since optical devices have extremely high requirements for coating thickness uniformity (generally requiring a thickness error of less than 5 nanometers), and the temperature, humidity and flow rate in the production environment fluctuate greatly, the existing PID control system cannot meet this demanding requirement. Although the traditional PID control system can handle small-range errors, it is slow to respond when dealing with complex nonlinear dynamic changes in the coating process, especially when the ambient temperature changes cause the viscosity of the coating liquid to fluctuate. The system's feedback speed is not fast enough, often resulting in large-scale product scrapping.

[0160] In order to solve this problem, the company decided to adopt the high-precision coating die coating thickness adjustment method based on intelligent feedback control proposed in this invention. In practical applications, the entire coating system introduces a hybrid control scheme of fuzzy control, PID control, model predictive control (MPC) and dual adaptive neural network to ensure that the system can quickly respond to changes in working conditions and continuously optimize the control strategy. The system is equipped with laser thickness measurement sensors and environmental sensors to collect coating thickness and environmental parameters such as temperature and humidity in real time. Through these sensors, the system establishes a dynamic data set and inputs it into the controller module.

[0161] During the production process, the system first uses a fuzzy controller to roughly adjust the error between the coating thickness and the target value. The laser sensor collects coating thickness data 1,000 times per second. The system calculates the error in real time based on this data, normalizes it, and inputs it into the fuzzy controller for preliminary adjustment. The fuzzy controller generates a rough control signal based on a preset set of rules to eliminate a wide range of errors. This rough adjustment signal is then transmitted to the PID controller for further fine correction. The PID controller makes real-time adjustments to the flow rate of the coating liquid and the position of the die head based on the real-time error, error change rate, and error integral. Through an adaptive algorithm, the system dynamically adjusts the PID parameters according to the current error to ensure that the fine control error is always maintained at the nanometer level.

[0162] During the production process, the system also uses model predictive control (MPC) to predict future changes in operating conditions in advance. Based on the current ambient temperature, coating liquid flow rate, and historical data on coating thickness, the system predicts the coating thickness change trend in the next few seconds in real time, and sends out adjustment signals in advance to prevent uneven coating thickness due to changes in operating conditions. By introducing MPC, the system can still maintain the accuracy of coating thickness when the temperature fluctuates by more than 3 degrees Celsius, avoiding the error accumulation problem caused by delayed response in traditional control systems.

[0163] In the application, the system also introduces dual adaptive neural networks. The real-time adaptive network of the system processes the thickness, temperature, flow rate and other data obtained in real time during the coating process, and dynamically adjusts the control parameters; the historical adaptive network continuously optimizes the fuzzy control rules and PID control parameters through long-term learning of control strategies under different environmental conditions. In actual production, the dual adaptive neural network effectively improves the system's ability to adapt to environmental changes.

[0164] In order to verify the beneficial effects of the present invention, the company compared the performance of the traditional PID control system and the intelligent feedback control system of the present invention under the same production conditions during a 6-month production process. The test environment was a constant temperature of 25°C and a humidity of 45%, and an anti-reflection film for optical glass was produced. The comparison results are shown in the following table:

[0165] Table 1 Performance comparison between coating system based on intelligent feedback control and traditional PID control system

[0166]

[0167]

[0168] The above table analyzes the performance of the traditional PID control system and the intelligent feedback control system of the present invention in the coating process in multiple dimensions. It can be clearly seen from the data that the intelligent feedback control system of the present invention has significant advantages in all key performance indicators.

[0169] First, from the perspective of the average error and maximum error of the coating thickness, the average error of the traditional PID control system is 12.25 nanometers and the maximum error is 21.68 nanometers. The high error value indicates that it is difficult for the system to accurately control the coating thickness, especially when encountering environmental fluctuations, the error accumulation problem is more serious. The intelligent feedback control system of the present invention combines fuzzy control, PID control and adaptive neural network, which greatly reduces the average error of the coating thickness to 3.85 nanometers and the maximum error to only 4.15 nanometers, significantly improving the accuracy of coating thickness control.

[0170] In terms of system response time, the traditional PID control system is relatively slow, and the system response time takes about 35 seconds on average, which brings challenges to working conditions with large changes in coating liquid flow or temperature, making it difficult for the system to respond and adjust in time. The intelligent feedback control system of the present invention shows a fast response capability, and the system can complete the adjustment within 8 seconds, ensuring real-time control of coating thickness and avoiding excessive error accumulation.

