Surface antistatic treatment method and system for polyester knitted fabric

Abstract

The application discloses a surface antistatic treatment method and system for polyester knitted fabric, and relates to the technical field of polyester knitted fabric.The method comprises the following steps: according to a treatment process scheme determined according to a target application scenario, calling a sample process record, and constructing a standard process training set; based on the standard process training set, combining a knowledge distillation mechanism, constructing and training an antistatic performance prediction model; constructing a cost evaluation factor, combining the antistatic performance prediction model and the cost evaluation factor, and defining a process evaluation function, wherein the process evaluation function is associated with a penalty space; taking the process evaluation function as an optimization target, iteratively optimizing self-adaptive expansion and contraction of treatment process parameters, and determining a target process parameter combination for the polyester knitted fabric according to an iterative optimization result to perform surface antistatic treatment. In this way, the technical effects of reducing parameter adjustment cost, improving adjustment efficiency and enhancing antistatic treatment flexibility are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of polyester knitted fabric, and particularly relates to a surface antistatic treatment method and system for polyester knitted fabric. BACKGROUND

[0002] The polyester knitted fabric is widely used in many fields such as clothing, home textiles and industrial fabric due to its excellent performance. In actual use, the polyester knitted fabric is prone to static electricity, which not only affects the use experience, but also may cause safety hazards in some specific scenarios. Therefore, the surface antistatic treatment is needed. At present, the surface antistatic treatment method for the polyester knitted fabric mainly relies on the traditional empirical process adjustment. For example, based on fixed process parameters, small-scale parameter adjustment is made by manual experience, which is difficult to flexibly adjust according to actual production data and specific application scene requirements, resulting in inaccurate process parameter optimization and being unable to effectively meet the antistatic requirements in different situations. Although some existing methods introduce optimization algorithms, it is difficult to balance between global exploration and local convergence when dealing with complex process parameter optimization problems, resulting in long optimization time and low precision. SUMMARY

[0003] The present application provides a surface antistatic treatment method and system for polyester knitted fabric to solve the technical problems of high cost and low efficiency of parameter adjustment in the prior art, and achieve the technical effects of reducing parameter adjustment cost, improving adjustment efficiency and enhancing antistatic treatment flexibility.

[0004] In a first aspect, the present application provides a surface antistatic treatment method for polyester knitted fabric, wherein the surface antistatic treatment method for polyester knitted fabric comprises:

[0005] According to the processing process scheme determined by the target application scene, a sample process record is called to construct a standard process training set.

[0006] Based on the standard process training set, an antistatic performance prediction model is constructed and trained in combination with a knowledge distillation mechanism.

[0007] A cost evaluation factor is constructed, and a process evaluation function is defined in combination with the antistatic performance prediction model and the cost evaluation factor, wherein the process evaluation function is associated with a penalty space.

[0008] The processing process parameters are iteratively optimized by self-adaptive expansion and contraction, and the target process parameter combination for the polyester knitted fabric is determined according to the iteration optimization result to perform surface antistatic treatment.

[0009] In a feasible implementation manner, the process evaluation function is defined in combination with the antistatic performance prediction model and the cost evaluation factor, comprising:

[0010] An antistatic treatment task sheet is obtained and parsed to extract a compliance constraint.

[0011] The penalty space of the process evaluation function is defined according to the compliance constraint.

[0012] The output of the antistatic performance prediction model and the cost evaluation factor are weighted and fused to obtain the process evaluation function, and the process evaluation function is associated with the penalty space.

[0013] In a feasible implementation, the processing process parameters are iteratively optimized for adaptive expansion and contraction with the process evaluation function as the optimization target, including:

[0014] Based on the standard process training set, a process parameter particle swarm containing multiple candidate process parameter combinations is initialized.

[0015] According to the initialized process parameter particle swarm, an iterative optimization process based on a particle swarm optimization algorithm is performed, the optimization direction of each particle is calculated in each iteration process, and the bidirectional virtual intersection point of the particle in the search space is identified according to the optimization direction.

[0016] According to the bidirectional virtual intersection point identification result, the process evaluation function, the process parameter particle swarm is adaptively expanded and contracted.

[0017] According to the preset iteration constraint condition, the iteration optimization result is output.

[0018] In a feasible implementation, the process parameter particle swarm is adaptively expanded and contracted in combination with the process evaluation function, further including:

[0019] When the optimization direction of the first preset number of particles in the preset range has a common virtual intersection point in the multi-dimensional parameter space, the process evaluation function value corresponding to the current position of each particle is obtained, and the fusion weight is defined correspondingly.

[0020] Based on the fusion weight, the first preset number of particles are weighted and fused, and the first fusion process evaluation function value corresponding to the weighted fusion result is calculated based on the process evaluation function.

[0021] The virtual intersection point process evaluation function value at the common virtual intersection point is determined based on the process evaluation function value.

[0022] The spatial point corresponding to the larger value between the first fusion process evaluation function value and the virtual intersection point process evaluation function value is compared and selected as a new particle and added to the process parameter particle swarm.

[0023] In a feasible implementation, the adaptive expansion and contraction of the process parameter particle swarm in combination with the process evaluation function further includes:

[0024] In the multi-dimensional parameter space, a reverse virtual intersection point formed by the intersection of the reverse extensions of the second preset number of optimization directions is identified.

[0025] When the reverse virtual intersection point is located within a preset range, the angle variance of adjacent angles formed by the second preset number of optimization directions is calculated.

[0026] Based on the angle variance, the number of new particles to be inserted is determined, and new particles are inserted on the central axes of the adjacent angles of the second preset number of optimization directions in descending order of adjacent angles.

[0027] In a feasible implementation, the adaptive expansion and contraction of the process parameter particle swarm in combination with the process evaluation function further includes:

[0028] A dynamic fluctuation range of the particle swarm size is set, and the adaptive regulation of the particle swarm size is performed within the dynamic fluctuation range based on the initial particle swarm size.

