Real-time monitoring and closed-loop control system and method for quality of composite material pultrusion

Abstract

This invention provides a real-time monitoring and closed-loop control system and method for composite material pultrusion molding quality, belonging to the field of process quality monitoring and intelligent control technology. It includes: collecting real-time multi-source raw data during the production process; constructing a training set and validation set by combining historical production data with the corresponding production site conditions to generate a fusion model; inputting the real-time multi-source raw data into the fusion model for production quality assessment and defect prediction to generate process parameter adjustment strategies; adjusting the parameters of production actuators to obtain production feedback data; evaluating the quality of the production feedback data; and obtaining and displaying a real-time production quality report. Although no manual intervention is required throughout the process, it significantly improves the production yield and efficiency of pultruded products and reduces production costs.

Description

Technical Field

[0001] This invention relates to the field of process quality monitoring and intelligent control technology, and in particular to a real-time monitoring and closed-loop control system and method for the quality of composite material pultrusion molding. Background Technology

[0002] Pultruded composite materials, with their core advantages of light weight, high strength, and strong corrosion resistance, have been widely used in various fields such as electrical engineering, construction, and rail transportation. Their production quality directly determines the reliability and service life of the final products. In the continuous production of pultruded composite materials, online real-time quality monitoring and precise dynamic control of process parameters are crucial for ensuring high yield and controlling production costs.

[0003] Existing quality inspection and process control technologies for pultruded products suffer from numerous technical drawbacks: Detection of defects such as poor fiber impregnation often relies on destructive offline cutting and sampling methods, which not only fail to provide real-time feedback on production status but also easily lead to product scrapping; internal porosity detection depends on costly X-ray CT technology, which has a high barrier to entry and is difficult to adapt to online continuous inspection production scenarios; delamination and debonding defects are mainly detected through contact scanning, exhibiting significant detection lag and failing to reflect the real-time defect status of the production process; surface defects such as white filaments rely on manual visual inspection, which is highly susceptible to human factors, resulting in a high rate of missed detections and low inspection efficiency. Furthermore, traditional technologies lack a comprehensive real-time online non-destructive testing system. The lag in test results prevents rapid adjustment of production process parameters, leading to a disconnect between process control and production, ultimately hindering the improvement of pultruded product yield and resulting in poor control over production efficiency and costs.

[0004] Therefore, the present invention provides a real-time monitoring and closed-loop control system and method for the quality of composite material pultrusion molding. Summary of the Invention

[0005] This invention relates to a real-time monitoring and closed-loop control system and method for composite material pultrusion molding quality. It overcomes the technical shortcomings of existing technologies, such as destructive processes, lag, high false negative rates, disconnect between process control and quality inspection, and the lack of a closed-loop control system in composite material pultrusion molding quality detection. This invention provides a real-time monitoring and closed-loop control system for composite material pultrusion molding quality. Through multi-module collaborative operation, it achieves blind-spot-free acquisition of multi-source data during the pultrusion molding production process, fusion modeling based on physical models and machine learning, accurate production quality assessment and defect prediction, and dynamic closed-loop control of process parameters. It also generates a visualized real-time production quality report, requiring no manual intervention throughout the entire process. This significantly improves the production yield and efficiency of pultruded products while reducing production costs.

[0006] This invention provides a real-time monitoring and closed-loop control system for the quality of composite material pultrusion molding, including: The multi-source sensor acquisition module is used to collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal and real-time infrared thermal field information during the production process, and to construct real-time multi-source raw data; The modeling preprocessing module is used to filter several historical production data related to the production process and construct training and validation sets by combining the production condition type corresponding to each historical production data. The model building module is used to build a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model. The evaluation and prediction module is used to input the real-time multi-source raw data into the fusion model to perform production quality evaluation and defect prediction, and generate process parameter adjustment strategies. The feedback control module is used to adjust the parameters of the production execution equipment using the process parameter adjustment strategy to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

[0007] In one implementable manner, the multi-source sensing acquisition module includes: A first temperature sensor is installed at the mold inlet, a second temperature sensor is installed in the middle section of the mold, and a third temperature sensor is installed at the mold outlet. The first temperature sensor, the second temperature sensor, and the third temperature sensor are used to monitor the mold temperature distribution and temperature change information during the product curing process in real time, so as to obtain real-time temperature information during the production process. A first vision sensor is installed at the mold exit, a second vision sensor is installed behind the traction machine, and a third vision sensor is installed behind the preforming stage. The first vision sensor, the second vision sensor, and the third vision sensor are used to monitor the cross-sectional shape after preforming and the surface state after curing in real time to obtain real-time video information during the production process. A tension sensor installed at the traction clamp of the traction machine is used to collect tension data in real time during the product traction process, so as to obtain real-time tension information during the production process. A dielectric sensor embedded in the preforming stage is used to collect real-time dielectric signals during the production process; An acoustic emission probe is installed at the traction clamp of the traction machine to collect real-time acoustic emission signals during the production process; Infrared thermal imagers are installed at the production site to collect real-time infrared thermal field information during the production process. The data integration unit is used to integrate all real-time data to construct the real-time generated raw data in the production process.

[0008] One feasible approach also includes: The acquisition setting unit is used to adjust the acquisition frequency of the first temperature sensor, the second temperature sensor and the third temperature sensor to 1 second / time; Set the sampling frequency of the tension sensor to 1 second / time; The sampling frequency of the dielectric sensor is set to 1 second / time; The acquisition frequency of the infrared thermal imager is set to 50 sets of thermal field distribution maps per second.

