Method and system for controlling bonding strength of grid interlayer and paper

Abstract

The invention provides a method and a system for controlling the bonding strength of a grid interlayer and paper. The method comprises the following steps: acquiring input parameters such as environment temperature and humidity, paper porosity, adhesive viscosity and production line speed; fuzzy processing is carried out on each parameter, a discourse domain is divided into non-equidistant fuzzy level subsets, the division density of a target bonding strength sensitive interval is improved, an asymmetric membership function is constructed, a Gaussian function is adopted in an ascending section, and adjustment is realized through historical data weighted smooth fitting according to a confidence index of a previous control period in a descending section; performing fuzzy reasoning on the basis of a preset fuzzy rule base, adjusting the weight of each rule, and generating a fuzzy output quantity in combination with the weight and a rule of rule triggering intensity aggregation activation; and performing defuzzification and correction on the fuzzy output quantity to generate a final control quantity, and updating the fuzzy rule weight with the maximum contribution by using a difference value.

Description

Technical Field

[0001] This application belongs to the field of adhesive strength control, and particularly relates to a method and system for controlling the adhesive strength between a mesh interlayer and paper. Background Technology

[0002] In the bonding process between the mesh interlayer and paper, precise control of the bonding strength is crucial to product quality, reliability, and durability. Numerous factors influence bonding strength, including ambient temperature and humidity, the porosity of the paper material, the viscosity characteristics of the adhesive used, and the production line speed. These parameters are not only interrelated but also fluctuate during production, making it difficult for classic control algorithms like PID to achieve ideal control results and exhibiting poor adaptability. Fuzzy control can handle uncertainty and fuzzy information. In the fuzzification stage, the universe of discourse of the input variables is typically divided into equidistant regions, and a fixed, symmetrical membership function is used. This ignores the differences in sensitivity across different operating ranges, making it impossible to finely adjust the sensitive range and adapt to changes in operating conditions, thus limiting control accuracy. Furthermore, fuzzy inference often uses fixed-weight fuzzy rules, failing to reflect changes in the degree of influence of each input parameter on the output during production. When operating conditions drift, adaptability and control performance deteriorate. In the defuzzification process, only a definite control variable is usually output. It is impossible to evaluate the confidence of the output result, establish an online correction mechanism based on historical best performance, or optimize the rule base based on feedback from the control effect, which makes it difficult to continuously improve the long-term control performance. Summary of the Invention

[0003] To address the aforementioned problems, in a first aspect, the present invention proposes a method for controlling the adhesive strength between a mesh interlayer and paper, comprising the following steps:

[0004] Obtain input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity, and production line speed;

[0005] Each input parameter is fuzzified, the parameter domain is divided into non-equidistant fuzzy level subsets and the partition density of the sensitive interval of the target adhesion strength is increased. An asymmetric membership function is constructed for the subset, with the ascending segment being a Gaussian function and the descending segment being adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle.

[0006] Fuzzy inference is performed based on a pre-set fuzzy rule base. The weights of each fuzzy rule are adjusted according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters. The adjusted weights are combined with the rules activated by the rule trigger intensity aggregation to generate fuzzy output.

[0007] The initial control quantity is obtained by using a weighted average method, and the coefficient of variation of the corresponding fuzzy output membership degree is calculated as the confidence index for the current period. The initial control quantity is compensated based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity. The weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

[0008] Optionally, the step of fuzzifying each input parameter, dividing the parameter universe into non-equidistant fuzzy level subsets, and increasing the partitioning density of the target adhesion strength sensitive region includes:

[0009] The range of target adhesive strength value of 8-12 N / m is designated as the sensitive range. Within this range, the universe of discourse of each input parameter is divided into 7 fuzzy level subsets. Outside the sensitive range, the universe of discourse of each input parameter is divided into 3 fuzzy level subsets.

[0010] Optionally, the construction of an asymmetric membership function for the subset, with the ascending segment being a Gaussian function and the descending segment adjusted based on the confidence index of the previous control cycle through weighted smoothing fitting of historical data, includes:

[0011] The ascending segment of the membership function is expressed by the formula. The Gaussian function is defined, where c is the center of the subset. The standard deviation is defined as follows: the descending segment is a parameterized curve controlled by one or more shape parameters, the shape parameters being updated using confidence indices from the previous 10 control periods; the update is achieved by calculating a weighted moving average of the confidence indices and adjusting the shape parameters based on the deviation of the average from a preset target confidence level.