[0171] The production defect rate is also a key indicator. The defect rate of the traditional PID control system is 2.72%, while the intelligent control system of the present invention significantly reduces the defect rate to 0.48%. This shows that the present invention can more stably control the coating quality during the production process, reduce product scrapping and rework, and directly improve production efficiency and product qualification rate.

[0172] From the perspective of increasing production, the traditional PID system failed to bring about a significant increase in production, while the intelligent feedback control system of the present invention achieved a production increase of about 5.15%, mainly due to the system's ability to reduce errors, reduce manual intervention, and improve overall production efficiency.

[0173] In terms of the number of manual interventions, the traditional PID control system requires about 15 manual adjustments in 6 months of operation, mainly when the system faces environmental changes or equipment status fluctuations. In contrast, the system of the present invention, thanks to its adaptive learning mechanism, does not require manual intervention in the same operating time, and all adjustments are automatically completed by the system, further improving the level of automation.

[0174] In addition, the present invention also shows significant advantages in the response speed of temperature changes and flow fluctuations. The traditional PID system has a temperature change response time of 32 seconds and a flow fluctuation response time of 28 seconds, while the system of the present invention only takes 12 seconds and 9 seconds to adjust respectively. The fast response capability greatly enhances the robustness of the system, ensuring that the coating thickness can be maintained stably even under complex working conditions.

[0175] Finally, the intelligent feedback control system of the present invention also showed excellent performance in terms of system stability and production cycle improvement. In terms of monthly trouble-free operation time, the system of the present invention is 5 days higher than the traditional PID system, which means that the system of the present invention is more stable and reduces downtime caused by failures or adjustments. In addition, the improvement of the production cycle reached 6.48%, which further improved production efficiency and saved the production cost of the enterprise.

[0176] In summary, the present invention fully demonstrates its superior performance under complex working conditions in the optical device production application scenario. By combining a variety of intelligent control algorithms and adaptive mechanisms, the system achieves high-precision coating thickness control, and has the advantages of rapid response, high automation and long-term stability when facing complex environmental changes. Production data proves that the present invention not only significantly improves product quality and reduces the defective rate, but also improves the system's adaptability and production efficiency through self-learning mechanisms and intelligent feedback mechanisms.

[0177] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A high-precision coating die head coating thickness adjustment method based on intelligent feedback control, characterized in that: The steps include: S1, using a laser thickness measurement sensor to collect coating thickness data during the coating process and environmental parameters collected by an environmental sensor in real time, constructing a data set and transmitting it to a controller module; S2, calculating the error between the currently measured coating thickness and the preset target thickness, and inputting the error into the fuzzy controller for preliminary control. The fuzzy controller roughly adjusts the large-scale error according to the preset fuzzy rules and generates preliminary control instructions; S3, inputting the generated preliminary control instruction into a PID controller, wherein the PID controller further fine-tunes the small-scale error based on proportional, integral and differential parameters; S4. Use the model prediction controller to predict the future trend of coating thickness changes and send out adjustment signals in advance; S5, based on the dual adaptive neural network control mechanism, the parameters of the controller module are dynamically adjusted through parallel learning of historical data and real-time data, including the parameters of the PID controller, the fuzzy control rules and the parameters of the model predictive control; S6. Utilize feedback data during system operation to continuously optimize control strategies through a "trial-feedback" mechanism; S7, transmitting the control signal generated by the controller module to the actuator module, wherein the actuator module includes a servo motor and a flow control valve, wherein the servo motor adjusts the position of the coating die head in real time according to the control signal, and the flow control valve adjusts the flow rate of the coating liquid; S8. Through the data processing and self-learning module, all thickness data, error data and control signals in the coating process are recorded and stored.

