Intelligent textile printing and dyeing process optimization method and system

Abstract

The invention provides an intelligent textile printing and dyeing process optimization method and system, and the method comprises the steps: determining parameters of a membership function according to the features of a current batch of materials, and carrying out the fuzzification of a real-time technological parameter through employing the membership function, and obtaining fuzzy technological features; the activation contribution degree of each fuzzy rule is calculated according to the historical prediction accuracy of the fuzzy rules and the correlation degree of the process benefits, and then the process cooperation state is obtained; the sensor time sequence data and the process cooperation state are input into a multi-mode time sequence main capsule layer to obtain a main capsule output vector, the confidence coefficient of the vector is calculated, the influence of the vector during follow-up routing to a high-layer capsule layer is adjusted according to the confidence coefficient, and the high-layer capsule layer generates a regulation and control parameter set for follow-up procedures; and obtaining quality detection data of a subsequent process, obtaining a multi-target deviation vector according to the quality detection data and the regulation and control parameter set, and optimizing the high-layer capsule layer by using the multi-target deviation vector to obtain an optimized regulation and control parameter set.

Description

Technical Field

[0001] This application belongs to the textile field, and in particular relates to an intelligent method and system for optimizing textile printing and dyeing processes. Background Technology

[0002] As a component of traditional manufacturing, the textile printing and dyeing industry involves complex and variable production processes, encompassing numerous physicochemical processes that impact product quality, resource consumption, and environmental protection. Traditional textile printing and dyeing process control relies on manual experience and fixed process parameter settings. This approach struggles to adapt to batch-to-batch variations in raw materials, equipment fluctuations, and evolving market demands for product diversification and high quality. Firstly, the physicochemical properties of raw materials such as dyes, auxiliaries, and fabrics themselves vary between batches. Even with the same material from the same supplier, subtle variations in characteristics can lead to quality issues like color differences and poor dyeing evenness in the final product. Operators often need to make trial-and-error adjustments based on experience, which is not only time-consuming and labor-intensive but also difficult to guarantee the stability and consistency of optimization results. Secondly, the performance of printing and dyeing equipment undergoes subtle changes over long-term operation, such as pipe scaling and sensor aging. These factors also interfere with the accurate execution of process parameters and the stable maintenance of process states, rendering fixed process parameters no longer optimal. Traditional control systems are poorly adaptable to such dynamic changes, often lagging behind the occurrence of problems and resulting in unnecessary defects and waste. To enhance the intelligence level of textile printing and dyeing, academia and industry have introduced artificial intelligence technologies such as expert systems, fuzzy logic, and neural networks to model and optimize the printing and dyeing process. Fuzzy logic demonstrates advantages in handling the fuzziness and uncertainty inherent in the printing and dyeing process, transforming operator experience into executable fuzzy rules. Neural networks excel at extracting nonlinear relationships from large amounts of historical data for predicting key process indicators or diagnosing faults. However, the determination of membership functions in traditional fuzzy systems often relies on expert experience, making it difficult to dynamically adapt to real-time changes in material properties and process disturbances, thus limiting the accuracy of fuzzy inference. Furthermore, the construction and optimization of fuzzy rule bases are time-consuming, and the correlation between rule effectiveness evaluation and actual production benefits is not strong enough to ensure the continued applicability of the rules. For the fusion processing of time-series data and multi-source heterogeneous information, traditional methods often struggle to fully explore their deep features and dynamic correlations. For example, how to effectively combine high-frequency sensor data with macroscopic process coordination status information to achieve accurate prediction and proactive control of future process trends remains a pressing technical challenge. Meanwhile, existing optimization methods mostly focus on optimizing local processes, lacking a systematic consideration of the impact of the preceding and following processes in the entire dyeing and printing process and the ability to optimize globally. Furthermore, the feedback and correction mechanisms for optimization results are usually not refined or efficient, making it difficult to achieve continuous self-evolution of the model and accurate achievement of multi-dimensional quality goals. Summary of the Invention

[0003] To address the aforementioned problems, this application proposes an intelligent method for optimizing textile printing and dyeing processes, comprising: Obtain real-time and historical process parameters of the dyeing and printing process, determine the parameters of the membership function based on the characteristics of the current batch of materials, and use the membership function to fuzzify the real-time process parameters to obtain fuzzy process features; The activation contribution of each fuzzy rule is calculated based on the correlation between the historical prediction accuracy of the fuzzy rules and the process benefits. The process synergy status is obtained by using the activation contribution of the fuzzy rules and the fuzzy process characteristics. Sensor timing data and process coordination status are input into the multimodal timing master capsule layer to obtain the master capsule output vector. The confidence level of the vector is calculated, and the influence of the vector in subsequent routing to the higher capsule layer is adjusted according to the confidence level. The higher capsule layer generates a set of control parameters for subsequent processes. The quality inspection data of subsequent processes is obtained, and a multi-objective deviation vector is obtained based on the quality inspection data and the set of control parameters. The multi-objective deviation vector is then used to optimize the high-level capsule layer to obtain the adjusted and optimized set of control parameters.