[0029] If the current particle swarm size is lower than the lower limit of the fluctuation, the adaptive contraction is suspended until the current particle swarm size is higher than the initial particle swarm size.

[0030] If the current particle swarm size is higher than the upper limit of the fluctuation, the adaptive expansion is suspended until the current particle swarm size is lower than the initial particle swarm size.

[0031] In a feasible implementation, according to the processing process scheme determined according to the target application scenario, a sample process record is called to construct a standard process training set, including:

[0032] Sample process parameter data is obtained, wherein the process parameters at least include the proportion of auxiliary agents, processing temperature, processing time, and fabric moisture content.

[0033] Sample antistatic performance data associated with the sample process parameter data is obtained, wherein the antistatic performance at least includes static voltage, charge surface density, and half-life.

[0034] The sample process parameter data and the sample antistatic performance data are preprocessed and target value calibrated, and are divided into positive samples and negative samples according to the target value calibration result, and are output as the standard process training set.

[0035] In a feasible implementation, based on the standard process training set, an antistatic performance prediction model is constructed and trained in combination with a knowledge distillation mechanism, including:

[0036] The benchmark prediction model based on the deep neural network is constructed and trained through the standard process training set.

[0037] Based on the knowledge distillation mechanism, the prediction output of the benchmark prediction model is combined with the standard process training set to train a simplified prediction model.

[0038] The accuracy of the simplified prediction model is verified, and if the positive samples and negative samples in the standard process training set both meet the set performance evaluation threshold, the simplified prediction model is output and deployed as the antistatic performance prediction model.

[0039] In a feasible implementation manner, the cost evaluation factor at least includes a processing cost item per unit fabric, a processing energy consumption item, a processing time item, and a resource consumption item.

[0040] In a second aspect, the application also provides a surface antistatic treatment system for polyester knitted fabric, wherein the surface antistatic treatment system for polyester knitted fabric comprises:

[0041] The training set construction module is configured to determine a processing process scheme according to a target application scenario, call a sample process record, and construct a standard process training set.

[0042] The prediction model training module is configured to construct and train an antistatic performance prediction model based on the standard process training set and in combination with a knowledge distillation mechanism.

[0043] The evaluation function definition module is configured to construct a cost evaluation factor, and define a process evaluation function in combination with the antistatic performance prediction model and the cost evaluation factor, wherein the process evaluation function is associated with a penalty space.

[0044] The process parameter optimization module is configured to perform iterative optimization of self-adaptive expansion and contraction of the processing process parameters with the process evaluation function as an optimization target, determine a target process parameter combination for the polyester knitted fabric according to an iterative optimization result, and perform surface antistatic treatment.

[0045] The application discloses a polyester knitted fabric surface antistatic treatment method and system, which comprises the following steps: determining a treatment process scheme based on a target application scenario, calling historical sample process records, and constructing a standardized process training dataset; based on the training dataset, a machine learning model for antistatic performance prediction is established and trained in combination with a knowledge distillation mechanism; a cost-related evaluation factor is constructed, and the antistatic performance prediction model is combined with the cost evaluation factor to set a process evaluation function containing a penalty term, which is used to quantify the comprehensive effect of the treatment process; the process evaluation function is used as an optimization target, and an adaptive expansion and contraction strategy is adopted to iteratively optimize the treatment process parameters, so as to finally determine the target process parameter combination for the surface antistatic treatment of the polyester knitted fabric. The polyester knitted fabric surface antistatic treatment method and system disclosed by the application solve the technical problems of high treatment parameter adjustment cost and low adjustment efficiency, and achieve the technical effects of reducing parameter adjustment cost, improving adjustment efficiency and enhancing antistatic treatment flexibility. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 It is a flowchart of the polyester knitted fabric surface antistatic treatment method.

[0047] Figure 2 It is a structural schematic diagram of the polyester knitted fabric surface antistatic treatment system.

[0048] Mark explanation: training set construction module 11, prediction model training module 12, evaluation function definition module 13, process parameter optimization module 14. DETAILED DESCRIPTION

[0049] The above technical solutions will be described in detail below in combination with the drawings and specific embodiments, so that the above technical solutions can be better understood. Obviously, the described embodiments are only part of the embodiments of the application, not all embodiments of the application, and it should be understood that the application is not limited to the example embodiments for explaining the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application. In addition, it should be noted that, for convenience of description, only parts related to the application are shown in the drawings, not all.

[0050] Embodiment one, as Figure 1 It is a flowchart of the polyester knitted fabric surface antistatic treatment method, wherein the polyester knitted fabric surface antistatic treatment method comprises the following steps:

[0051] S100: According to the treatment process scheme determined according to the target application scenario, calling sample process records, and constructing a standard process training set.

[0052] Specifically, the target application scenario refers to the end use of the polyester knitted fabric (such as clothing, home textiles, industrial fabrics, etc.), and based on the target application scenario, specific requirements for anti-static performance, hand feeling, color fastness, etc. can be determined.

[0053] Specifically, different target application scenarios correspond to different types of processes, such as surface coating method, chemical modification method, and conductive fiber embedding. In addition, different processing schemes involve different numbers and types of process parameters, including but not limited to, for example, antistatic agent type, addition ratio, padding process parameters, drying temperature and time, etc.

[0054] Specifically, the sample process record is the process parameter verified effective in the historical production process and its corresponding performance test data. For example, obtain the processing records of all polyester knitted fabrics produced in the past year, including different temperature, time, additive type and amount, etc. parameter combination and corresponding surface resistivity (index for measuring anti-static performance) data, from which the sample that meets the requirements of the target application scenario is selected, such as the surface coating treatment record in the clothing field, which is constructed into a standard process training set.

[0055] Through the above process, accurate and targeted data basis can be provided for subsequent anti-static performance prediction models. In other words, the standard process training set constructed closely around the target application scenario ensures that the rules learned by the model match the actual application requirements.