[0009] In one implementable manner, the modeling preprocessing module includes: The data filtering unit is used to acquire field parameters in the production site, deduce several production conditions in the production site, and filter sensor data, process parameter data and product quality test results corresponding to each production condition from Internet big data. The data aggregation unit is used to aggregate the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. The dataset partitioning unit is used to label the training dataset according to the defects in the working conditions, and to divide the labeled training dataset into a training set and a validation set in a ratio of 7:3.

[0010] In one implementable manner, the model building module includes: The reasoning building unit is used to build a physical model based on the on-site parameters in the production site, the principles of curing reaction kinetics, heat and mass transfer laws, and the mechanical properties of materials. The model training unit is used to train the physical model using the training set to obtain several parameters of the physical model to be adjusted, to determine the parameter value corresponding to each parameter to be adjusted using the validation set, and to adjust the parameters of the physical model to obtain an adjusted model. The functional verification unit is used to construct functional test data using the training set and the verification set, and to perform functional verification on the adjusted model. When the defect prediction accuracy of the fusion model is ≥95% and the false alarm rate is ≤2.5%, training is stopped to obtain the fusion model.

[0011] In one implementable manner, the evaluation and prediction module includes: The data processing unit is used to mine the nonlinear relationships of the real-time multi-source raw data using gradient boosting trees, perform feature enhancement on the real-time multi-source raw data using support vector machines, and input the nonlinear relationships and the enhanced real-time multi-source raw data into the fusion model. The quality assessment unit is used to recreate the finished product in the production process in real time in the fusion model, analyze the internal curing, surface flatness and cross-sectional dimensions of the finished product, and generate quality assessment results. The defect prediction unit is used to analyze abnormal changes in the production process using the fusion model, analyze the defects and their severity corresponding to the abnormal changes, and generate defect prediction results. The strategy production unit is used to determine the process parameters that need to be adjusted in the production process based on the quality assessment results and the defect prediction results, and to generate a process parameter adjustment strategy in combination with the production execution equipment corresponding to each process parameter that needs to be adjusted.

[0012] In one implementable manner, the feedback control module includes: The sequence determination unit is used to identify the influence weight of each process parameter to be adjusted in the process parameter adjustment strategy on the production process, and determine the parameter adjustment sequence of this parameter adjustment work according to the order of production weight from high to low. The step size determination unit is used to obtain the parameter deviation value and parameter limit threshold corresponding to each of the production actuators, and determine the parameter adjustment step size corresponding to each of the production actuators in the current adjustment work by combining the parameter adjustment amount corresponding to the process parameter adjustment strategy. The monitoring unit is used to dynamically and iteratively adjust each of the production execution devices according to the parameter adjustment sequence and the parameter adjustment step size, and at the same time obtain the production feedback parameters of the current adjustment work, and determine the real-time production defects of the production process based on the production feedback parameters. The continuous monitoring unit is used to end the current adjustment work when the real-time production defect disappears, determine that the production site has entered a high-quality production state, and at the same time use the production feedback data to analyze and display the real-time production quality report of the production process.

[0013] This invention provides a method for real-time monitoring and closed-loop control of the quality of composite material pultrusion molding, including: Step 1: Collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal, and real-time infrared thermal field information during the production process to construct real-time multi-source raw data; Step 2: Filter several historical production data related to the production process, and construct a training set and a validation set by combining the production condition type corresponding to each historical production data. Step 3: Construct a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model; Step 4: Input the real-time multi-source raw data into the fusion model for production quality assessment and defect prediction, and generate process parameter adjustment strategies; Step 5: Use the process parameter adjustment strategy to adjust the parameters of the production execution equipment to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

[0014] In one feasible embodiment, step 2 includes: Step 21: Obtain the field parameters in the production site, deduce several production conditions in the production site, and filter the sensor data, process parameter data and product quality test results corresponding to each production condition in the Internet big data. Step 22: Summarize the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. Step 23: Label the training dataset according to the defects in the working conditions, and divide the labeled training dataset into a training set and a validation set in a ratio of 7:3.