[0004] Optionally, determining the parameters of the membership function based on the characteristics of the current batch of materials includes: Obtain the material feature vector of the current batch of materials, input the material feature vector into the radial basis function network, and obtain the center point value and standard deviation value of the Gaussian membership function of the preset fuzzy subset for each process parameter.

[0005] Optionally, the step of using membership functions to fuzzify real-time process parameters to obtain fuzzy process features includes: Obtain the Gaussian membership function of the fuzzy subset corresponding to each process parameter, as well as the corresponding center point value and standard deviation value; Substitute the numerical values ​​of the process parameters into the Gaussian membership function of each corresponding fuzzy subset. The membership degree values ​​of the process parameters for each fuzzy subset are obtained, where x is the parameter value, c is the center point value, and σ is the standard deviation. All membership values ​​of each process parameter constitute the fuzzy process feature of the process parameter.

[0006] Optionally, the step of calculating the activation contribution of each fuzzy rule based on the correlation between the historical prediction accuracy of the fuzzy rule and the process benefits includes: For each fuzzy rule in the fuzzy rule base, select all historical process batches from the database of historical process parameters whose membership product of each fuzzy variable in the condition part is greater than the preset value. For each historical process batch, the root mean square error (RMSE) between the predicted intermediate process target value defined in the result portion of the fuzzy rule and the actual recorded intermediate process target value for that batch is calculated. Then, the historical prediction accuracy of the rule is... ; Obtain at least three final product quality indicators and at least two process benefit indicators corresponding to each historical process batch; calculate the Pearson correlation coefficient between the prediction accuracy of the rule and each quality indicator and benefit indicator, and use the weighted average of each Pearson correlation coefficient as the process benefit correlation index CB; According to the formula The activation contribution of the fuzzy rule is calculated, where , The preset coefficients, and Optionally, obtaining the process cooperative state using the activation contribution of fuzzy rules and fuzzy process features includes: For each fuzzy rule, the fuzzy process features are substituted into the condition part of the rule, and the total activation intensity of the condition part of the rule is calculated using the algebraic product T-norm operator. The total activation intensity of the rule is multiplied by the activation contribution degree ACD of the rule to obtain the weighted activation intensity of the rule. Multiply the weighted activation intensity of each rule by the feature value of the process coordination state defined in the fuzzy rule result part, sum the product results of all rules, and then divide by the sum of the weighted activation intensities of all rules to obtain the process coordination state at the current moment.

[0007] Optionally, calculating the confidence level of the vector includes: Within the current processing time window, which is preset or dynamically adjusted according to the characteristics of the process stage, all main capsule output vectors output by the multimodal time-series main capsule layer are obtained; the L2 norm of each main capsule output vector is calculated; and the maximum value among the L2 norms of all main capsule output vectors within the time window is taken as the reference value. The confidence level of the main capsule output vector is obtained by dividing the L2 norm of each main capsule output vector by the benchmark value.

[0008] Optionally, adjusting the influence of the vector on subsequent routing to higher-level capsule layers based on the confidence level, wherein the higher-level capsule layer generates a set of control parameters for subsequent processes, includes: For each master capsule layer, the output master capsule vector and their corresponding confidence levels : For each high-level capsule j, through the transformation matrix Main capsule vector Convert to prediction vector ; Initialize the input of high-level capsule j It is a zero vector; During the routing iteration process, the prediction vector for the current iteration round is calculated. With the current state vector of high-level capsule j The dot product is then normalized across all higher-level capsules using the softmax function to obtain the vector for each master capsule. The connection routing coupling coefficient between the high-level capsule j and the high-level capsule j ;according to Update the input of high-level capsule j by updating the state vector of high-level capsule j through the squeezing function; After the iteration is completed, the state vectors of all high-level capsules are concatenated to obtain a feature vector, which is then input into a multilayer perceptron (MLP) to obtain a set of control parameters.

[0009] Optionally, the optimization of the high-level capsule layer using the multi-objective deviation vector to obtain the adjusted and optimized set of control parameters includes: The loss function value is calculated based on the multi-objective bias vector; the gradient of the loss function value with respect to the weight parameters of each layer of the MLP is calculated through backpropagation, and the gradient of the loss function value with respect to each element in the transformation matrix connecting the main capsule layer and the higher capsule layers is also calculated through backpropagation. The weight parameters of the MLP and the transformation matrix are updated according to the gradient; the control parameter set is regenerated using the updated MLP and transformation matrix.