[0056] In some embodiments, according to the processing scheme determined according to the target application scenario, the sample process record is called to construct a standard process training set, including:

[0057] Obtain sample process parameter data, wherein the process parameters at least include additive ratio, treatment temperature, treatment time and fabric moisture content; obtain sample anti-static performance data associated with the sample process parameter data, wherein the anti-static performance at least includes static voltage, charge surface density and half-life; pre-process and target value calibration are performed on the sample process parameter data and the sample anti-static performance data, and are divided into positive samples and negative samples according to the target value calibration result, and the output is the standard process training set.

[0058] Specifically, the sample process parameter data refers to the process operation parameters actually collected or historically recorded in the surface anti-static treatment process of the polyester knitted fabric, mainly including additive ratio (such as the ratio of antistatic agent to water), treatment temperature, treatment time and fabric moisture content, etc. The sample anti-static performance data includes static voltage, charge surface density and half-life, which are used to measure the anti-static effect of the polyester knitted fabric.

[0059] Specifically, first, the target anti-static performance value is determined according to the target application scenario (such as clothing, home textiles, etc.). Then, the historical sample process records are retrieved, the relevant process parameter data and corresponding anti-static performance data are extracted, and the data is standardized (such as removing outliers, normalization, etc.). Then, according to the preset performance threshold, the samples are labeled, and the static voltage is not higher than a certain set value, the charge surface density is lower than the specified upper limit, and the half-life is less than the specified upper limit as positive samples (reach the target performance), otherwise as negative samples (not reach the target performance). Finally, the dataset containing positive and negative samples is output as the standard process training set.

[0060] For example, if the fabric has a static voltage of 350V (lower than the target 400V), a charge surface density of 0.7μC / m 2 (lower than the target 1.0μC / m 2 ), and a half-life of 0.510s (less than the target 2s) under a certain set of process parameters, the sample is labeled as a positive sample.

[0061] Through the above process, a standard process training set that meets the target application requirements can be systematically constructed, improving the scientificity and accuracy of process development, providing strong data support for parameter optimization of polyester knitted fabric surface anti-static treatment process, and thus improving the consistency, stability and production efficiency of products.

[0062] S200: Based on the standard process training set, an anti-static performance prediction model is constructed and trained in combination with a knowledge distillation mechanism.

[0063] In some embodiments, based on the standard process training set, an anti-static performance prediction model is constructed and trained in combination with a knowledge distillation mechanism, including:

[0064] A benchmark prediction model based on a deep neural network is constructed and trained through the standard process training set; a simplified prediction model is trained based on the knowledge distillation mechanism, combining the prediction output of the benchmark prediction model and the standard process training set; the accuracy of the simplified prediction model is verified, and if the positive and negative samples in the standard process training set both meet the set performance evaluation threshold, the simplified prediction model is output and deployed as the anti-static performance prediction model.

[0065] Specifically, the knowledge distillation mechanism is used to migrate the knowledge in the complex and high-performance benchmark prediction model (usually a deep neural network model) to a simplified prediction model with a simpler structure and higher computational efficiency. The benchmark prediction model refers to a deep neural network model trained based on the complete standard process training set, which has high anti-static performance prediction accuracy. The simplified prediction model refers to a model with smaller structure and faster reasoning speed obtained through knowledge distillation technology, which is convenient for rapid application in actual deployment.

[0066] Specifically, first, the process parameters and antistatic performance data in the standard process training set are used to train a benchmark prediction model based on a deep neural network (including, for example, a multilayer perceptron, a convolutional neural network, or a long short-term memory network) so that it can accurately predict the antistatic performance under given process parameters.

[0067] Further, after training is completed, knowledge distillation is used to train a simplified student model based on the prediction output (such as a probability distribution, a feature representation, or a soft label) of the benchmark prediction model as a teacher model and the real label of the standard process training set, and by designing a distillation loss function, the simplified prediction model not only fits the real label but also tries to approximate the output performance of the teacher model.

[0068] For example, the distillation loss function can use the following composite loss:

[0069] ;

[0070] wherein, is the cross-entropy loss of the real label, is the Kullback-Leibler divergence of the output of the teacher model (benchmark prediction model) and the student model (simplified prediction model), and a is a weight coefficient (such as 0.5), are the outputs of the teacher model and the student model at temperature T, respectively; and y is the real label, such as the real antistatic performance category or actual performance value of the sample; is the prediction output of the student model.

[0071] Further, the performance of the simplified prediction model is evaluated on the positive and negative samples in the standard process training set. If the prediction accuracy, recall rate, and other indicators all meet the preset threshold, it is considered that the performance of the simplified prediction model meets the requirements, and the simplified prediction model can be output as the final antistatic performance prediction model for actual deployment.

[0072] Through the above process, the inference efficiency and deployment flexibility of the model can be significantly improved while ensuring the accuracy of the antistatic performance prediction, which helps to realize real-time process parameter optimization in actual production environment and further improve the automation and intelligent level of the antistatic treatment process of polyester knitted fabric.

[0073] S300: Construct a cost evaluation factor, and define a process evaluation function in combination with the antistatic performance prediction model and the cost evaluation factor, wherein the process evaluation function is associated with a penalty space.

[0074] Specifically, the cost evaluation factor is used to quantify the index of the economic aspect of the anti-static treatment process of the polyester knitted fabric, so as to facilitate the quantitative comparison of the cost of different process schemes. For example, the cost evaluation factor can include but is not limited to: the cost of the amount of auxiliary agent, such as the market price of the auxiliary agent required per square meter of fabric multiplied by the actual amount; the cost of energy consumption, such as the power consumption cost corresponding to the treatment temperature and time; the cost of water, such as the consumption of water resources in the treatment process; the cost of equipment depreciation and labor, such as the equipment depreciation and labor cost allocated for each batch of production.