[0015] The beneficial effects achievable by this invention are as follows: To effectively improve product yield and production efficiency while reducing production costs, real-time production data from multiple dimensions, including temperature, video, tension, dielectric, acoustic emission, and infrared thermal field, are simultaneously collected at key nodes in the pultrusion molding production process. This data is integrated to form unified real-time multi-source raw data, achieving blind-spot-free acquisition of multi-physics field data throughout the entire production process. This provides comprehensive, accurate, and real-time raw data support for subsequent quality assessment, defect prediction, and model building, overcoming the limitations of traditional single-data acquisition. Then, historical production data highly matching the current production process is precisely selected. The data is standardized based on production condition types, and training and validation sets are constructed, effectively avoiding interference from invalid data in model building. This provides a standardized data source that fits the actual production scenario for training and parameter adjustment of the fusion model, ensuring the relevance and accuracy of subsequent model building. Furthermore, based on relevant physical knowledge of the production process... A physical model is constructed, and training and parameter tuning are completed using training and validation sets. The resulting fusion model combines the support of physical theory with the data-driven nature of machine learning, accurately meeting the actual needs of pultrusion production. This improves the model's accuracy in production quality assessment and defect prediction, laying a reliable model foundation for subsequent assessment and prediction work. Further inputting real-time multi-source raw data into the fusion model enables real-time quantitative assessment of production quality and accurate prediction of product defects. Simultaneously, based on the assessment and prediction results, targeted process parameter adjustment strategies are generated, providing a clear and scientific basis for subsequent process control. This solves the problems of strong lag and high false negative rate in traditional detection methods, enabling early prediction and precise policy implementation of production anomalies. Finally, precise parameter adjustments are made to production actuators based on the process parameter adjustment strategies. Production feedback data after parameter adjustment is collected, and quality evaluation is completed, generating and displaying a real-time production quality report, forming an "assessment- The complete closed-loop control of "control-verification" enables dynamic iterative optimization of process parameters, timely elimination of defects in the production process, avoidance of batch production of unqualified products, and visualization of quality reports facilitates real-time monitoring of production status by staff, while providing complete data reference for subsequent process optimization analysis.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the composition of the real-time monitoring and closed-loop control system for composite material pultrusion molding quality in an embodiment of the present invention; Figure 2 This is a schematic diagram of the workflow of the real-time monitoring and closed-loop control method for composite material pultrusion molding quality in an embodiment of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Example 1: This example provides a real-time monitoring and closed-loop control system for the quality of composite material pultrusion molding, such as... Figure 1 As shown, it includes: The multi-source sensor acquisition module is used to collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal and real-time infrared thermal field information during the production process, and to construct real-time multi-source raw data; The modeling preprocessing module is used to filter several historical production data related to the production process and construct training and validation sets by combining the production condition type corresponding to each historical production data. The model building module is used to build a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model. The evaluation and prediction module is used to input the real-time multi-source raw data into the fusion model to perform production quality evaluation and defect prediction, and generate process parameter adjustment strategies. The feedback control module is used to adjust the parameters of the production execution equipment using the process parameter adjustment strategy to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

[0021] In this example, the real-time temperature information refers to real-time data collected by temperature sensors deployed at key nodes in the pultrusion production process, which reflects the temperature distribution of the mold and the temperature changes during the curing process of the composite material product.

[0022] In this example, real-time video information refers to real-time video image data collected by vision sensors deployed at locations such as after the preforming stage, at the mold exit, and after the traction machine, which can reflect the cross-sectional shape of the product after preforming and the surface state after curing.

[0023] In this example, the real-time tension information refers to the real-time tension data collected by the tension sensor installed at the traction clamp of the traction machine, which reflects the stress changes experienced by the product during traction.

[0024] In this example, the real-time dielectric signal refers to real-time signal data collected by a dielectric sensor embedded in the preforming stage, which reflects the changes in the dielectric properties of the material during the pultrusion production process.

[0025] In this example, the real-time acoustic emission signal refers to the real-time acoustic emission data collected by the acoustic emission probe integrated into the traction clamp of the traction machine, which can reflect the changes in internal stress, delamination and debonding of the product during the production process.

[0026] In this example, the real-time infrared thermal field information refers to real-time data collected by infrared thermal imagers covering key monitoring locations throughout the entire production process, which reflects the distribution of infrared thermal fields at each key node in the production process.

[0027] In this example, the real-time multi-source raw data refers to the multi-dimensional, integrated real-time production data set formed by the multi-source sensor acquisition module after integrating the real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal and real-time infrared thermal field information collected in the pultrusion production process.

[0028] In this example, historical production data refers to the total sensor data, process parameter data, and corresponding product quality inspection results generated under different production conditions during the pultrusion molding process.

[0029] In this example, the production condition type refers to the different production operation status types classified according to different conditions such as the product type, mold specifications, raw material characteristics, and process parameters of pultrusion production.

[0030] In this example, the relevant physical knowledge refers to the theoretical knowledge in the physical fields related to the curing reaction kinetics, heat and mass transfer laws, and material mechanical properties associated with the pultrusion molding process of composite materials.

[0031] In this example, the physical model represents a theoretical model constructed based on relevant physical knowledge of pultrusion production, which can clearly define the quantitative relationship between process parameters such as mold temperature, traction speed, and tension and the degree of internal curing and surface quality of the product.

[0032] In this example, the training set refers to the dataset used to train the physical model after the historical production data selected by the modeling preprocessing module is divided according to a certain proportion.

[0033] In this example, the validation set refers to the dataset used to adjust the parameters and verify the effect of the trained physical model after the historical production data selected by the modeling preprocessing module is divided according to a certain proportion.

[0034] In this example, the fusion model represents a comprehensive evaluation model that combines the support of physical theory and the ability to analyze machine learning data, formed by training a physical model based on relevant physical knowledge using a training set and adjusting parameters using a validation set.

[0035] In this example, the process parameter adjustment strategy refers to the specific control plan formulated by the evaluation and prediction module based on the production quality evaluation results and defect prediction results. This plan includes the process parameters to be adjusted, the adjustment sequence, the adjustment step size, and the corresponding production execution equipment.

[0036] In this example, production execution equipment refers to various types of equipment that can execute process parameter adjustment instructions during the pultrusion molding process of composite materials, including mold temperature control systems, traction machines, and glue injection equipment.

[0037] In this example, the production feedback data represents the real-time data of the pultrusion production process that is re-collected by the multi-source sensor acquisition module after the parameters of the production actuators have been adjusted according to the process parameter adjustment strategy.