[0012] Optionally, the step of performing fuzzy inference based on a preset fuzzy rule base, and adjusting the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, includes:

[0013] For each rule in the fuzzy rule base, the weight adjustment factor is calculated as follows: calculate the normalized gradient norm of the time series of all input parameters included in the premise of the rule over the past 5 sampling periods, and calculate the arithmetic mean; calculate the absolute value of the Pearson cross-correlation coefficient between all pairs of input parameters included in the premise of the rule over the past 20 sampling periods, and calculate the arithmetic mean; sum the two averages with weights to obtain the weight adjustment factor; update the weight of the rule based on the weight adjustment factor.

[0014] Optionally, the rule that combines the adjusted weights with the rule-triggered intensity aggregation activation includes:

[0015] For the i-th activated rule, calculate the trigger strength. With adjusted weights product The product is used to truncate the fuzzy set of the rule consequent; the maximum value method is used to merge all truncated fuzzy sets to obtain the aggregated fuzzy output.

[0016] Optionally, the calculation of the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period includes:

[0017] Calculate the standard deviation of the membership function over the universe of discourse of the fuzzy output. and mean ; through formula The coefficient of variation is calculated and used as the confidence index for the current period.

[0018] Optionally, the step of compensating the initial control quantity based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity includes:

[0019] A radial basis function neural network model is established, taking the confidence index as input and the optimal control compensation value as output. This model is trained offline using 300 sets of historical best production data. The confidence index calculated in the current period is input into the model to obtain the compensation value. ; set the initial control quantity Compensation value Add them together to get the final control quantity.

[0020] Optionally, updating the weight of at least one fuzzy rule that contributes most to generating the initial control quantity using the difference between the final control quantity and the initial control quantity includes:

[0021] The top three fuzzy rules that contribute the most to generating the initial control quantity in fuzzy inference are identified. Based on the difference between the final control quantity and the initial control quantity, the weights of the top three fuzzy rules are updated using gradient descent with a learning rate of 0.1.

[0022] In a second aspect, the present invention provides a bonding strength control system for a mesh interlayer and paper, comprising the following modules:

[0023] The acquisition module is used to acquire input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity and production line speed.

[0024] The adjustment module is used to fuzzify each input parameter, divide the parameter domain into non-equidistant fuzzy level subsets and increase the partition density of the sensitive interval of the target adhesion strength, and construct an asymmetric membership function for the subset. The ascending segment is a Gaussian function, and the descending segment is adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle.

[0025] The generation module is used to perform fuzzy inference based on a preset fuzzy rule library. It adjusts the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, and combines the adjusted weights with the rules activated by the rule trigger intensity aggregation to generate fuzzy output.

[0026] The update module is used to obtain the initial control quantity using a weighted average method, and calculate the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period; based on the deviation between the confidence index and the historical best control quantity model, the initial control quantity is compensated to obtain the final control quantity; and the weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

[0027] Preferably, the step of fuzzifying each input parameter, dividing the parameter universe into non-equidistant fuzzy level subsets, and increasing the partitioning density of the sensitive region of the target adhesion strength includes:

[0028] The range of target adhesive strength value of 8-12 N / m is designated as the sensitive range. Within this range, the universe of discourse of each input parameter is divided into 7 fuzzy level subsets. Outside the sensitive range, the universe of discourse of each input parameter is divided into 3 fuzzy level subsets.

[0029] Preferably, the construction of an asymmetric membership function for the subset, wherein the ascending segment is a Gaussian function and the descending segment is adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle, includes:

[0030] The ascending segment of the membership function is expressed by the formula. The Gaussian function is defined, where c is the center of the subset. The standard deviation is defined as follows: the descending segment is a parameterized curve controlled by one or more shape parameters, the shape parameters being updated using confidence indices from the previous 10 control periods; the update is achieved by calculating a weighted moving average of the confidence indices and adjusting the shape parameters based on the deviation of the average from a preset target confidence level.