[0010] This invention also proposes an intelligent textile printing and dyeing process optimization system, comprising: The feature extraction unit is used to obtain real-time and historical process parameters of the dyeing and printing process, determine the parameters of the membership function based on the characteristics of the current batch of materials, and use the membership function to fuzzify the real-time process parameters to obtain fuzzy process features. The state acquisition unit is used to calculate the activation contribution of each fuzzy rule based on the historical prediction accuracy of the fuzzy rules and the correlation between process benefits, and to obtain the process coordination state using the activation contribution of the fuzzy rules and fuzzy process characteristics. The parameter adjustment unit is used to input sensor timing data and process coordination status into the multimodal timing main capsule layer to obtain the main capsule output vector, calculate the confidence level of the vector, and adjust the influence of the vector when routing to the higher capsule layer according to the confidence level. The higher capsule layer generates a set of control parameters for subsequent processes. The optimization unit is used to acquire quality inspection data of subsequent processes, obtain a multi-objective deviation vector based on the quality inspection data and the set of control parameters, and optimize the high-level capsule layer using the multi-objective deviation vector to obtain the adjusted and optimized set of control parameters.

[0011] Optionally, determining the parameters of the membership function based on the characteristics of the current batch of materials includes: Obtain the material feature vector of the current batch of materials, input the material feature vector into the radial basis function network, and obtain the center point value and standard deviation value of the Gaussian membership function of the preset fuzzy subset for each process parameter.

[0012] Optionally, the step of using membership functions to fuzzify real-time process parameters to obtain fuzzy process features includes: Obtain the Gaussian membership function of the fuzzy subset corresponding to each process parameter, as well as the corresponding center point value and standard deviation value; Substitute the numerical values ​​of the process parameters into the Gaussian membership function of each corresponding fuzzy subset. The membership degree values ​​of the process parameters for each fuzzy subset are obtained, where x is the parameter value, c is the center point value, and σ is the standard deviation. All membership values ​​of each process parameter constitute the fuzzy process feature of the process parameter.

[0013] Optionally, the step of calculating the activation contribution of each fuzzy rule based on the correlation between the historical prediction accuracy of the fuzzy rule and the process benefits includes: For each fuzzy rule in the fuzzy rule base, select all historical process batches from the database of historical process parameters whose membership product of each fuzzy variable in the condition part is greater than the preset value. For each historical process batch, the root mean square error (RMSE) between the predicted intermediate process target value defined in the result part of the fuzzy rule and the actual recorded intermediate process target value of the batch is calculated. Then, the historical prediction accuracy of the rule is AP = 1 / (1+RMSE). Obtain at least three final product quality indicators and at least two process benefit indicators corresponding to each historical process batch; calculate the Pearson correlation coefficient between the prediction accuracy of the rule and each quality indicator and benefit indicator, and use the weighted average of each Pearson correlation coefficient as the process benefit correlation index CB; According to the formula The activation contribution of the fuzzy rule is calculated, where , The preset coefficients, and .

[0014] Optionally, obtaining the process cooperative state using the activation contribution of fuzzy rules and fuzzy process features includes: For each fuzzy rule, the fuzzy process features are substituted into the condition part of the rule, and the total activation intensity of the condition part of the rule is calculated using the algebraic product T-norm operator. The total activation intensity of the rule is multiplied by the activation contribution degree ACD of the rule to obtain the weighted activation intensity of the rule. Multiply the weighted activation intensity of each rule by the feature value of the process coordination state defined in the fuzzy rule result part, sum the product results of all rules, and then divide by the sum of the weighted activation intensities of all rules to obtain the process coordination state at the current moment.

[0015] Optionally, calculating the confidence level of the vector includes: Within the current processing time window, which is preset or dynamically adjusted according to the characteristics of the process stage, all main capsule output vectors output by the multimodal time-series main capsule layer are obtained; the L2 norm of each main capsule output vector is calculated; and the maximum value among the L2 norms of all main capsule output vectors within the time window is taken as the reference value. The confidence level of the main capsule output vector is obtained by dividing the L2 norm of each main capsule output vector by the benchmark value.

[0016] Optionally, adjusting the influence of the vector on subsequent routing to higher-level capsule layers based on the confidence level, wherein the higher-level capsule layer generates a set of control parameters for subsequent processes, includes: For each master capsule layer, the output master capsule vector and their corresponding confidence levels : For each high-level capsule j, through the transformation matrix Main capsule vector Convert to prediction vector ; Initialize the input of high-level capsule j It is a zero vector; During the routing iteration process, the prediction vector for the current iteration round is calculated. With the current state vector of high-level capsule j The dot product is calculated and normalized across all higher-level capsules using the softmax function to obtain the connection routing coupling coefficient between each master capsule vector U_i and the higher-level capsule j. ;according to Update the input of high-level capsule j by updating the state vector of high-level capsule j through the squeezing function; After the iteration is completed, the state vectors of all high-level capsules are concatenated to obtain a feature vector, which is then input into a multilayer perceptron (MLP) to obtain a set of control parameters.