[0038] In this example, the real-time production quality report is a visual report generated by the feedback control module after completing the quality evaluation based on the production feedback data. It includes information such as production quality status, defect status, process parameter adjustment process, and quality evaluation results.

[0039] The working principle and beneficial effects of the above technical solution are as follows: To effectively improve product yield and production efficiency while reducing production costs, real-time production data from multiple dimensions, including temperature, video, tension, dielectric, acoustic emission, and infrared thermal field, are simultaneously collected at key nodes in the pultrusion molding production process. This data is integrated to form unified real-time multi-source raw data, achieving blind-spot-free acquisition of multi-physics field data throughout the entire production process. This provides comprehensive, accurate, and real-time raw data support for subsequent quality assessment, defect prediction, and model building, overcoming the limitations of traditional single-data acquisition. Then, historical production data highly matching the current production process is precisely selected. The data is standardized based on production condition types, and training and validation sets are constructed. This effectively avoids interference from invalid data in model building, providing a standardized data source that fits the actual production scenario for model training and parameter adjustment. This ensures the relevance and accuracy of subsequent model building. Furthermore, based on the relevant physical properties of the production process... The knowledge-based physical model is trained and its parameters adjusted using training and validation sets. The resulting fusion model combines the support of physical theory with the data-driven nature of machine learning, accurately meeting the actual needs of pultrusion production. This improves the model's accuracy in production quality assessment and defect prediction, laying a reliable model foundation for subsequent assessment and prediction work. Further inputting real-time multi-source raw data into the fusion model enables real-time quantitative assessment of production quality and accurate prediction of product defects. Simultaneously, based on the assessment and prediction results, targeted process parameter adjustment strategies are generated, providing a clear and scientific basis for subsequent process control. This solves the problems of strong lag and high false negative rate in traditional detection methods, enabling early prediction and precise policy implementation of production anomalies. Finally, precise parameter adjustments are made to production actuators based on the process parameter adjustment strategies. Production feedback data after parameter adjustments is collected and quality evaluation is completed, generating and displaying a real-time production quality report, forming an "assessment- The complete closed-loop control of "control-verification" enables dynamic iterative optimization of process parameters, timely elimination of defects in the production process, avoidance of batch production of unqualified products, and visualization of quality reports facilitates real-time monitoring of production status by staff, while providing complete data reference for subsequent process optimization analysis.

[0040] Example 2: Based on Example 1, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality includes a multi-source sensor acquisition module comprising: A first temperature sensor is installed at the mold inlet, a second temperature sensor is installed in the middle section of the mold, and a third temperature sensor is installed at the mold outlet. The first temperature sensor, the second temperature sensor, and the third temperature sensor are used to monitor the mold temperature distribution and temperature change information during the product curing process in real time, so as to obtain real-time temperature information during the production process. A first vision sensor is installed at the mold exit, a second vision sensor is installed behind the traction machine, and a third vision sensor is installed behind the preforming stage. The first vision sensor, the second vision sensor, and the third vision sensor are used to monitor the cross-sectional shape after preforming and the surface state after curing in real time to obtain real-time video information during the production process. A tension sensor installed at the traction clamp of the traction machine is used to collect tension data in real time during the product traction process, so as to obtain real-time tension information during the production process. A dielectric sensor embedded in the preforming stage is used to collect real-time dielectric signals during the production process; An acoustic emission probe is installed at the traction clamp of the traction machine to collect real-time acoustic emission signals during the production process; Infrared thermal imagers are installed at the production site to collect real-time infrared thermal field information during the production process. The data integration unit is used to integrate all real-time data to construct the real-time generated raw data in the production process.

[0041] In this example, the mold temperature distribution represents the temperature values ​​at different locations of the mold, such as the mold inlet, mold middle section, and mold outlet, during the pultrusion molding process of composite materials, as well as the temperature change trends between each location and the overall temperature distribution state. It is a key process environment parameter that affects the curing reaction of composite materials.

[0042] In this example, the product curing process refers to the entire process in pultrusion production where, after the resin-impregnated fiber composite material enters the mold, the resin undergoes a series of physical and chemical changes under the action of the mold temperature, gradually curing and hardening from the initial viscous flow state to finally forming a composite material product with a predetermined cross-sectional shape and stable mechanical properties.

[0043] The working principle and beneficial effects of the above technical solution are as follows: By strategically deploying various sensors at key nodes in the pultrusion molding production process, temperature sensors are installed in different sections of the mold, vision sensors are arranged at key locations during pre-forming and post-forming, and tension sensors and acoustic emission probes are adapted and installed at the traction fixture. Simultaneously, dielectric sensors are embedded in the pre-forming stage, and infrared thermal imagers are set up in the production area. This enables accurate and real-time acquisition of multi-dimensional data on temperature, video, tension, dielectric, acoustic emission, and infrared thermal field during the production process. Each sensor performs its specific function, and the acquisition range covers the entire production process without blind spots. Furthermore, the data integration unit unifies and integrates all real-time acquired data to construct real-time raw production data. This ensures the comprehensiveness, relevance, and real-time nature of the acquired data, and provides complete and accurate raw data support for subsequent quality assessment, defect prediction, and process parameter control. It effectively avoids problems such as misjudgment of production status and missed defect detection caused by single data acquisition or scattered data, significantly improving the effectiveness and scientific nature of production data acquisition.