[0031] Preferably, the step of performing fuzzy inference based on a preset fuzzy rule base, and adjusting the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, includes:

[0032] For each rule in the fuzzy rule base, the weight adjustment factor is calculated as follows: calculate the normalized gradient norm of the time series of all input parameters included in the premise of the rule over the past 5 sampling periods, and calculate the arithmetic mean; calculate the absolute value of the Pearson cross-correlation coefficient between all pairs of input parameters included in the premise of the rule over the past 20 sampling periods, and calculate the arithmetic mean; sum the two averages with weights to obtain the weight adjustment factor; update the weight of the rule based on the weight adjustment factor.

[0033] Preferably, the rule that combines the adjusted weights with the rule-triggered intensity aggregation activation includes:

[0034] For the i-th activated rule, calculate the trigger strength. With adjusted weights product The product is used to truncate the fuzzy set of the rule consequent; the maximum value method is used to merge all truncated fuzzy sets to obtain the aggregated fuzzy output.

[0035] Preferably, the calculation of the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period includes:

[0036] Calculate the standard deviation of the membership function over the universe of discourse of the fuzzy output. and mean ; through formula The coefficient of variation is calculated and used as the confidence index for the current period.

[0037] Preferably, the step of compensating the initial control quantity based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity includes:

[0038] A radial basis function neural network model is established, taking the confidence index as input and the optimal control compensation value as output. This model is trained offline using 300 sets of historical best production data. The confidence index calculated in the current period is input into the model to obtain the compensation value. ; set the initial control quantity Compensation value Add them together to get the final control quantity.

[0039] Preferably, updating the weight of at least one fuzzy rule that contributes most to generating the initial control quantity using the difference between the final control quantity and the initial control quantity includes:

[0040] The top three fuzzy rules that contribute the most to generating the initial control quantity in fuzzy inference are identified. Based on the difference between the final control quantity and the initial control quantity, the weights of the top three fuzzy rules are updated using gradient descent with a learning rate of 0.1.

[0041] This invention represents the state of input variables by dividing the input parameters affecting adhesion strength into non-uniform fuzzy levels and further refining the division within key sensitive intervals. It also incorporates an asymmetric membership function modified based on historical data and confidence indices. Furthermore, by analyzing the temporal trends and interrelationships of each parameter, different weights are assigned to the fuzzy rules, allowing the inference process to focus on the dominant influencing factors under the current operating conditions. When generating the final control quantity, a comparison and correction step between the confidence assessment and the historical best model is incorporated to compensate for the initial control quantity, improving the reliability of the output control. By using the correction difference of the control quantity to update the weights of the core fuzzy rules, the performance of the entire control system can be continuously improved. Attached Figure Description

[0042] Figure 1 A flowchart of the first embodiment;

[0043] Figure 2 This is a schematic diagram of an asymmetric membership function;

[0044] Figure 3 This is a schematic diagram of the output format based on the confidence index;

[0045] Figure 4 This is a schematic diagram for the correction of the control quantity. Detailed Implementation

[0046] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0047] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0048] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0049] In the first embodiment, the present invention proposes a method for controlling the adhesive strength between the mesh interlayer and the paper, such as... Figure 1 This includes the following steps:

[0050] S1, Obtain input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity and production line speed;

[0051] Sensors are installed at appropriate locations on the production line to collect data in real time. Preferably, ambient temperature and humidity are obtained through temperature and humidity sensors positioned above the glue application area; paper porosity is measured online using a non-contact airflow penetration rate detector; adhesive viscosity is monitored using a rotary online viscometer installed in the glue supply pipeline; and production line speed is obtained through a photoelectric encoder connected to the main drive roller.

[0052] S2, fuzzify each input parameter, divide the parameter domain into non-equidistant fuzzy level subsets and increase the partition density of the sensitive interval of the target adhesion strength, and construct an asymmetric membership function for the subset. The ascending segment is a Gaussian function, and the descending segment is adjusted by weighted smoothing fitting of historical data according to the confidence index of the previous control cycle.