[0017] Optionally, the optimization of the high-level capsule layer using the multi-objective deviation vector to obtain the adjusted and optimized set of control parameters includes: The loss function value is calculated based on the multi-objective bias vector; the gradient of the loss function value with respect to the weight parameters of each layer of the MLP is calculated through backpropagation, and the gradient of the loss function value with respect to each element in the transformation matrix connecting the main capsule layer and the higher capsule layers is also calculated through backpropagation. The weight parameters of the MLP and the transformation matrix are updated according to the gradient; the control parameter set is regenerated using the updated MLP and transformation matrix.

[0018] This invention improves the adaptability of the fuzzy system to material diversity and processes by adjusting the parameters of the membership function and calibrating the activation contribution of fuzzy rules based on historical prediction accuracy and correlation with actual process benefits. It enhances the ability to acquire intermodal interactions by introducing a multimodal temporal master capsule layer to process heterogeneous sensor data and process coordination states, and by combining a dynamic routing mechanism based on confidence assessment to transmit information to higher-level capsule layers. The higher-level capsule layers not only generate predictive control parameters for subsequent processes but also use multi-objective deviation vectors generated by subsequent actual quality inspections to correct their own state representation and routing protocols. This achieves deep and precise control of the dyeing and printing process and optimization of the model, thereby improving the quality stability of textile dyeing and printing products, reducing resource consumption, and enhancing the intelligence and adaptability of the production process. Attached Figure Description

[0019] Figure 1 This is a flowchart of Example 1; Figure 2 This is a schematic diagram of membership at a temperature of 40 degrees Celsius. Figure 3 This is a schematic diagram of a capsule network; Figure 4 This is a schematic diagram of an optimization loop based on a multi-objective deviation vector. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] In a specific embodiment, this application proposes an intelligent method for optimizing textile printing and dyeing processes, such as... Figure 1 As shown, including: S1, obtain real-time and historical process parameters of the dyeing and printing process, determine the parameters of the membership function based on the characteristics of the current batch of materials, and use the membership function to fuzzify the real-time process parameters to obtain fuzzy process features; Collect real-time data and historically accumulated data from the dyeing and printing production line. Based on the specific properties of the current batch of fabric being processed, determine the membership function used to describe the degree of fuzziness of the process parameters. For example, for the fuzzy concept of high temperature, the center point of the membership function might be set at 95℃ for heat-resistant cotton, with a larger standard deviation; while for silk, the center point might be around 80℃, with a smaller standard deviation. The membership function transforms the precise real-time process parameter values ​​into a set of features describing their fuzzy state. For example, a pH value of 6.8 is converted to a pH value indicating a moderately acidic level of 0.7 and a weakly acidic level of 0.3.

[0022] S2, calculate the activation contribution of each fuzzy rule based on the historical prediction accuracy of the fuzzy rule and the correlation between the process benefits, and obtain the process synergy state using the activation contribution of the fuzzy rule and the fuzzy process characteristics; A rule's high accuracy in predicting process results in the past does not necessarily mean it's a good rule. Furthermore, it's crucial to determine the correlation between prediction accuracy and process benefits such as final product quality and production efficiency. For example, if a rule has historically accurately predicted the temperature of a certain intermediate step, and historical data shows that this accurate prediction has mostly resulted in better colorfastness and lower energy consumption, then this rule's activation contribution will be relatively high. Then, by combining the activation contribution with the obtained real-time fuzzy process characteristics, a comprehensive assessment of the coordination among various stages of the entire dyeing and printing process is made to determine the process synergy status.

[0023] S3, input the sensor timing data and process coordination status into the multimodal timing main capsule layer to obtain the main capsule output vector, calculate the confidence level of the vector, and adjust the influence of the vector when routing to the higher capsule layer according to the confidence level. The higher capsule layer generates a set of control parameters for subsequent processes. Data from different sensors and the overall process status are input into the multimodal time-series master capsule layer to obtain the patterns of these different data types changing over time, and output a set of master capsule output vectors. If a master capsule vector is generated based on sensor data with significant noise, its confidence level is low. The confidence level determines the influence of the vector when it is passed to higher-level capsule layers. The higher-level capsule layers use this information to generate a set of specific operational instructions, i.e., a set of control parameters, for subsequent dyeing and printing processes.

[0024] S4. Obtain quality inspection data for subsequent processes, obtain a multi-objective deviation vector based on the quality inspection data and the set of control parameters, and optimize the high-level capsule layer using the multi-objective deviation vector to obtain the adjusted and optimized set of control parameters.

[0025] By obtaining actual product quality inspection results and comparing the actual inspection data with the expected quality predicted based on control parameters, a multi-objective deviation vector containing differences in multiple aspects is obtained. For example, the color difference deviation is ΔE=0.5, and the intensity is 2N lower than expected. The deviation vector is used to provide feedback for adjusting and regulating the internal parameters of the higher-level capsule layers, enabling more accurate predictions and controls in the future, thereby outputting a set of control parameters that have been further adjusted and optimized.