[0075] For example, if the amount of auxiliary agent in a single batch is 0.8%, the unit price of the auxiliary agent is 100 yuan / kg, and 8g of auxiliary agent is used per square meter of fabric, then the cost of the auxiliary agent is 0.8 yuan / square meter; the treatment temperature is 140℃, the electricity consumption is 0.5 degrees, and the electricity price is 0.8 yuan / degree, then the energy consumption cost is 0.4 yuan / square meter; the water consumption is 0.2 tons, the water price is 3 yuan / ton, and the water cost is 0.6 yuan / square meter; the total cost of equipment depreciation and labor is 0.5 yuan / square meter. Then the total cost is 2.3 yuan / square meter.

[0076] By combining the cost evaluation process evaluation function, the comprehensive trade-off between anti-static performance and economic cost of the process scheme can be realized, and the recommended process can have excellent anti-static effect and consider production cost control. Through the punishment space mechanism, the low-efficiency or unusable process parameter combination that does not meet the performance requirements can be further avoided, and the practicability and reliability of process optimization and intelligent recommendation are improved.

[0077] In some embodiments, the cost evaluation factor at least includes a treatment cost item per unit fabric, a treatment energy consumption item, a treatment time item, and a resource consumption item.

[0078] Preferably, the cost evaluation factor at least includes the following sub-items:

[0079] The treatment cost item per unit fabric refers to the direct economic input required to complete the anti-static treatment of a unit area (or unit weight) of fabric, including auxiliary agent cost, labor cost, etc. The treatment energy consumption item refers to the energy cost consumed in the process, such as electricity, steam, etc. The treatment time item refers to the actual time required to complete a batch of anti-static treatment, which reflects the process efficiency and has a direct impact on production capacity. The resource consumption item refers to the cost of water, chemicals and other resources consumed in the process, which reflects the green environmental protection characteristics of the process.

[0080] In some embodiments, in combination with the anti-static performance prediction model and the cost evaluation factor, a process evaluation function is defined, including:

[0081] An antistatic treatment task sheet is obtained, and the antistatic treatment task sheet is parsed to extract a compliance constraint; the compliance constraint is used to define a penalty space of the process evaluation function; the output of the antistatic performance prediction model and the cost evaluation factor are weighted and fused to obtain the process evaluation function, and the process evaluation function is associated with the penalty space.

[0082] Specifically, first, an antistatic treatment task sheet is obtained, and the antistatic treatment task sheet is parsed to extract a compliance constraint, for example, a certain task sheet stipulates that the static voltage is less than or equal to 500V, the half-life is greater than or equal to 20s, and the unit area treatment cost is less than or equal to 3 yuan / m 2 The above constraint conditions can be correspondingly converted into a structured compliance constraint parameter set. Then, the compliance constraint is used to define the penalty space of the process evaluation function, which is used to negatively correct the process parameter combination that does not meet the above constraint. Exemplarily, the expression of the penalty space is:

[0083] ;

[0084] Among them, is the predicted static voltage, is the predicted half-life, is the predicted unit cost, and λ is a penalty coefficient (such as 1000) for amplifying the negative impact of the out-of-limit part.

[0085] Further, the output of the antistatic performance prediction model, the cost evaluation factor and the expression of the above penalty space are fused in a weighted sum manner to reduce the comprehensive score of the unqualified scheme.

[0086] Through the above process, automatic, multi-objective and constraint optimization evaluation of different antistatic treatment process parameter combinations can be realized, and the optimal comprehensive score is obtained under the condition that the recommended scheme meets the performance and cost standards.

[0087] S400: Iterative optimization of self-adaptive expansion and contraction of the process parameters is performed with the process evaluation function as the optimization target, and a target process parameter combination for polyester knitted fabric is determined according to the iterative optimization result to perform surface antistatic treatment.

[0088] Specifically, the iterative optimization of self-adaptive expansion and contraction can efficiently find the optimal process parameter combination by dynamically adjusting the particle swarm size, ensure that the optimization algorithm balances between global exploration and local convergence, avoid falling into a local optimal solution, and improve the accuracy and efficiency of optimization.

[0089] In some embodiments, the iterative optimization of self-adaptive expansion and contraction of the process parameters with the process evaluation function as the optimization target comprises:

[0090] Based on the standard process training set, a process parameter particle swarm containing a plurality of candidate process parameter combinations is initialized; based on the initialized process parameter particle swarm, an iterative optimization process based on a particle swarm optimization algorithm is performed, the optimization direction of each particle is calculated in each iteration process, and the bidirectional virtual intersection point of the particle in the search space is identified according to the optimization direction; according to the bidirectional virtual intersection point identification result, the process evaluation function, the process parameter particle swarm is adaptively expanded and scaled; according to the preset iteration constraint condition, the iteration optimization result is output.

[0091] Specifically, the process parameter particle swarm refers to a group composed of a plurality of candidate process parameter combinations in the optimization process, wherein each particle represents a group of possible process parameters. The bidirectional virtual intersection point refers to a virtual point that the particle may intersect in the search space during the particle swarm optimization process, which is identified according to the motion trend and optimization direction of the particle, and is used to judge whether the aggregation degree and optimization direction of the particle are consistent.

[0092] Specifically, first, based on the standard process training set, the initial value range of the process parameters is set, and a plurality of candidate process parameter combinations are randomly generated in the range as the initial particle swarm, and the optimization direction and speed of the particle are initialized. For example, a particle swarm containing 50 particles is initialized, each particle represents a group of process parameters such as auxiliary agent ratio, treatment temperature, treatment time and fabric moisture content, and the current fitness (i.e. process evaluation function value) is recorded.

[0093] Then, based on the initialized process parameter particle swarm, an iterative optimization process based on a particle swarm optimization algorithm is performed: in each iteration process, first, the fitness value of each particle is calculated, that is, the advantages and disadvantages of the process parameter combination corresponding to the particle are evaluated according to the process evaluation function. Then, according to the speed of the particle and the update rule of the particle swarm algorithm, the speed and position of each particle are updated.