[0044] Example 3: Based on Example 2, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality further includes: The acquisition setting unit is used to adjust the acquisition frequency of the first temperature sensor, the second temperature sensor and the third temperature sensor to 1 second / time; Set the sampling frequency of the tension sensor to 1 second / time; The sampling frequency of the dielectric sensor is set to 1 second / time; The acquisition frequency of the infrared thermal imager is set to 50 sets of thermal field distribution maps per second.

[0045] The working principle and beneficial effects of the above technical solution are as follows: Suitable acquisition frequencies are set for various sensors, with temperature and dielectric sensors set to 1 second / time, tension sensors set to 1 second / time, and infrared thermal imagers set to 50 sets of thermal field distribution maps / second. This allows for the capture of faster-changing tension data at a higher frequency, ensuring accurate acquisition of subtle data such as stress changes.

[0046] Example 4: Based on Example 1, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality includes a modeling preprocessing module comprising: The data filtering unit is used to acquire field parameters in the production site, deduce several production conditions in the production site, and filter sensor data, process parameter data and product quality test results corresponding to each production condition from Internet big data. The data aggregation unit is used to aggregate the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. The dataset partitioning unit is used to label the training dataset according to the defects in the working conditions, and to divide the labeled training dataset into a training set and a validation set in a ratio of 7:3.

[0047] In this example, the field parameters refer to various parameters that define the basic production conditions in the pultrusion molding production site of composite materials, including core parameters such as product type, mold specifications, and raw material characteristics.

[0048] In this example, the sensor data represents multi-dimensional monitoring data collected by various multi-source sensors during the pultrusion molding process of composite materials, including real-time temperature information, real-time video information, and real-time tension information.

[0049] In this example, the process parameter data represents various process-related data used to control the production process during the pultrusion molding of composite materials, including parameters such as mold temperature, traction speed, injection pressure, and resin viscosity.

[0050] In this example, the product quality inspection results represent the product quality index test data and the product defect judgment results obtained after the composite material pultruded product has undergone professional testing methods such as non-destructive testing and destructive testing.

[0051] In this example, "condition defects" refers to various product defects that are likely to occur or actually occur under a specific production condition in the pultrusion molding of composite materials, such as poor fiber impregnation, internal porosity, delamination and debonding, and white filament defects.

[0052] In this example, defect labeling refers to the operation of adding corresponding feature labels such as defect type and defect severity to the relevant historical production data in the training dataset based on the defects corresponding to each production condition.

[0053] The working principle and beneficial effects of the above technical solution are as follows: To significantly improve the relevance, effectiveness, and accuracy of model training, the actual production conditions are first derived based on the on-site parameters of the production site. This allows for precise identification of various data that match the current production site. From the big data, sensor data, process parameter data, and product quality inspection results corresponding to each condition are selectively filtered, effectively eliminating invalid data irrelevant to the current production. This avoids interference from irrelevant data in subsequent model construction, ensuring that the raw data for subsequent data processing highly matches the actual production site, thus improving the relevance and effectiveness of the data. Finally, various relevant data under the same production condition are uniformly aggregated. The system generates corresponding historical production data, enabling the classification, integration, and systematic organization of various types of production data. The constructed training dataset is both complete and specific to different operating conditions. It also clearly identifies the defects corresponding to each production condition, establishing a direct correlation between production data and actual defects in production. This provides a clear and accurate basis for subsequent defect labeling, allowing the training dataset to reflect actual production defect patterns. Finally, defect labeling is performed on the training dataset based on operating condition defects, giving the data clear quality characteristics and defect information. This provides a clear label basis for the fusion model to learn defect patterns under different operating conditions, helping the model accurately identify defect features. The training and validation sets are divided in a fixed 7:3 ratio, ensuring a scientific fit between the two datasets. This guarantees sufficient labeled data for the fusion model's training while effectively verifying the model's training performance through the validation set. This standardization of data processing provides standardized and reasonable dataset support for subsequent model training and parameter adjustment.

[0054] Example 5: Based on Example 1, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality includes a model building module comprising: The reasoning building unit is used to build a physical model based on the on-site parameters in the production site, the principles of curing reaction kinetics, heat and mass transfer laws, and the mechanical properties of materials. The model training unit is used to train the physical model using the training set to obtain several parameters of the physical model to be adjusted, to determine the parameter value corresponding to each parameter to be adjusted using the validation set, and to adjust the parameters of the physical model to obtain an adjusted model. The functional verification unit is used to construct functional test data using the training set and the verification set, and to perform functional verification on the adjusted model. When the defect prediction accuracy of the fusion model is ≥95% and the false alarm rate is ≤2.5%, training is stopped to obtain the fusion model.

[0055] In this example, the principle of curing reaction kinetics represents the kinetic theory that studies the quantitative relationship between the reaction rate and degree of curing of resin and process parameters such as temperature, time, and pressure during the pultrusion process of composite materials. It is the core principle for revealing the law of the resin from viscous flow to curing and hardening process in pultrusion, and it is also an important theoretical basis for clarifying the relationship between mold temperature and product curing degree when constructing physical models.

[0056] In this example, the heat and mass transfer law represents the basic law of heat transfer between the mold and the composite material, and within the composite material, as well as the migration and diffusion of various substances in the material system during the pultrusion molding process of composite materials. It is an important theoretical basis for controlling the relationship between the temperature distribution of the mold and the curing state of the product.