[0026] In an optional embodiment, determining the parameters of the membership function based on the characteristics of the current batch of materials includes: Obtain the material feature vector of the current batch of materials, input the material feature vector into the radial basis function network, and obtain the center point value and standard deviation value of the Gaussian membership function of the preset fuzzy subset for each process parameter.

[0027] Specifically, suppose we need to process a batch of special blended fabrics. Their material feature vectors contain information such as 70% cotton content, 30% polyester content, 150D yarn denier, and 180g / m² fabric weight. The feature vectors are input into a pre-trained radial basis function (RBF) network. Based on the input material characteristic combinations, the network outputs optimal membership function parameters for each fuzzy subset of subsequent process parameters. For example, for this batch of blended fabrics, the RBF network might output a Gaussian membership function with a center point value of 92℃ and a standard deviation of 3℃ for the "high temperature" fuzzy subset of dyeing temperature; while for the long dyeing time subset, the center point value of its Gaussian membership function is determined to be 55 minutes, with a standard deviation of 5 minutes. Furthermore, if we switch to a batch of thicker pure cotton fabrics with a larger weight, after inputting their material feature vectors into the same RBF network, the RBF network outputs a center point value of 98℃ and a standard deviation of 2℃ for the high temperature subset of dyeing temperature; and a center point value of 70 minutes and a standard deviation of 4 minutes for the long dyeing time subset. Furthermore, it can identify that thick pure cotton fabrics require higher dyeing temperatures and longer dyeing times. Determining membership functions based on material characteristic parameters allows for flexible definition, avoids the subjectivity of manually setting parameters, and enables the entire optimization method to be applied to various types of textile printing and dyeing.

[0028] In an optional embodiment, the step of using membership functions to fuzzify real-time process parameters to obtain fuzzy process features includes: Obtain the Gaussian membership function of the fuzzy subset corresponding to each process parameter, as well as the corresponding center point value and standard deviation value; Substitute the numerical values ​​of the process parameters into the Gaussian membership function of each corresponding fuzzy subset. The membership degree values ​​of the process parameters for each fuzzy subset are obtained, where x is the parameter value, c is the center point value, and σ is the standard deviation. All membership values ​​of each process parameter constitute the fuzzy process feature of the process parameter.

[0029] Each real-time process parameter requiring fuzzification is predefined with a set of fuzzy subsets describing its different state ranges. For example, for dyeing temperature, three fuzzy subsets are defined: low temperature, medium temperature, and high temperature. Each such fuzzy subset is described by a Gaussian membership function, which is determined by two key parameters: the center point value c and the standard deviation σ. The center point value represents the most typical value of the fuzzy subset; for example, the center point of the medium temperature might be 70℃. The standard deviation determines the range of parameter values ​​near the center point that still belong to this subset.

[0030] Once the specific real-time process parameter values ​​are obtained, such as the current dyeing temperature of 72℃, the value x=72 is substituted sequentially into the Gaussian membership function formula for each fuzzy subset corresponding to the dyeing temperature parameter. This yields the membership values ​​of the current temperature 72℃ for the three fuzzy subsets: low temperature, medium temperature, and high temperature. For example, the calculated membership value for low temperature is 0.05, for medium temperature is 0.8, and for high temperature is 0.2. The membership values ​​of all fuzzy subsets corresponding to a process parameter are combined to form the fuzzy process feature [0.05, 0.8, 0.2]. Figure 2 The membership degree at a temperature of 40 degrees Celsius is shown.

[0031] In an optional embodiment, calculating the activation contribution of each fuzzy rule based on the correlation between the historical prediction accuracy of the fuzzy rules and the process benefits includes: For each fuzzy rule in the fuzzy rule base, select all historical process batches from the database of historical process parameters whose membership product of each fuzzy variable in the condition part is greater than the preset value. For each historical process batch, the root mean square error (RMSE) between the predicted intermediate process target value defined in the result portion of the fuzzy rule and the actual recorded intermediate process target value for that batch is calculated. Then, the historical prediction accuracy of the rule is... ; Obtain at least three final product quality indicators and at least two process benefit indicators corresponding to each historical process batch; calculate the Pearson correlation coefficient between the prediction accuracy of the rule and each quality indicator and benefit indicator, and use the weighted average of each Pearson correlation coefficient as the process benefit correlation index CB; According to the formula The activation contribution of the fuzzy rule is calculated, where , The preset coefficients, and .

[0032] For each historical production batch, determine the extent to which the process parameters at that time satisfied the conditional part of the rule. If the product of these membership degrees exceeds a preset value, then this historical batch is highly correlated with the premise of the rule. For all filtered historical batches, based on the prediction effect of the rule's conclusion at that time, calculate the corresponding value between the intermediate process target value predicted by the rule and the actual historical record. The smaller the error, the more accurate the rule prediction, and the higher its historical prediction accuracy.