2. The high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to claim 1 is characterized in that: The S2 specifically includes: S21, real-time acquisition of the currently measured coating thickness T actual (t) and the preset target coating thickness T target , and calculate the current error e(t): e(t)=T target -T actual (t); S22, normalize the obtained error e(t) to obtain the normalized error e norm (t); S23. According to the actual process requirements, a fuzzy rule set is predefined, and the normalized error e is calculated based on the "if-then" rule structure. norm (t) performing preliminary control, wherein the fuzzy rule set is optimized and adjusted according to production conditions; S24, normalized error e norm (t) Perform fuzzy processing and define the membership function: Among them, μ complex (e norm (t)) represents the membership degree after error fuzzification, exp represents the exponential function, c represents the center point of the membership function, α, β and ζ represent the shape adjustment parameters of the membership function, σ represents the standard deviation of the normalized error, η represents the amplitude of the oscillation term, ω represents the frequency of the oscillation term, and λ represents the adjustment coefficient; S25, fuzzy reasoning generates a rough control signal based on the output of the membership function and the fuzzy rule set: Among them, u fuzzy (t) represents the rough control signal after fuzzy reasoning, z i represents the output value of the fuzzy rule set, ν represents the amplitude of the sinusoidal adjustment factor, ω1 represents the frequency of the sinusoidal adjustment factor, κ represents the weight factor of the squared error term, ζ1 represents the sensitivity of the control membership function output, and Ω represents the definition domain of the fuzzy membership function; S26, defuzzifying the rough control signal generated by the fuzzy reasoning to generate a preliminary control instruction.

3. The high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to claim 1, characterized in that: The S3 specifically includes: S31, input the preliminary control instruction generated by the fuzzy controller to the PID controller, the PID controller simultaneously obtains the currently measured coating thickness and the preset target coating thickness, and calculates the small range error e PID (t), as the input error signal of the PID controller; S32, PID controller based on real-time error e PID (t), the derivative of the error and the integral of the error, the proportional gain K p (t), integral gain K i (t) and differential gain K d (t) Perform adaptive adjustment: Among them, K p0 Indicates the initial proportional gain value, K i0 Indicates the initial integral gain value, K d0 represents the initial differential gain value, α p , α i and α d represents the first-order adaptive adjustment coefficient, β p , β i and β d represents the second-order adaptive adjustment coefficient, γ p , γ i and γ d represents the coefficient of the adjusted error derivative, δ d represents the adjustment factor related to error accumulation, represents the rate of change of error, represents the acceleration of error change, Indicates the cumulative amount of error; S33, introduce nonlinear dynamic compensation term u in PID controller NL (t), enhance the correction capability of nonlinear and complex dynamic errors: Among them, γ1, γ2 and γ3 represent nonlinear compensation coefficients, ω2 represents angular frequency, δ1 and δ2 represent adjustment factors of compensation terms; S34, the adaptively adjusted PID control signal u PID (t) and the nonlinear dynamic compensation term u NL (t) Synthesize the final control signal u final (t): Among them, γ4, γ5, γ6, δ3 and ∈ represent adjustment coefficients; S35, control signal u final (t) Perform amplitude limiting processing and eliminate high-frequency noise, filter the control signal after amplitude limiting processing, and obtain the smoothed control signal u filtered (t); S36, smoothing the control signal u filtered (t) Output to the actuator module to accurately adjust the position of the coating die head and the flow rate of the coating liquid in real time, thereby achieving further fine correction of small-range errors.

4. The high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to claim 1, characterized in that: The S4 specifically includes: S41, based on the multidimensional data set formed by the coating thickness, coating die position, coating liquid flow rate and ambient temperature and humidity collected in real time by the sensors deployed in the system, the model predictive controller establishes a dynamic mathematical model to describe the state changes and influencing factors of the coating process; S42, the model predictive controller uses the current system status and historical data to predict the coating thickness change trend in multiple time steps in the future, performs multi-time domain prediction on the thickness in multiple time periods in the future, gradually estimates the coating thickness at each time point, and predicts the short-term thickness change trend and the long-term thickness change trend: Among them, T pred (t+k) represents the predicted coating thickness at the future time t+k, A k and B k represents the prediction matrix, x(t) represents the current system state vector, u(t) represents the current control signal, γ k represents the adjustment coefficient, δ k and η k represents the adjustment factor used to control the prediction accuracy, T env (t) represents the ambient temperature, Indicates the accumulated value of the control signal, Indicates the historical cumulative value of actual coating thickness; S43, the model prediction controller introduces a dynamic correction mechanism for prediction errors. After each prediction is completed, the system automatically calculates the error between the future predicted thickness and the target thickness, corrects the prediction parameters in the model according to the obtained error size, and performs error tolerance analysis to define the allowable error range according to different process requirements; S44. Based on the prediction of the future coating thickness change trend, the model predictive controller automatically generates a control adjustment signal: Among them, u adj (t) represents the adjustment signal generated at the current moment, μ1, μ2 and μ3 represent the adjustment coefficients, and e pred (t+k) represents the prediction error at the future time t+k, P head (t) represents the current position of the coating die, ν1 and ν2 represent dynamic adjustment factors, T actual (t) represents the actual measured coating thickness; S45. When the adjustment signal is generated and sent, the model predictive controller monitors the thickness change during the coating process in real time, and corrects the adjustment signal according to the actual situation. It adopts a multi-level feedback mechanism to compare the gap between the actual change of coating thickness and the predicted value in real time, and uses the feedback signal to fine-tune the adjustment signal to adapt to the fluctuation of the current working conditions.