[0057] In this example, the mechanical properties of materials represent the inherent mechanical properties and variation laws of the fibers, resins and impregnated composite materials used in composite pultrusion under production conditions such as stress and heat, including strength, stiffness, stress-strain, and adhesion. They are an important basis for analyzing the correlation between process parameters such as tension and defects such as delamination and debonding inside the product when constructing a physical model.

[0058] In this example, the parameters to be adjusted represent various model parameters that have not yet been optimized after the physical model has been trained on the training set and need to be further optimized by combining actual data from the production site.

[0059] In this example, "adjusting the model" means that the physical model is processed by the model training unit, and each parameter to be adjusted is replaced with the optimal parameter value determined by the validation set, resulting in a model that has completed the initial parameter optimization.

[0060] In this example, the functional test data refers to the dataset formed by integrating the training set and validation set obtained in the modeling preprocessing stage, which is used to comprehensively test and adjust the core functional performance of the model, such as defect prediction and quality assessment.

[0061] In this example, functional verification means inputting functional test data into the adjustment model and verifying whether the adjustment model has the ability to accurately complete the production quality assessment and defect prediction functions by testing the model's core performance indicators such as defect prediction accuracy and false alarm rate. The functional verification is considered to be passed only when the model's defect prediction accuracy is ≥95% and the false alarm rate is ≤2.5%.

[0062] The working principle and beneficial effects of the above technical solution are as follows: To provide an accurate, reliable, and highly adaptable model, firstly, a physical model is constructed by combining on-site parameters from the production site with core physical knowledge of pultrusion molding processes, such as curing reaction kinetics, heat and mass transfer laws, and material mechanical properties. This ensures that the physical model is not a generic, generalized model, but rather precisely adapted to the actual working conditions of the current production site, including product type and mold specifications. Simultaneously, it relies on professional physical theories to form a solid theoretical support, clearly defining the quantitative relationship between process parameters such as mold temperature and traction speed and the internal curing degree and surface quality of the product. This provides a basic model framework for subsequent model training and parameter adjustment. Then, a training set is used for training, and a validation set is used for parameter determination. The physical model is adjusted using a step-by-step optimization approach. First, the correlation between production data and product quality / defects is mined from the training set to extract parameters requiring optimization. Then, the optimal parameter values ​​for each parameter are determined using the validation set, ensuring that the entire parameter adjustment process is supported by standardized real-world production data. This avoids model bias caused by blind parameter tuning, allowing the adjusted model to accurately fit the characteristics of the current production dataset. This effectively improves the model's ability to analyze production data and its adaptability to actual production. Finally, the training and validation sets are integrated to construct complete functional test data for comprehensive functional verification of the adjusted model, with a defect prediction accuracy target of ≥95%. Furthermore, a clear and stringent performance threshold of ≤2.5% for false positive rate is set. Training is stopped only when this threshold is reached, and the adjusted model is determined to be the fusion model. The final performance of the model is controlled through unified and strict verification standards, which fundamentally ensures that the fusion model put into use has the characteristics of high accuracy and low false positive rate in defect prediction. This effectively avoids problems such as production quality assessment deviation, false positive and false negative defect judgment caused by the use of substandard models, thus ensuring the practicality and reliability of the model.

[0063] Example 6: Based on Example 1, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality, including the evaluation and prediction module, comprises: The data processing unit is used to mine the nonlinear relationships of the real-time multi-source raw data using gradient boosting trees, perform feature enhancement on the real-time multi-source raw data using support vector machines, and input the nonlinear relationships and the enhanced real-time multi-source raw data into the fusion model. The quality assessment unit is used to recreate the finished product in the production process in real time in the fusion model, analyze the internal curing, surface flatness and cross-sectional dimensions of the finished product, and generate quality assessment results. The defect prediction unit is used to analyze abnormal changes in the production process using the fusion model, analyze the defects and their severity corresponding to the abnormal changes, and generate defect prediction results. The strategy production unit is used to determine the process parameters that need to be adjusted in the production process based on the quality assessment results and the defect prediction results, and to generate a process parameter adjustment strategy in combination with the production execution equipment corresponding to each process parameter that needs to be adjusted.

[0064] In this example, nonlinear relationships represent the nonlinear interrelationships between real-time multi-source raw data in composite material pultrusion production, such as the complex dynamic relationships between mold temperature and product curing degree, and traction tension and the risk of internal delamination in the product.

[0065] In this example, feature enhancement means that the support vector machine algorithm is used to process real-time multi-source raw data to enhance the identification of feature information related to product defects and quality anomalies in the data, weaken invalid and interfering information, and make it easier for the fusion model to capture the processing of quality and defect-related features in the data.

[0066] In this example, the finished product representation evaluation and prediction module relies on a fusion model to simulate and reconstruct the virtual state of the composite material product that will be produced during the pultrusion molding process based on real-time multi-source raw data.

[0067] In this example, internal curing refers to the degree to which the resin inside the composite pultruded product undergoes a curing reaction under the influence of process conditions such as mold temperature, as well as the uniformity and integrity of the cured product. It is a core indicator for measuring the internal quality of the product.

[0068] In this example, surface smoothness refers to the smoothness and flatness of the surface of the composite pultruded product, and the quality indicator of whether there are surface defects such as bumps, scratches, and white streaks.