[0053] Taking adhesive viscosity as an example, the domain of discourse is 100 to 500 mPa·s. Data analysis shows that the viscosity has the most significant impact on adhesive strength in the 280 to 350 mPa·s range, which is the sensitive range. Therefore, three fuzzy-level subsets are defined within this range: low, moderate, and high. Outside this range, only two sparser subsets are defined: too low and too high, resulting in a non-equidistant partition. For the moderate subset, the ascending segment of the membership function adopts a standard Gaussian function with a center point of 315 mPa·s; the shape of the descending segment is not fixed but is based on the confidence index calculated in the previous period. To determine this, store the viscosity values ​​and corresponding control effect data points for the most recent N control cycles. Use an exponentially weighted moving average method to fit these historical data points to obtain a baseline curve. The confidence index for the previous cycle... As an adjustment factor, if A higher value indicates greater uncertainty in the previous control result, suggesting that the current fit is more biased towards the long-term trend of historical data; conversely, if... If the value is low, the weight of the latest data point is increased, so that the descending curve can respond to changes in operating conditions more quickly.

[0054] S3 performs fuzzy inference based on a preset fuzzy rule base, adjusts the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, and combines the adjusted weights with the rules activated by the rule trigger intensity aggregation to generate fuzzy output.

[0055] The fuzzy rule base is a series of IF-THEN rules, such as the rule... For example, if the ambient humidity is high and the paper porosity is high, the glue amount setting will increase. The initial weight for each rule is 1. In each control cycle, the time-series gradient norm of each input parameter is calculated; for example, the absolute value of the difference between the current humidity and the humidity at the previous moment is calculated. The larger this value, the more drastic the humidity change. The Pearson cross-correlation coefficient between any two input parameters is calculated; for example, the correlation between humidity and paper porosity sequences over the past minute is calculated. For each rule... The weights are determined by the gradient norm and cross-correlation coefficient of the parameters in the preconditions. In an optional embodiment, the weight adjustment formula is as follows: Where a, b, and c are adjustment coefficients. The trigger strength of each rule is calculated based on the membership degree of each input parameter, for example, using the minimum value method. The trigger strength of each activation rule is multiplied by the adjusted weight, and then the weighted output results of all rules are aggregated into a fuzzy output set using the maximum value method.

[0056] S4. The initial control quantity is obtained by using the weighted average method, and the coefficient of variation of the corresponding fuzzy output membership degree is calculated as the confidence index of the current period. The initial control quantity is compensated based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity. The weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

[0057] The weighted average method is used to defuzzify the aggregated fuzzy output to obtain the initial control quantity. For example, the glue content is set to 5.2 g / m². Calculate the standard deviation of the fuzzy output membership distribution and compare it with the mean. The ratio is used as a confidence index for the current period. A sharp membership distribution corresponds to low... The value indicates a high confidence level. It incorporates a radial basis function neural network model trained on historical production data, which can provide a historically optimal control input based on the current input parameters. The control quantity is generated through compensation using, for example, the following formula: When the confidence level is low, the final control input will rely more on the historical best model. Calculate the difference. Identify the top three fuzzy rules in terms of weighted trigger strength during fuzzy inference, and fine-tune the base weights according to the magnitude and direction of ΔC, for example... η is the learning rate. The adjusted base weights will be used for weight calculation in the next control cycle, thereby enabling online learning and optimization of the rule base.

[0058] In an optional embodiment, the step of fuzzifying each input parameter, dividing the parameter universe of discourse into non-equidistant fuzzy level subsets, and increasing the partition density of the target adhesion strength sensitive region includes:

[0059] The range of target adhesive strength value of 8-12 N / m is designated as the sensitive range. Within this range, the universe of discourse of each input parameter is divided into 7 fuzzy level subsets. Outside the sensitive range, the universe of discourse of each input parameter is divided into 3 fuzzy level subsets.

[0060] The current control is determined based on the target adhesive strength value to determine if it is within the sensitive range. For example, if the target adhesive strength of the current batch of products is 10 N / m, it is determined to be within the sensitive range, requiring fine-grained control. For input parameters such as temperature, the universe of discourse ranges from 150℃ to 250℃ and is divided into seven fuzzy subsets: very low, lower, slightly lower, medium, slightly higher, higher, and very high. This fine-grained division can represent the parameter state and provide richer information for subsequent fuzzy inference. Conversely, if the target adhesive strength of another batch of products is 6 N / m, which falls outside the sensitive range, a coarser control strategy is adopted. For the same temperature parameter, the universe of discourse will be divided into three fuzzy subsets: low, medium, and high. This simplifies the complexity of the fuzzy rule base, reduces computational load, and improves response speed while ensuring basic control effectiveness. By adjusting the number of fuzzy subsets, control under different accuracy requirements is achieved.