[0094] Next, after the particle position is updated, the moving trend of the particle in the historical optimal direction or the global optimal direction is analyzed, the theoretical bidirectional virtual intersection point is calculated, and according to the virtual intersection point distribution, it is judged whether the current particle swarm tends to converge or disperse, which provides a basis for subsequent expansion and scaling adjustment. Among them, the bidirectional virtual intersection point includes the virtual intersection point on the extension line of the optimization direction and the reverse virtual intersection point on the reverse extension line of the optimization direction.

[0095] Further, according to the bidirectional virtual intersection recognition result and the process evaluation function, the process parameter particle swarm is adaptively expanded or shrunk. For example, if the diversity of the particle swarm is high and the bidirectional virtual intersection indicates that the particles are dispersed in a large search space, the size of the particle swarm can be appropriately increased to enhance the global search capability; on the contrary, if the diversity of the particle swarm is low and the bidirectional virtual intersection indicates that the particles are concentrated in a small area, the size of the particle swarm can be appropriately reduced to improve the search efficiency.

[0096] Finally, according to a preset iteration constraint condition (such as a maximum iteration number, a convergence threshold of the evaluation function, etc.), when the termination condition is met, the optimal process parameter combination in the current particle swarm is output as the target process parameter of the polyester knitted fabric surface antistatic treatment.

[0097] Through the above process, the efficiency of process development and the reliability of optimal solution are improved. The adaptive expansion and shrinkage mechanism can effectively avoid local optimum and low search efficiency, and ensure that the output process parameter combination is optimal in performance, cost and other multiple objectives.

[0098] In some implementations, the adaptive expansion and shrinkage of the process parameter particle swarm in combination with the process evaluation function includes:

[0099] When the optimization directions of the first preset number of particles in the preset range have a common virtual intersection point in the multi-dimensional parameter space, the process evaluation function value corresponding to the current position of each particle is obtained, and a fusion weight is defined accordingly; the first preset number of particles are weighted and fused based on the fusion weight, and a first fused process evaluation function value corresponding to the weighted fusion result is calculated based on the process evaluation function; the virtual intersection point process evaluation function value at the common virtual intersection point is determined based on the process evaluation function value; the space point corresponding to the larger value between the first fused process evaluation function value and the virtual intersection point process evaluation function value is selected as a new particle and added to the process parameter particle swarm.

[0100] Specifically, the common virtual intersection point refers to the intersection point of the straight lines extended by the optimization directions (i.e. speed vectors) of a plurality of particles in the multi-dimensional process parameter space, which can be regarded as the potential convergence target of the plurality of particles. The fusion weight refers to the proportion of each particle in the fusion process according to its current position and process evaluation function value. The first fused process evaluation function value is the process evaluation function value corresponding to the new process parameter combination obtained by weighting and fusing a plurality of particles. The virtual intersection point process evaluation function value refers to the process evaluation function value calculated at the common virtual intersection point assuming that there is a corresponding process parameter combination.

[0101] Specifically, first, in each iteration, it is determined whether there is an intersection of the optimization directions of a first preset number (such as 2-3) of particles in a preset range (such as a region with a specific radius of the parameter space), and a particle set meeting the condition is recorded as a candidate particle group participating in fusion. Then, the process evaluation function value of the current position of each candidate particle is obtained, and a fusion weight (such as a weight distribution based on normalization) is assigned according to the size. Then, the parameter combination of each particle is weighted and fused based on the above fusion weight to obtain a new parameter combination (i.e., a first fusion point), and the process evaluation function value of the weighted fusion point (i.e., the first fusion point) is calculated by the process evaluation function, which is recorded as the first fusion process evaluation function value.

[0102] Further, the parameter combination of the common virtual intersection point is taken as input, the process evaluation function value is calculated by the process evaluation function, which is recorded as the virtual intersection process evaluation function value; then, the virtual intersection process evaluation function value and the first fusion process evaluation function value are compared, and the parameter space point (i.e., the point with better performance) corresponding to the larger value is taken as a new particle to join the process parameter particle group for subsequent iterations. At the same time, the first preset number of particles are removed from the particle group.

[0103] Through the above process, the information of multiple high-quality parameter combinations in the particle group can be fully utilized, and new particles with better performance are dynamically generated through weighted fusion and virtual intersection analysis, which improves the exploration ability of the search space and the global optimal search probability. At the same time, according to the feedback of the process evaluation function, the particle group structure is adjusted in real time to realize adaptive scaling, which further speeds up the optimization convergence speed and improves the intelligent level and optimization efficiency of process parameter optimization.

[0104] In some implementations, the adaptive expansion and scaling of the process parameter particle group in combination with the process evaluation function further includes:

[0105] In a multi-dimensional parameter space, a reverse virtual intersection point formed by the intersection of the reverse extensions of a second preset number of optimization directions is identified; when the reverse virtual intersection point is located in a preset range, the angle variance of adjacent angles formed by the second preset number of optimization directions is calculated; the number of new particles to be inserted is determined based on the angle variance, and the new particles are inserted on the central axes of the adjacent angles of the second preset number of optimization directions in descending order of adjacent angles.

[0106] Specifically, the reverse virtual intersection point refers to a theoretical point formed by the intersection of the reverse extensions (i.e., the reverse extensions of velocity vectors) of the optimization directions of a plurality of particles in a multi-dimensional parameter space, which reflects the potential divergence trend of the particle group or the space region that has not been fully explored. The angle variance refers to the variance calculated from the angles between the optimization directions, which reflects the dispersion degree of the particle distribution. The central axis of the adjacent angles refers to the direction of the bisector of the angle between two optimization directions in the parameter space.

[0107] Specifically, first, based on the same method principle of the above-mentioned common virtual intersection, in the process parameter space, the optimization directions of a second preset number (such as 3-5) of particles are selected, the opposite directions of the velocity vectors thereof are respectively extended, and the intersection points of the reverse extension lines are calculated; if there is an intersection point and its coordinates fall within the preset parameter range (i.e. meet the preset range requirement), the intersection point of the reverse extension line is recorded as the reverse virtual intersection point. Then, based on the optimization directions of the above-mentioned second preset number of particles, the included angles between every two adjacent optimization directions (i.e. velocity vectors) are calculated, a set of angle values (and the above-mentioned adjacent angles, including multiple) are obtained, and the variance of the angles is calculated.