5. The high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to claim 1, characterized in that: The S5 specifically includes: S51, constructing a dual adaptive neural network structure, wherein the dual adaptive neural network structure includes a real-time adaptive network and a historical adaptive network, wherein the real-time adaptive network is used to process real-time data acquired during the current coating process, and the historical adaptive network performs deep learning based on long-term accumulated data, and the real-time adaptive network and the historical adaptive network learn in parallel; S52, real-time adaptive network continuously collects the error between coating thickness and target thickness real (t), dynamically adjust the parameters of the PID controller, the weights of the fuzzy control rules, and the prediction matrix of the model predictive controller; S53, historical adaptive network analyzes coating performance and control strategies under different environments by deep learning of coating data accumulated over a long period of time, and identifies the long-term control mode of the system under different materials and environmental conditions by looking back on historical data; S54, based on the outputs of the real-time adaptive network and the historical adaptive network, nonlinear feedback control is performed through the weight fusion mechanism, and the weights of the real-time adaptive network and the historical adaptive network are automatically adjusted according to the current working conditions and error fluctuations: Among them, W real (t) represents the dynamic weight of the real-time network, ξ, γ w and λ w represents the dynamic weight of the real-time network, f env (t) represents the influence function of environmental factors on the system state, exp represents the exponential function, Indicates the accumulated value of the control signal; S55, based on the optimization results of the dual adaptive neural network output, the system proportional gain K of the PID controller p (t), integral gain K i (t) and differential gain K d (t) Dynamically adjust and adaptively adjust the fuzzy control rules: Among them, K p0 represents the initial proportional gain, β p ,θ p and p Represents the adaptive adjustment coefficient, W real (t) represents the weight of the real-time network, f hist (t) represents the impact function of historical data on the current system, Indicates the rate of change of real-time coating thickness; S56, the control matrix of model predictive control is dynamically optimized through a dual adaptive network. The real-time network is responsible for quickly adjusting the control matrix based on the current error, and the historical network optimizes the control matrix according to the long-term trend: Among them, A(t) represents the optimized control matrix, A0 represents the initial control matrix, λ real and κ a Represents the adaptive adjustment coefficient.

6. The high-precision coating die head coating thickness adjustment method based on intelligent feedback control according to claim 1, characterized in that: The S6 specifically includes: S61. Based on the analysis of the current feedback data, a nonlinear tentative adjustment signal is generated to test the system's response to a small adjustment of the control parameters: Among them, u probe (t) represents the generated tentative adjustment signal, u0 represents the original control signal, α u and β u represents the adjustment coefficient, e real (t) represents the current error, P head (t) represents the current position of the coating die head, T actual (t represents the actual measured coating thickness, F flow (t) represents the current coating liquid flow rate, Indicates the accumulated value of the control signal; S62. After the tentative adjustment signal is applied, the system immediately starts to monitor the response of various feedback data, using a nonlinear feedback mechanism to introduce the high-order derivatives of the error into the feedback process: Among them, e feedback (t) represents the corrected feedback error, γ e and e represents the regulating factor; S63, by establishing a correlation model between feedback data and control signals, identifying which control parameters have the greatest impact on the current coating process, and obtaining feedback analysis results, wherein the feedback analysis results are processed by a weighted regression model to automatically adjust the contribution weight of each feedback signal to the controller; S64. Based on the feedback analysis results, the system optimizes the weight of the trial adjustment signal through an adaptive mechanism: Among them, W opt (t) represents the optimized feedback weight, σ represents the activation function, f env (t) represents the influence function of environmental factors on the system state, u probe (t) represents a tentative adjustment signal; S65. The system records the feedback data during the adjustment process to form an adaptive learning mechanism. In long-term operation, the system gradually reduces the frequency of trial adjustments and tends to the optimal control parameter combination.

Images