[0061] In an optional embodiment, the construction of an asymmetric membership function for the subset, with the ascending segment being a Gaussian function and the descending segment being adjusted based on the confidence index of the previous control cycle by performing a weighted smoothing fit on historical data, includes:

[0062] The ascending segment of the membership function is expressed by the formula. The Gaussian function is defined, where c is the center of the subset. The standard deviation is defined as follows: the descending segment is a parameterized curve controlled by one or more shape parameters, the shape parameters being updated using confidence indices from the previous 10 control periods; the update is achieved by calculating a weighted moving average of the confidence indices and adjusting the shape parameters based on the deviation of the average from a preset target confidence level.

[0063] The membership function consists of an ascending segment and a descending segment. The ascending segment preferably uses a Gaussian function, and the descending segment preferably uses a third-order Bézier curve. The coordinates of its two intermediate control points are used as the shape parameters to calculate the weighted moving average of the confidence index for the first 10 control periods. The weight of the confidence index for each period decays exponentially over time, with the most recent period having the highest weight; the weighted moving average is then calculated. confidence level of the preset target Deviation between Based on the deviation E, the coordinates P(t) of the two intermediate control points are adjusted using proportional-integral control, and the adjustment formula is as follows: Where P(t) represents the coordinates of the control point in the current cycle, and P(t-1) represents the coordinates of the control point in the previous cycle. This is the proportional gain coefficient. ΣE represents the integral gain coefficient, and ΣE represents the cumulative sum of historical deviations. By adjusting the coordinates of the control points, the shape of the Bézier curve is changed, thereby optimizing the form of the membership function and causing the system's confidence index to converge to the target confidence level. However, those skilled in the art should know that the aforementioned descent is not limited to a third-order Bézier curve.

[0064] Taking a subset of temperature parameters, such as medium, as an example, the central value c is 200℃, and the standard deviation is... The value is 10. When the input temperature value approaches 200℃ from below 200℃, the membership calculation follows a standard Gaussian function, resulting in a smooth and symmetrical change. For the descending segment above 200℃, the shape is determined by one or more adjustable shape parameters, allowing the function shape to be adjusted according to the actual control effect, thereby achieving asymmetry. To update the shape parameters, the confidence index for the past 10 control cycles is recorded, assuming the values ​​are 0.85, 0.88, 0.86, 0.89, 0.91, 0.90, 0.87, 0.88, 0.92, and 0.93. The weighted moving average of these values ​​is calculated, giving higher weight to recent data, resulting in an average value of 0.89. If the preset target confidence level is 0.95, the current deviation is 0.06. Based on the positive deviation, it is determined that the current control effect needs improvement, and the shape parameters of the descending curve are adjusted accordingly. Figure 2 For example, making the slope steeper can lead to more explicit outputs in subsequent controls, thereby increasing confidence.

[0065] In an optional embodiment, the fuzzy inference based on a preset fuzzy rule base, adjusting the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, includes:

[0066] For each rule in the fuzzy rule base, the weight adjustment factor is calculated as follows: calculate the normalized gradient norm of the time series of all input parameters included in the premise of the rule over the past 5 sampling periods, and calculate the arithmetic mean; calculate the absolute value of the Pearson cross-correlation coefficient between all pairs of input parameters included in the premise of the rule over the past 20 sampling periods, and calculate the arithmetic mean; sum the two averages with weights to obtain the weight adjustment factor; update the weight of the rule based on the weight adjustment factor.

[0067] A fuzzy rule states that if both temperature and pressure are high, then the amount of adhesive is high. Obtain the temperature and pressure data sequences for the most recent 5 sampling periods. For example, the temperature sequence is 220, 221, 223, 222, 225, and the pressure sequence is 1.5, 1.6, 1.5, 1.7, 1.6. Calculate the normalized gradient norm for both sequences, representing recent volatility. Assuming the obtained values ​​are 0.4 and 0.6 respectively, the arithmetic mean is 0.5. The obtained value reflects the overall activity level of the rule's preconditions. Analyze the correlation between temperature and pressure data over the past 20 sampling periods. By calculating the Pearson cross-correlation coefficient for the two time series, a value representing the strength of their linear relationship is obtained. Assuming an absolute value of 0.85, it indicates a strong positive correlation between temperature and pressure. Weight the gradient norm mean of 0.5 and the cross-correlation coefficient mean of 0.85, for example, with weights of 0.3 and 0.7 respectively, resulting in a weight adjustment factor of 0.745. The original weight of the rule is multiplied by the adjustment factor to update the weight, thereby giving rules with active preconditions and strong internal parameter correlation a more dominant position in reasoning.