[0069] In this example, the cross-sectional dimensions represent the quality indicators of whether the actual cross-sectional specifications and dimensional parameters of the composite pultruded product after preforming and curing conform to the preset product cross-sectional design requirements.

[0070] In this example, abnormal changes refer to changes in the real-time multi-source raw data during the pultrusion molding process of composite materials, which deviate from the data fluctuation range under normal production conditions, such as sudden large fluctuations in temperature, sudden changes in traction tension, and abnormal textures in visual images.

[0071] In this example, defects and defect severity refer to quality problems such as poor fiber impregnation, internal porosity, delamination and debonding, and white threads that occur in the product during the pultrusion molding process of composite materials due to improper process parameters, as well as the severity, scope of impact, and location of these quality problems.

[0072] In this example, the defect prediction result represents the comprehensive result output by the defect prediction unit after analysis using the fusion model, which includes information such as the type of defect, the degree of defect, the location of defect, and the probability of defect occurrence in the production process.

[0073] In this example, the process parameters that need to be adjusted refer to the pultrusion molding process parameters that are determined to cause substandard product quality or production defects based on the quality assessment results and defect prediction results. These parameters mainly include mold temperature and traction speed, as well as auxiliary parameters such as injection pressure and resin viscosity.

[0074] The working principle and beneficial effects of the above technical solution are as follows: To achieve seamless integration from production data to quality analysis and control strategies, making quality monitoring and defect response in pultrusion production more timely and effective, two algorithms, gradient boosting tree and support vector machine, are specifically used to differentiate the real-time multi-source raw data. Gradient boosting tree can accurately uncover complex nonlinear relationships between data and capture potential correlations behind production data, while support vector machine can perform feature enhancement processing on the raw data, strengthening the identification of defect-related features. The results of both processing are input into the fusion model, which not only enriches the analytical dimensions of the data but also improves the adaptability of the data to the fusion model, laying a high-quality data foundation for subsequent quality assessment and defect prediction, and avoiding the bias caused by direct analysis of raw data. Then, the fusion model is used to recreate the finished product state in real time during the production process, focusing on the three core quality indicators of pultruded composite products—internal curing, surface smoothness, and cross-sectional dimensions—to conduct precise analysis and generate quantitative results. The quality assessment results can intuitively and comprehensively reflect the actual quality status of the product. The assessment process is based on real-time production data, realizing online, real-time, and non-destructive assessment of product quality. Furthermore, by utilizing a fusion model, abnormal data changes during the production process are accurately captured and deeply analyzed. This allows for accurate determination of the defect type and severity corresponding to abnormal changes and the generation of defect prediction results. This not only enables early prediction and precise location of defects but also clarifies the severity of defects, providing a basis for prioritizing subsequent process parameter adjustments. This significantly reduces the rate of missed defects and false alarms. Finally, by combining the quality assessment results and defect prediction results, the process parameters that need adjustment during the production process are scientifically determined. At the same time, the corresponding production execution equipment for each parameter is precisely matched. The generated process parameter adjustment strategy is both targeted and operable. This avoids production quality fluctuations caused by blindly adjusting process parameters and provides clear execution targets and operation objects for subsequent parameter adjustment work of the feedback control module, improving the efficiency and accuracy of process control.

[0075] Example 7: Based on Example 1, the real-time monitoring and closed-loop control system for composite material pultrusion molding quality includes a feedback control module comprising: The sequence determination unit is used to identify the influence weight of each process parameter to be adjusted in the process parameter adjustment strategy on the production process, and determine the parameter adjustment sequence of this parameter adjustment work according to the order of production weight from high to low. The step size determination unit is used to obtain the parameter deviation value and parameter limit threshold corresponding to each of the production actuators, and determine the parameter adjustment step size corresponding to each of the production actuators in the current adjustment work by combining the parameter adjustment amount corresponding to the process parameter adjustment strategy. The monitoring unit is used to dynamically and iteratively adjust each of the production execution devices according to the parameter adjustment sequence and the parameter adjustment step size, and at the same time obtain the production feedback parameters of the current adjustment work, and determine the real-time production defects of the production process based on the production feedback parameters. The continuous monitoring unit is used to end the current adjustment work when the real-time production defect disappears, determine that the production site has entered a high-quality production state, and at the same time use the production feedback data to analyze and display the real-time production quality report of the production process.

[0076] In this example, the parameter adjustment order refers to the order in which the process parameters are adjusted, determined by ranking the weight of each process parameter to be adjusted in relation to the production process from highest to lowest.

[0077] In this example, the parameter deviation value represents the difference between the actual operating parameter values ​​of the production actuator and the standard parameter values ​​required for pultrusion production.

[0078] In this example, the parameter limit threshold represents the minimum production threshold value based on the production of pultruded products. It is the critical value of the process parameter obtained by abbreviating the statistical results of the normal process parameter fluctuation range on both sides.

[0079] In this example, the parameter adjustment step size represents the specific value of each adjustment to the process parameter calculated and determined by the step size determination unit based on the parameter deviation value and parameter limit threshold of the production actuator, combined with the corresponding parameter adjustment amount in the process parameter adjustment strategy.

[0080] In this example, real-time production defects refer to product quality defects that still exist in the current production process, as determined by the adjustment monitoring unit based on the collected production feedback parameters after the parameters of the production execution equipment have been adjusted.