[0033] Further analyze the actual correlation between accuracy and the quality of the final product and production efficiency, i.e., the correlation between process efficiency and efficiency. Obtain the final product quality data and process efficiency data corresponding to historical batches for which the rule predictions were accurate. Final product quality indicators include, but are not limited to, color difference, levelness grade, and color fastness; process efficiency indicators include, but are not limited to, unit energy consumption and dye utilization rate.

[0034] The overall CB value is obtained by calculating the Pearson correlation coefficient between the historical prediction accuracy (AP) of a rule and quality and efficiency indicators, and then performing a weighted average. If a rule has a high AP value, historical data shows that the product's color fastness is generally high, and the unit energy consumption is low; therefore, the CB value of this rule will be relatively high. This is achieved through a weighted formula. By combining historical prediction accuracy (AP) and process benefit correlation (CB), the final activation contribution (ACD) of the fuzzy rule is obtained. Rules that ensure accurate predictions and deliver better quality and higher production efficiency are given greater weight and influence in current process optimization decisions.

[0035] In an optional embodiment, obtaining the process coordination state using the activation contribution of fuzzy rules and fuzzy process features includes: For each fuzzy rule, the fuzzy process features are substituted into the condition part of the rule, and the total activation intensity of the condition part of the rule is calculated using the algebraic product T-norm operator. The total activation intensity of the rule is multiplied by the activation contribution degree ACD of the rule to obtain the weighted activation intensity of the rule. Multiply the weighted activation intensity of each rule by the feature value of the process coordination state defined in the fuzzy rule result part, sum the product results of all rules, and then divide by the sum of the weighted activation intensities of all rules to obtain the process coordination state at the current moment.

[0036] For example, one rule states that if the temperature is high and the pH is low, the dyeing rate tends to be fast, and the activation contribution coefficient (ACD) of this rule is calculated to be 0.9 based on historical performance. Substituting the fuzzy characteristics of the current process parameters into the condition part of the rule, the algebraic product operator is used to obtain the original activation intensity of this rule triggered by the current conditions. Adding 0.8 for high temperature and 0.7 for low pH, the original activation intensity is 0.56. Multiplying the original activation intensity by the rule's ACD yields a value of 0.504, which is the weighted activation intensity of this rule in relation to the current overall state. This approach considers not only whether the rule is triggered by the current situation but also its historical effectiveness and actual contribution to process efficiency. The result part of each rule corresponds to a preset characteristic value of the process synergy state; for example, fast corresponds to a synergy state value of 8, medium to 5, and slow to 2. The weighted activation intensity of each rule is multiplied by the characteristic value of the process synergy state it points to. For example, if the weighted activation strength of rule A is 0.504, and its result points to the cooperative state characteristic value 8, then the product is 4.032; if the weighted activation strength of rule B is 0.3, and its result points to the cooperative state characteristic value 5, then the product is 1.5. Summing up these products of all rules and then dividing by the sum of the weighted activation strengths of all rules yields the current value of the process cooperative state.

[0037] In an optional embodiment, calculating the confidence level of the vector includes: Within the current processing time window, which is preset or dynamically adjusted according to the characteristics of the process stage, all main capsule output vectors output by the multimodal time-series main capsule layer are obtained; the L2 norm of each main capsule output vector is calculated; and the maximum value among the L2 norms of all main capsule output vectors within the time window is taken as the reference value. The confidence level of the main capsule output vector is obtained by dividing the L2 norm of each main capsule output vector by the benchmark value.

[0038] Specifically, assuming that in the fixation stage of textile dyeing, the current processing time window is determined to be, for example, the past 60 seconds based on the current fixation rate feedback. Within these 60 seconds, the multimodal temporal master capsule layer outputs multiple master capsule vectors, which represent short-term dynamic features extracted from temporal data from different sensors. For example, one master capsule vector might characterize the stability of the temperature curve, while another might characterize the trend of rapid decrease in dye concentration. The L2 norm of each such vector is calculated; for example, the L2 norm of the temperature stability vector is 0.8, the L2 norm of the dye concentration decrease trend vector is 1.5, and the L2 norm of other vectors may fall between these two. The maximum value of the L2 norm among all master capsule output vectors within these 60 seconds, for example, 1.5, is found and used as the baseline value. The confidence level of each master capsule output vector is obtained by dividing the L2 norm of each vector by the baseline value of 1.5. The confidence level of the dye concentration decrease trend vector is 1.5 / 1.5 = 1.0, while the confidence level of the temperature stability vector is 0.53. Figure 3 The relationship between the main capsule layer and the upper capsule layers is shown.