[0068] In an optional embodiment, the rule that combines the adjusted weights with the rule-triggered intensity aggregation activation includes:

[0069] For the i-th activated rule, calculate the trigger strength. With adjusted weights product The product is used to truncate the fuzzy set of the rule consequent; the maximum value method is used to merge all truncated fuzzy sets to obtain the aggregated fuzzy output.

[0070] Suppose that at a certain moment, two rules are activated. The prerequisite matching degree of rule one is its trigger strength. The adjusted weight is 0.8. The value is 1.1, and the consequent of the rule is a fuzzy set A, such as medium glue content. The trigger strength of rule two. The adjusted weight is 0.6. The value is 0.9, and the consequent of the rule is fuzzy set B, for example, a slightly larger glue quantity. The weighted trigger strength of the two rules is calculated separately. For rule one, the value is 0.88. For rule two, the value is 0.54. The calculated weighted trigger strength is used to truncate the respective consequent fuzzy sets. The membership function of fuzzy set A, exceeding 0.88 on the vertical axis, is flattened to 0.88, forming a new truncated fuzzy set A'. Similarly, the membership function of fuzzy set B, exceeding 0.54 on the vertical axis, is flattened to 0.54, forming a truncated fuzzy set B'. A' and B' are merged using the maximum value method. That is, at each point in the entire glue quantity universe, the larger membership value between A' and B' is taken as the membership value of the aggregated output fuzzy set. This fuzzy set integrates the contributions of all activation rules and is prepared for the next step of defuzzification.

[0071] In an optional embodiment, calculating the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period includes:

[0072] Calculate the standard deviation of the membership function over the universe of discourse of the fuzzy output. and mean ; through formula The coefficient of variation is calculated and used as the confidence index for the current period.

[0073] After fuzzy inference and rule aggregation, a fuzzy set is obtained for the output control quantity, such as glue quantity. The membership function of this fuzzy set defines the degree of membership for each possible value in the glue quantity domain, ranging from 0 g / s to 10 g / s. The membership function is then treated as a numerical distribution, and the arithmetic mean of all membership values ​​is calculated. For example, if the output fuzzy set has a relatively concentrated and tall shape, the average membership degree might be 0.6. Calculate the standard deviation of the membership degree values ​​relative to the mean. A sharp output fuzzy set will have membership values ​​closely distributed around the mean, resulting in a small standard deviation, such as 0.1. Conversely, a flat output fuzzy set indicates high uncertainty, with a dispersed distribution of membership values ​​and a larger standard deviation. The coefficient of variation (CV) is calculated using the formula and is approximately 0.167. This CV value is used as a confidence index for the current control cycle; the smaller the value, the lower the confidence level. Figure 3 The more explicit and reliable the result of fuzzy reasoning, the higher the confidence level of the output.

[0074] In an optional embodiment, the step of compensating the initial control quantity based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity includes:

[0075] A radial basis function neural network model is established, taking the confidence index as input and the optimal control compensation value as output. This model is trained offline using 300 sets of historical best production data. The confidence index calculated in the current period is input into the model to obtain the compensation value. ; set the initial control quantity Compensation value Add them together to get the final control quantity.

[0076] Using, for example, 300 sets of historical data containing confidence indices and corresponding optimal compensation values, a radial basis function neural network is trained offline. For instance, one set of historical data might show that when the confidence indices are 0.45, to achieve the optimal binding effect, the initial control input of the fuzzy controller needs to be increased by 0.2 units. By learning from these 300 sets of data, the neural network masters the nonlinear mapping relationship between confidence and the required compensation. In actual operation, assuming the initial control input given by the fuzzy controller in the current control cycle... The adhesive volume is 15.5 units, and the calculated confidence level is 0.5. This confidence level of 0.5 is input into the pre-trained neural network. The neural network calculates based on its internal model and outputs a compensation value. For example, -0.15 units. Adding the initial control value to the compensation value yields a final control value of 15.35 units. Figure 4 The compensated final control quantity is then sent to the actuator, realizing intelligent correction of the initial fuzzy control result.