[0108] Further, according to the size of the angle variance, the number of new particles to be inserted is determined, wherein the larger the angle variance is, the more dispersed the particle distribution is, and more new particles need to be inserted to strengthen the search in this area. Optionally, the angle variance is mapped to the number of new particles to be inserted through a preset mapping rule or mapping function.

[0109] Further, according to the sorting of all adjacent angles from large to small, the direction pair (i.e. velocity vector pair) with a larger angle is preferentially selected, and a new particle is inserted at the reverse virtual intersection point as the reference on the median axis (i.e. the bisector direction of the two optimization directions).

[0110] Optionally, the parameter combination of the newly inserted particle can be set as the intermediate value between the two adjacent original particles in the median axis direction or directly set as the position of a certain step distance from the reverse virtual intersection point on the median axis.

[0111] Through the above process, the particle swarm structure can be dynamically adjusted according to the dispersion of the particle swarm distribution and the spatial characteristics of the reverse virtual intersection point, the potential exploration blind area in the parameter space can be actively filled, the global search capability can be enhanced, and the local optimum can be prevented, so as to further improve the efficiency and effect of process optimization.

[0112] In some implementations, in combination with the process evaluation function, the adaptive expansion and contraction of the process parameter particle swarm is also included:

[0113] The dynamic fluctuation range of the particle swarm size is set, and the adaptive regulation of the particle swarm number is carried out in the dynamic fluctuation range based on the initial particle swarm number; if the current particle swarm number is lower than the lower limit of the fluctuation, the adaptive contraction is suspended until the current particle swarm number is higher than the initial particle swarm number; if the current particle swarm number is higher than the upper limit of the fluctuation, the adaptive expansion is suspended until the current particle swarm number is lower than the initial particle swarm number.

[0114] Specifically, the dynamic fluctuation range of the particle swarm size refers to the upper and lower limit interval of the allowed particle number change based on the initial particle swarm number. For example, if the initial particle swarm number is N0, the dynamic fluctuation range can be defined as [N min , N max ], where N min =N0−ΔN, N max =N0+ΔN, and ΔN is the fluctuation allowance (e.g., 20% of N0).

[0115] Specifically, first, the dynamic fluctuation range of the particle swarm number is set so that the particle swarm number can be flexibly expanded or contracted according to the search needs during the optimization process. For example, the initial particle swarm number is 50, and the dynamic fluctuation range is 40-60.

[0116] Then, the current particle swarm number is monitored in real time based on the initial particle swarm number. When the particle swarm number is lower than the lower limit (e.g., 40) due to adaptive contraction operation, the contraction operation is automatically suspended, and the contraction is resumed only when the particle swarm number rises (e.g., through new particle insertion or natural growth of iterations) to be higher than the initial particle swarm number (e.g., 50). Similarly, when the particle swarm number exceeds the upper limit (e.g., 60) due to adaptive expansion operation, the expansion operation is automatically suspended until the particle swarm number falls below the initial particle swarm number (e.g., 50), and then the expansion is resumed.

[0117] Through the above process, it can effectively prevent the search ability from being insufficient due to excessive contraction of the particle swarm size or waste of computing resources caused by excessive expansion of the particle swarm size, achieve dynamic balance of the global exploration ability and computing efficiency of the optimization process, and improve the stability and efficiency of process parameter optimization.

[0118] In summary, the surface antistatic treatment method for polyester knitted fabric provided by the present application has the following technical effects:

[0119] Through the process scheme determined based on the target application scenario, the historical sample process record is called to construct a standardized process training data set. Based on the training data set, a machine learning model for antistatic performance prediction is established and trained in combination with a knowledge distillation mechanism. A cost-related evaluation factor is constructed, and the antistatic performance prediction model is combined with the cost evaluation factor to set a process evaluation function containing a penalty term, which is used to quantify the comprehensive effect of the treatment process. The process evaluation function is used as the optimization target, and an adaptive expansion and contraction strategy is used to iteratively optimize the treatment process parameters, and finally the target process parameter combination for polyester knitted fabric surface antistatic treatment is determined, thereby achieving the technical effects of reducing parameter adjustment cost, improving adjustment efficiency, and enhancing antistatic treatment flexibility.

[0120] Example two, as Figure 2It is a structure schematic view of a surface antistatic treatment system of a polyester knitted fabric according to the present application. For example, Figure 1 The flowchart of a surface antistatic treatment method of a polyester knitted fabric according to the present application can be implemented by the structure shown in Figure 2

[0121] Based on the same idea as the surface antistatic treatment method of a polyester knitted fabric according to the present application, the present application also provides a surface antistatic treatment system of a polyester knitted fabric, which comprises:

[0122] A training set construction module 11 is configured to determine a processing scheme according to a target application scenario, call a sample process record, and construct a standard process training set.

[0123] A prediction model training module 12 is configured to construct and train an antistatic performance prediction model based on the standard process training set and in combination with a knowledge distillation mechanism.

[0124] An evaluation function definition module 13 is configured to construct a cost evaluation factor and define a process evaluation function in combination with the antistatic performance prediction model and the cost evaluation factor, wherein the process evaluation function is associated with a penalty space.

[0125] A process parameter optimization module 14 is configured to perform adaptive scaling iteration optimization on the processing parameters with the process evaluation function as the optimization target, determine a target process parameter combination for the polyester knitted fabric according to the iteration optimization result, and perform surface antistatic treatment.

[0126] In some embodiments, the training set construction module 11 comprises:

[0127] A sample process parameter data acquisition unit is configured to acquire sample process parameter data, wherein the process parameters at least include an additive ratio, a treatment temperature, a treatment time, and a fabric moisture content.