[0039] In an optional embodiment, adjusting the influence of the vector on subsequent routing to higher-level capsule layers based on the confidence level, wherein the higher-level capsule layer generates a set of control parameters for subsequent processes, includes: For each master capsule layer, the output master capsule vector and their corresponding confidence levels : For each high-level capsule j, through the transformation matrix Main capsule vector Convert to prediction vector ; Initialize the input of high-level capsule j It is a zero vector; During the routing iteration process, the prediction vector for the current iteration round is calculated. With the current state vector of high-level capsule j The dot product is then normalized across all higher-level capsules using the softmax function to obtain the vector for each master capsule. The connection routing coupling coefficient between the high-level capsule j and the high-level capsule j ;according to Update the input of high-level capsule j by updating the state vector of high-level capsule j through the squeezing function; After the iteration is completed, the state vectors of all high-level capsules are concatenated to obtain a feature vector, which is then input into a multilayer perceptron (MLP) to obtain a set of control parameters.

[0040] Specifically, the process involves obtaining the master capsule vectors representing low-level features and their respective confidence levels. Each master capsule vector, through its corresponding transformation matrix, generates a prediction for each high-level capsule. Then, through an iterative routing mechanism, the contribution weights of the master capsules to the high-level capsules are assigned based on the matching degree between the master capsule's prediction and the current state of the high-level capsules, as well as the master capsule's own confidence level. This weighted and iterative process allows the high-level capsules to effectively integrate information from multiple master capsules, with master capsules having higher confidence levels having a greater impact. After iteration, the state vectors of all high-level capsules are concatenated into a single feature vector. This feature vector is input into a multilayer perceptron network, which maps the integrated features into a set of specific control parameters for subsequent processes.

[0041] In an optional embodiment, the optimization of the high-level capsule layer using the multi-objective deviation vector to obtain an adjusted and optimized set of control parameters includes: The loss function value is calculated based on the multi-objective bias vector; the gradient of the loss function value with respect to the weight parameters of each layer of the MLP is calculated through backpropagation, and the gradient of the loss function value with respect to each element in the transformation matrix connecting the main capsule layer and the higher capsule layers is also calculated through backpropagation. The weight parameters of the MLP and the transformation matrix are updated according to the gradient; the control parameter set is regenerated using the updated MLP and transformation matrix.

[0042] Specifically, the effect of the previously generated control parameter set after practical application is calculated, i.e., the multi-objective deviation vector between the quality inspection data and the expected target. Based on this multi-objective deviation vector, the loss function value is calculated, and the internal weights of the multilayer perceptron (MLP) and the transformation matrix connecting the main capsule layer and higher-level capsule layers are updated using the backpropagation algorithm. According to gradient information, the weights and transformation matrix of the MLP are fine-tuned to reduce the loss function. After completing the parameter update, a better control parameter set is regenerated using the optimized MLP and transformation matrix, such as... Figure 4 As shown.

[0043] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. In addition, various different implementations of the embodiments of the present invention can be arbitrarily combined, as long as they do not violate the spirit of the embodiments of the present invention, and they should also be regarded as the content disclosed in the embodiments of the present invention.

Claims

1. A method for optimizing an intelligent textile printing and dyeing process, characterized in that, include: Obtain real-time and historical process parameters of the dyeing and printing process, determine the parameters of the membership function based on the characteristics of the current batch of materials, and use the membership function to fuzzify the real-time process parameters to obtain fuzzy process features; The activation contribution of each fuzzy rule is calculated based on the correlation between the historical prediction accuracy of the fuzzy rules and the process benefits. The process synergy status is obtained by using the activation contribution of the fuzzy rules and the fuzzy process characteristics. Sensor timing data and process coordination status are input into the multimodal timing master capsule layer to obtain the master capsule output vector. The confidence level of the vector is calculated, and the influence of the vector in subsequent routing to the higher capsule layer is adjusted according to the confidence level. The higher capsule layer generates a set of control parameters for subsequent processes. The quality inspection data of subsequent processes is obtained, and a multi-objective deviation vector is obtained based on the quality inspection data and the set of control parameters. The multi-objective deviation vector is then used to optimize the high-level capsule layer to obtain the adjusted and optimized set of control parameters.

2. The method according to claim 1, characterized in that, The parameters for determining the membership function based on the characteristics of the current batch of materials include: Obtain the material feature vector of the current batch of materials, input the material feature vector into the radial basis function network, and obtain the center point value and standard deviation value of the Gaussian membership function of the preset fuzzy subset for each process parameter.

3. The method according to claim 1, characterized in that, The process of obtaining fuzzy process features by fuzzifying real-time process parameters using membership functions includes: Obtain the Gaussian membership function of the fuzzy subset corresponding to each process parameter, as well as the corresponding center point value and standard deviation value; Substitute the numerical values ​​of the process parameters into the Gaussian membership function of each corresponding fuzzy subset. The membership degree values ​​of the process parameters for each fuzzy subset are obtained, where x is the parameter value, c is the center point value, and σ is the standard deviation. All membership values ​​of each process parameter constitute the fuzzy process feature of the process parameter.