[0077] In an optional embodiment, updating the weight of at least one fuzzy rule that contributes most to generating the initial control quantity using the difference between the final control quantity and the initial control quantity includes:

[0078] The top three fuzzy rules that contribute the most to generating the initial control quantity in fuzzy inference are identified. Based on the difference between the final control quantity and the initial control quantity, the weights of the top three fuzzy rules are updated using gradient descent with a learning rate of 0.1.

[0079] Specifically, after calculating the initial control value of 15.5 and obtaining the final control value of 15.35 through neural network compensation, the fuzzy inference process is analyzed to identify the three rules that contribute most to generating the initial control value of 15.5, such as rule 7, rule 18, and rule 25. The contribution can be ranked according to the weighted trigger strength of each rule. The difference between the final control value and the initial control value is calculated, yielding an error of -0.15. This negative error indicates that the initial fuzzy inference result is too high and needs to be adjusted downwards. A gradient descent algorithm with a learning rate of 0.1 is then used to update the weights of the three key rules. The update direction is consistent with the error sign, i.e., reducing the weights of the rules to weaken their influence in similar future scenarios. For example, for rule 7, the new weight will be equal to the old weight minus a positive adjustment amount, which is related to the learning rate of 0.1, the error of -0.15, and the activation state of rule 7 itself. Through this method, online learning can be performed from the behavior of the compensation module, gradually optimizing the fuzzy rule base and reducing future reliance on the compensation module.

[0080] In a second embodiment, a bonding strength control system for the mesh interlayer and paper is provided, comprising the following modules:

[0081] The acquisition module is used to acquire input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity and production line speed.

[0082] The adjustment module is used to fuzzify each input parameter, divide the parameter domain into non-equidistant fuzzy level subsets and increase the partition density of the sensitive interval of the target adhesion strength, and construct an asymmetric membership function for the subset. The ascending segment is a Gaussian function, and the descending segment is adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle.

[0083] The generation module is used to perform fuzzy inference based on a preset fuzzy rule library. It adjusts the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, and combines the adjusted weights with the rules activated by the rule trigger intensity aggregation to generate fuzzy output.

[0084] The update module is used to obtain the initial control quantity using a weighted average method, and calculate the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period; based on the deviation between the confidence index and the historical best control quantity model, the initial control quantity is compensated to obtain the final control quantity; and the weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code.

[0086] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for controlling the adhesive strength between a mesh interlayer and paper, characterized in that, Includes the following steps: Obtain input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity, and production line speed; Each input parameter is fuzzified, the parameter domain is divided into non-equidistant fuzzy level subsets and the partition density of the sensitive interval of the target adhesion strength is increased. An asymmetric membership function is constructed for the subset, with the ascending segment being a Gaussian function and the descending segment being adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle. Fuzzy inference is performed based on a pre-set fuzzy rule base. The weights of each fuzzy rule are adjusted according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters. The adjusted weights are combined with the rules activated by the rule trigger intensity aggregation to generate fuzzy output. The initial control quantity is obtained by using a weighted average method, and the coefficient of variation of the corresponding fuzzy output membership degree is calculated as the confidence index for the current period. The initial control quantity is compensated based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity. The weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

2. The method according to claim 1, characterized in that, The process of fuzzifying each input parameter, dividing the parameter universe into non-equidistant fuzzy level subsets, and increasing the partitioning density of the sensitive region of the target adhesion strength includes: The range of target adhesive strength value of 8-12 N / m is designated as the sensitive range. Within this range, the universe of discourse of each input parameter is divided into 7 fuzzy level subsets. Outside the sensitive range, the universe of discourse of each input parameter is divided into 3 fuzzy level subsets.