[0128] A sample antistatic performance data acquisition unit is configured to acquire sample antistatic performance data associated with the sample process parameter data, wherein the antistatic performance at least includes an electrostatic voltage, a charge surface density, and a half-life.

[0129] A data preprocessing and standard process training set generation unit is configured to preprocess and target value calibrate the sample process parameter data and the sample antistatic performance data, divide the target value calibration results into positive samples and negative samples, and output the standard process training set.

[0130] In some embodiments, the prediction model training module 12 comprises:

[0131] ​The benchmark prediction model construction and training unit is configured to construct and train a benchmark prediction model based on a deep neural network by using the standard process training set.

[0132] The simplified prediction model training unit is configured to train a simplified prediction model based on a knowledge distillation mechanism, in combination with the prediction output of the benchmark prediction model and the standard process training set.

[0133] The simplified prediction model accuracy verification and output unit is configured to verify the accuracy of the simplified prediction model, and if the positive samples and negative samples in the standard process training set both meet the set performance evaluation threshold, the simplified prediction model is output and deployed as the antistatic performance prediction model.

[0134] In some embodiments, the cost evaluation factor in the evaluation function definition module 13 at least includes a processing cost item per unit fabric, a processing energy consumption item, a processing time item, and a resource consumption item.

[0135] In some embodiments, the evaluation function definition module 13 includes:

[0136] The antistatic treatment task document acquisition and analysis unit is configured to acquire an antistatic treatment task document and analyze the antistatic treatment task document to extract compliance constraints.

[0137] The penalty space definition unit is configured to define the penalty space of the process evaluation function according to the compliance constraints.

[0138] The process evaluation function acquisition and association unit is configured to obtain the process evaluation function by weightedly fusing the output of the antistatic performance prediction model and the cost evaluation factor, and associate the process evaluation function with the penalty space.

[0139] In some embodiments, the process parameter optimization module 14 includes:

[0140] The process parameter particle swarm initialization unit is configured to initialize a process parameter particle swarm containing multiple candidate process parameter combinations based on the standard process training set.

[0141] The iterative optimization and optimization direction calculation unit is configured to perform an iterative optimization process based on a particle swarm optimization algorithm according to the initialized process parameter particle swarm, calculate the optimization direction of each particle in each round of iteration, and identify the bidirectional virtual intersection point of the particle in the search space according to the optimization direction.

[0142] The adaptive scaling unit is configured to perform adaptive scaling on the process parameter particle swarm according to the bidirectional virtual intersection point identification result, the process evaluation function.

[0143] An iterative optimization result output unit is configured to output the iterative optimization result according to a preset iterative constraint condition.

[0144] In some implementations, the execution steps of the adaptive expansion and contraction unit in the process parameter optimization module 14 include:

[0145] When the optimization directions of the first preset number of particles in the preset range have a common virtual intersection point in the multi-dimensional parameter space, the process evaluation function value corresponding to the current position of each particle is obtained, and a fusion weight is defined correspondingly.

[0146] Based on the fusion weight, the first preset number of particles are weighted and fused, and a first fused process evaluation function value corresponding to the weighted and fused result is calculated based on the process evaluation function.

[0147] Based on the process evaluation function value, a virtual intersection point process evaluation function value at the common virtual intersection point is determined.

[0148] The space point corresponding to the larger one of the first fused process evaluation function value and the virtual intersection point process evaluation function value is selected as a new particle and added to the process parameter particle group.

[0149] In some implementations, the execution steps of the adaptive expansion and contraction unit in the process parameter optimization module 14 further include:

[0150] In the multi-dimensional parameter space, a reverse virtual intersection point formed by the intersection of the reverse extension lines of the second preset number of optimization directions is identified.

[0151] When the reverse virtual intersection point is located within the preset range, the angle variance of adjacent angles formed by the second preset number of optimization directions is calculated.

[0152] Based on the angle variance, the number of new particles to be inserted is determined, and the new particles are inserted on the central axes of the adjacent angles of the second preset number of optimization directions in descending order of the adjacent angles.

[0153] In some implementations, the adaptive expansion and contraction unit in the process parameter optimization module 14 further includes:

[0154] A scale constraint subunit is configured to set a dynamic fluctuation range of the particle group scale, and adaptively regulate the number of the particle group within the dynamic fluctuation range based on the initial particle group number.

[0155] An adaptive contraction suspension subunit is configured to suspend adaptive contraction if the current particle group number is lower than the lower limit of the fluctuation, until the current particle group number is higher than the initial particle group number.

[0156] The adaptive expansion suspension subunit is configured to suspend the adaptive expansion if the current particle group quantity is higher than the fluctuation upper limit, until the current particle group quantity is lower than the initial particle group quantity.

[0157] It should be understood that the embodiment mentioned in the specification focuses on its difference from other embodiments, and the specific embodiments in the foregoing embodiment one are also applicable to the surface antistatic treatment system of the polyester knitted fabric in embodiment two, which will not be further expanded here for the sake of brevity of the specification.