4. The method according to claim 1, characterized in that, The calculation of the activation contribution of each fuzzy rule based on the correlation between the historical prediction accuracy of the fuzzy rule and the process benefits includes: For each fuzzy rule in the fuzzy rule base, select all historical process batches from the database of historical process parameters whose membership product of each fuzzy variable in the condition part is greater than the preset value. For each historical process batch, the root mean square error (RMSE) between the predicted intermediate process target value defined in the result portion of the fuzzy rule and the actual recorded intermediate process target value for that batch is calculated. Then, the historical prediction accuracy of the rule is... ; Obtain at least three final product quality indicators and at least two process benefit indicators corresponding to each historical process batch; calculate the Pearson correlation coefficient between the prediction accuracy of the rule and each quality indicator and benefit indicator, and use the weighted average of each Pearson correlation coefficient as the process benefit correlation index CB; According to the formula The activation contribution of the fuzzy rule is calculated, where , The preset coefficients, and .

5. The method according to claim 1, characterized in that, The process coordination state obtained by utilizing the activation contribution of fuzzy rules and fuzzy process features includes: For each fuzzy rule, the fuzzy process features are substituted into the condition part of the rule, and the total activation intensity of the condition part of the rule is calculated using the algebraic product T-norm operator. The total activation intensity of the rule is multiplied by the activation contribution degree ACD of the rule to obtain the weighted activation intensity of the rule. Multiply the weighted activation intensity of each rule by the feature value of the process coordination state defined in the fuzzy rule result part, sum the product results of all rules, and then divide by the sum of the weighted activation intensities of all rules to obtain the process coordination state at the current moment.

6. The method according to claim 1, characterized in that, The calculation of the confidence level of the vector includes: Within the current processing time window, which is preset or dynamically adjusted according to the characteristics of the process stage, all main capsule output vectors output by the multimodal time-series main capsule layer are obtained; the L2 norm of each main capsule output vector is calculated; and the maximum value among the L2 norms of all main capsule output vectors within the time window is taken as the reference value. The confidence level of the main capsule output vector is obtained by dividing the L2 norm of each main capsule output vector by the benchmark value.

7. The method according to claim 1, characterized in that, The step of adjusting the influence of the vector on subsequent routing to higher-level capsule layers based on the confidence level, wherein the higher-level capsule layer generates a set of control parameters for subsequent processes, including: For each master capsule layer, the output master capsule vector and their corresponding confidence levels : For each high-level capsule j, through the transformation matrix Main capsule vector Convert to prediction vector ; Initialize the input of high-level capsule j It is a zero vector; During the routing iteration process, the prediction vector for the current iteration round is calculated. With the current state vector of high-level capsule j The dot product is then normalized across all higher-level capsules using the softmax function to obtain the vector for each master capsule. The connection routing coupling coefficient between the high-level capsule j and the high-level capsule j ;according to Update the input of high-level capsule j by updating the state vector of high-level capsule j through the squeezing function; After the iteration is completed, the state vectors of all high-level capsules are concatenated to obtain a feature vector, which is then input into a multilayer perceptron (MLP) to obtain a set of control parameters.

8. The method according to claim 1, characterized in that, The optimization of the high-level capsule layer using the multi-objective deviation vector yields an adjusted and optimized set of control parameters, including: The loss function value is calculated based on the multi-objective bias vector; the gradient of the loss function value with respect to the weight parameters of each layer of the MLP is calculated through backpropagation, and the gradient of the loss function value with respect to each element in the transformation matrix connecting the main capsule layer and the higher capsule layers is also calculated through backpropagation. The weight parameters of the MLP and the transformation matrix are updated according to the gradient; the control parameter set is regenerated using the updated MLP and transformation matrix.

9. An intelligent textile printing and dyeing process optimization system, characterized in that, include: The feature extraction unit is used to obtain real-time and historical process parameters of the dyeing and printing process, determine the parameters of the membership function based on the characteristics of the current batch of materials, and use the membership function to fuzzify the real-time process parameters to obtain fuzzy process features. The state acquisition unit is used to calculate the activation contribution of each fuzzy rule based on the historical prediction accuracy of the fuzzy rules and the correlation between process benefits, and to obtain the process coordination state using the activation contribution of the fuzzy rules and fuzzy process characteristics. The parameter adjustment unit is used to input sensor timing data and process coordination status into the multimodal timing main capsule layer to obtain the main capsule output vector, calculate the confidence level of the vector, and adjust the influence of the vector when routing to the higher capsule layer according to the confidence level. The higher capsule layer generates a set of control parameters for subsequent processes. The optimization unit is used to acquire quality inspection data of subsequent processes, obtain a multi-objective deviation vector based on the quality inspection data and the set of control parameters, and optimize the high-level capsule layer using the multi-objective deviation vector to obtain the adjusted and optimized set of control parameters.

10. The system according to claim 9, characterized in that, The parameters for determining the membership function based on the characteristics of the current batch of materials include: Obtain the material feature vector of the current batch of materials, input the material feature vector into the radial basis function network, and obtain the center point value and standard deviation value of the Gaussian membership function of the preset fuzzy subset for each process parameter.

Images