3. The method according to claim 1, characterized in that, The process involves constructing an asymmetric membership function for the subset, with the ascending segment being a Gaussian function and the descending segment being adjusted based on the confidence index of the previous control cycle through weighted smoothing fitting of historical data, including: The ascending segment of the membership function is expressed by the formula. The Gaussian function is defined, where c is the center of the subset. The standard deviation is defined as follows: the descending segment is a parameterized curve controlled by one or more shape parameters, the shape parameters being updated using confidence indices from the previous 10 control periods; the update is achieved by calculating a weighted moving average of the confidence indices and adjusting the shape parameters based on the deviation of the average from a preset target confidence level.

4. The method according to claim 1, characterized in that, The fuzzy inference based on a preset fuzzy rule base, adjusting the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, includes: For each rule in the fuzzy rule base, the weight adjustment factor is calculated as follows: calculate the normalized gradient norm of the time series of all input parameters included in the premise of the rule over the past 5 sampling periods, and calculate the arithmetic mean; calculate the absolute value of the Pearson cross-correlation coefficient between all pairs of input parameters included in the premise of the rule over the past 20 sampling periods, and calculate the arithmetic mean; sum the two averages with weights to obtain the weight adjustment factor; update the weight of the rule based on the weight adjustment factor.

5. The method according to claim 1, characterized in that, The rules for combining the adjusted weights with the rules for triggering intensity aggregation activation include: For the i-th activated rule, calculate the trigger strength. With adjusted weights product The product is used to truncate the fuzzy set of the rule consequent; the maximum value method is used to merge all truncated fuzzy sets to obtain the aggregated fuzzy output.

6. The method according to claim 1, characterized in that, The calculation of the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period includes: Calculate the standard deviation of the membership function over the universe of discourse of the fuzzy output. and mean ; through formula The coefficient of variation is calculated and used as the confidence index for the current period.

7. The method according to claim 1, characterized in that, The process of compensating for the initial control quantity based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity includes: A radial basis function neural network model is established, taking the confidence index as input and the optimal control compensation value as output. This model is trained offline using 300 sets of historical best production data. The confidence index calculated in the current period is input into the model to obtain the compensation value. ; set the initial control quantity Compensation value Add them together to get the final control quantity.

8. The method according to claim 1, characterized in that, The step of updating the weight of at least one fuzzy rule that contributes the most to generating the initial control quantity using the difference between the final control quantity and the initial control quantity includes: The top three fuzzy rules that contribute the most to generating the initial control quantity in fuzzy inference are identified. Based on the difference between the final control quantity and the initial control quantity, the weights of the top three fuzzy rules are updated using gradient descent with a learning rate of 0.

1.

9. A control system for the adhesion strength between a mesh interlayer and paper, characterized in that, Includes the following modules: The acquisition module is used to acquire input parameters that affect the bonding strength, including ambient temperature and humidity, paper porosity, adhesive viscosity and production line speed. The adjustment module is used to fuzzify each input parameter, divide the parameter domain into non-equidistant fuzzy level subsets and increase the partition density of the sensitive interval of the target adhesion strength, and construct an asymmetric membership function for the subset. The ascending segment is a Gaussian function, and the descending segment is adjusted by weighted smoothing fitting of historical data based on the confidence index of the previous control cycle. The generation module is used to perform fuzzy inference based on a preset fuzzy rule library. It adjusts the weights of each fuzzy rule according to the time series gradient norm of the input parameters and the cross-correlation coefficient between the parameters, and combines the adjusted weights with the rules activated by the rule trigger intensity aggregation to generate fuzzy output. The update module is used to obtain the initial control quantity using the weighted average method and calculate the coefficient of variation of the corresponding fuzzy output membership degree as the confidence index for the current period. The initial control quantity is compensated based on the deviation between the confidence index and the historical best control quantity model to obtain the final control quantity; and the weight of at least one fuzzy rule that contributes the most to the generation of the initial control quantity is updated using the difference between the final control quantity and the initial control quantity.

10. The system according to claim 9, characterized in that, The process of fuzzifying each input parameter, dividing the parameter universe into non-equidistant fuzzy level subsets, and increasing the partitioning density of the sensitive region of the target adhesion strength includes: The range of target adhesive strength value of 8-12 N / m is designated as the sensitive range. Within this range, the universe of discourse of each input parameter is divided into 7 fuzzy level subsets. Outside the sensitive range, the universe of discourse of each input parameter is divided into 3 fuzzy level subsets.

Images