[0158] It should be understood that the embodiments disclosed in the present application and the above description can enable those skilled in the art to implement the present application. Meanwhile, the present application is not limited to the above-mentioned part of the embodiments, and it should be understood that those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or make equivalent replacement for part of the technical features; and the modification or replacement does not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for surface antistatic treatment of a polyester knitted fabric, characterized by, The method comprises the following steps: According to the target application scenario, the processing process scheme is determined, the sample process record is called, and the standard process training set is constructed; Based on the standard process training set, combined with the knowledge distillation mechanism, an antistatic performance prediction model is constructed and trained; Construct a cost evaluation factor, and combine the antistatic performance prediction model and the cost evaluation factor to define a process evaluation function, wherein the process evaluation function is associated with a penalty space; Taking the process evaluation function as the optimization target, the processing process parameters are iteratively optimized by self-adaptive expansion and contraction, and the target process parameter combination for polyester knitted fabric is determined according to the iteration optimization result to perform surface antistatic treatment; Taking the process evaluation function as the optimization target, the processing process parameters are iteratively optimized by self-adaptive expansion and contraction, comprising: Based on the standard process training set, the process parameter particle swarm containing multiple candidate process parameter combinations is initialized; According to the initialized process parameter particle swarm, an iterative optimization process based on a particle swarm optimization algorithm is executed, the optimization direction of each particle is calculated in each iteration process, and the bidirectional virtual intersection point of the particle in the search space is identified according to the optimization direction; According to the bidirectional virtual intersection point identification result, the process evaluation function, the process parameter particle swarm is adaptively expanded and contracted; According to the preset iteration constraint condition, the iteration optimization result is outputted; In combination with the process evaluation function, the process parameter particle swarm is adaptively expanded and contracted, further comprising: When the optimization direction of the first preset number of particles in the preset range has a common virtual intersection point in the multi-dimensional parameter space, the process evaluation function value corresponding to the current position of each particle is obtained, and a fusion weight is defined accordingly; Based on the fusion weight, the first preset number of particles are weighted and fused, and the first fusion process evaluation function value corresponding to the weighted fusion result is calculated based on the process evaluation function; Based on the process evaluation function value, the virtual intersection point process evaluation function value at the common virtual intersection point is determined; The space point corresponding to the larger value between the first fusion process evaluation function value and the virtual intersection point process evaluation function value is selected as a new particle and added to the process parameter particle swarm.

2. The method for surface antistatic treatment of polyester knitted fabric according to claim 1, wherein In combination with the antistatic performance prediction model and the cost evaluation factor, the process evaluation function is defined, comprising: An antistatic treatment task sheet is obtained, and the antistatic treatment task sheet is parsed to extract the compliance constraints; According to the compliance constraints, the penalty space of the process evaluation function is defined; The output of the antistatic performance prediction model and the cost evaluation factor are weighted and fused to obtain the process evaluation function, and the process evaluation function is associated with the penalty space.

3. The method for surface antistatic treatment of polyester knitted fabric according to claim 1, wherein the antistatic agent is applied to the surface of the polyester knitted fabric in an amount of 0.1 to 5% by weight based on the weight of the polyester knitted fabric. In combination with the process evaluation function, the process parameter particle swarm is adaptively expanded and contracted, further comprising: In the multi-dimensional parameter space, the reverse virtual intersection point formed by the intersection of the reverse elongation lines of the second preset number of optimization directions is identified; When the reverse virtual intersection point is located in the preset range, the angle variance of the adjacent included angle formed by the second preset number of optimization directions is calculated; Determine the number of new particles to be inserted based on the angle variance, and insert new particles on the central axes of the adjacent angles of the second preset number of optimization directions in descending order of the adjacent angles.

4. The method for surface antistatic treatment of polyester knitted fabric according to claim 1, wherein The adaptive expansion and contraction of the process parameter particle group in combination with the process evaluation function further includes: Set a dynamic fluctuation range for the particle group size, and adaptively regulate the number of particles within the dynamic fluctuation range based on the initial number of particles; If the current number of particles is lower than the lower limit of the fluctuation, suspend the adaptive contraction until the current number of particles is higher than the initial number of particles; If the current number of particles is higher than the upper limit of the fluctuation, suspend the adaptive expansion until the current number of particles is lower than the initial number of particles.

5. The method for surface antistatic treatment of polyester knitted fabric according to claim 1, wherein According to the processing process scheme determined by the target application scenario, the sample process record is called to construct a standard process training set, including: Obtain sample process parameter data, wherein the process parameters at least include auxiliary agent ratio, processing temperature, processing time and fabric moisture content; Obtain sample antistatic performance data associated with the sample process parameter data, wherein the antistatic performance at least includes static voltage, charge surface density and half-life; Preprocess and target value calibrate the sample process parameter data and the sample antistatic performance data, and divide them into positive samples and negative samples according to the target value calibration result, and output the standard process training set.

6. The method for surface antistatic treatment of polyester knitted fabric according to claim 5, wherein the antistatic agent is applied to the surface of the polyester knitted fabric in an amount of 0.1 to 5% by weight based on the weight of the polyester knitted fabric. Based on the standard process training set, an antistatic performance prediction model is constructed and trained in combination with a knowledge distillation mechanism, including: A benchmark prediction model based on deep neural network is constructed and trained through the standard process training set; Based on the knowledge distillation mechanism, a simplified prediction model is trained in combination with the prediction output of the benchmark prediction model and the standard process training set; The accuracy of the simplified prediction model is verified, and if the positive samples and negative samples in the standard process training set both meet the set performance evaluation threshold, the simplified prediction model is output and deployed as the antistatic performance prediction model.

7. The method for surface antistatic treatment of polyester knitted fabric according to claim 1, wherein the antistatic agent is applied to the surface of the polyester knitted fabric in an amount of 0.1 to 5% by weight based on the weight of the polyester knitted fabric. The cost evaluation factor at least includes processing cost per unit fabric, processing energy consumption, processing time and resource consumption.

8. A surface antistatic treatment system for polyester knitted fabric, characterized by, A surface antistatic treatment method for polyester knitted fabric is implemented according to any one of claims 1-7, comprising: A training set construction module is configured to call sample process records to construct a standard process training set according to a processing process scheme determined by a target application scenario; A prediction model training module is configured to construct and train an antistatic performance prediction model based on the standard process training set in combination with a knowledge distillation mechanism; An evaluation function definition module is configured to construct a cost evaluation factor, and define a process evaluation function in combination with the antistatic performance prediction model and the cost evaluation factor, wherein the process evaluation function is associated with a penalty space; A process parameter optimization module is configured to iteratively optimize the adaptive expansion and contraction of the processing process parameters with the process evaluation function as the optimization target, and determine the target process parameter combination for the surface antistatic treatment of the polyester knitted fabric according to the iterative optimization result.

Images