[0081] In this example, a high-quality production status indicates that real-time production defects during the production process have completely disappeared, and all key quality indicators such as internal curing, surface smoothness, and cross-sectional dimensions of the product have met the production standard requirements for pultrusion production operation.

[0082] The working principle and beneficial effects of the above technology are as follows: To effectively avoid the production of batches of defective products and ensure high-quality operation throughout the pultrusion molding process, the impact weight of the process parameters to be adjusted on the production process is first identified. The parameters are then adjusted in descending order of weight, allowing the process parameter control to prioritize issues that play a core role in production quality and defect prevention. This avoids the problems of low control efficiency and unresolved core defects caused by disordered adjustments, improving the targeting and effectiveness of process parameter adjustments. Then, by obtaining the parameter deviation values ​​and parameter limit thresholds of the production actuators, and combining this with the amount of process parameter adjustment, a reasonable parameter adjustment step size is determined. This ensures that the process parameter adjustments are always within a reasonable range, avoiding parameter abrupt changes caused by excessively large single adjustments. This effectively prevents new production quality problems caused by parameter abrupt changes. The system dynamically iterates and adjusts the production actuators according to a predetermined parameter adjustment sequence and step size. Simultaneously, it acquires real-time production feedback parameters and identifies real-time production defects, achieving full real-time monitoring of the parameter adjustment process. This allows for timely understanding of the actual effect of each adjustment step, accurate identification of existing defects in the current production process, and the final determination of whether real-time production defects have disappeared. This ensures that quality defects in the production process are completely eliminated, ultimately leading to high-quality production. Furthermore, the system uses production feedback data to generate and display real-time production quality reports, achieving visualized monitoring of production quality, facilitating staff understanding of the production status, and preserving complete data for subsequent process optimization and production analysis. This forms a closed-loop optimization of process control, further ensuring the stability of production quality.

[0083] Example 8: This example provides a method for real-time monitoring and closed-loop control of composite material pultrusion molding quality, such as... Figure 2 As shown, it includes: Step 1: Collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal, and real-time infrared thermal field information during the production process to construct real-time multi-source raw data; Step 2: Filter several historical production data related to the production process, and construct a training set and a validation set by combining the production condition type corresponding to each historical production data. Step 3: Construct a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model; Step 4: Input the real-time multi-source raw data into the fusion model for production quality assessment and defect prediction, and generate process parameter adjustment strategies; Step 5: Use the process parameter adjustment strategy to adjust the parameters of the production execution equipment to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

[0084] In this example, the real-time temperature information refers to real-time data collected by temperature sensors deployed at key nodes in the pultrusion production process, which reflects the temperature distribution of the mold and the temperature changes during the curing process of the composite material product.

[0085] In this example, real-time video information refers to real-time video image data collected by vision sensors deployed at locations such as after the preforming stage, at the mold exit, and after the traction machine, which can reflect the cross-sectional shape of the product after preforming and the surface state after curing.

[0086] In this example, the real-time tension information refers to the real-time tension data collected by the tension sensor installed at the traction clamp of the traction machine, which reflects the stress changes experienced by the product during traction.

[0087] In this example, the real-time dielectric signal refers to real-time signal data collected by a dielectric sensor embedded in the preforming stage, which reflects the changes in the dielectric properties of the material during the pultrusion production process.

[0088] In this example, the real-time acoustic emission signal refers to the real-time acoustic emission data collected by the acoustic emission probe integrated into the traction clamp of the traction machine, which can reflect the changes in internal stress, delamination and debonding of the product during the production process.

[0089] In this example, the real-time infrared thermal field information refers to real-time data collected by infrared thermal imagers covering key monitoring locations throughout the entire production process, which reflects the distribution of infrared thermal fields at each key node in the production process.

[0090] In this example, the real-time multi-source raw data refers to the multi-dimensional, integrated real-time production data set formed by the multi-source sensor acquisition module after integrating the real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal and real-time infrared thermal field information collected in the pultrusion production process.

[0091] In this example, historical production data refers to the total sensor data, process parameter data, and corresponding product quality inspection results generated under different production conditions during the pultrusion molding process.

[0092] In this example, the production condition type refers to the different production operation status types classified according to different conditions such as the product type, mold specifications, raw material characteristics, and process parameters of pultrusion production.

[0093] In this example, the relevant physical knowledge refers to the theoretical knowledge in the physical fields related to the curing reaction kinetics, heat and mass transfer laws, and material mechanical properties associated with the pultrusion molding process of composite materials.

[0094] In this example, the physical model represents a theoretical model constructed based on relevant physical knowledge of pultrusion production, which can clearly define the quantitative relationship between process parameters such as mold temperature, traction speed, and tension and the degree of internal curing and surface quality of the product.

[0095] In this example, the training set refers to the dataset used to train the physical model after the historical production data selected by the modeling preprocessing module is divided according to a certain proportion.

[0096] In this example, the validation set refers to the dataset used to adjust the parameters and verify the effect of the trained physical model after the historical production data selected by the modeling preprocessing module is divided according to a certain proportion.

[0097] In this example, the fusion model represents a comprehensive evaluation model that combines the support of physical theory and the ability to analyze machine learning data, formed by training a physical model based on relevant physical knowledge using a training set and adjusting parameters using a validation set.

[0098] In this example, the process parameter adjustment strategy refers to the specific control plan formulated by the evaluation and prediction module based on the production quality evaluation results and defect prediction results. This plan includes the process parameters to be adjusted, the adjustment sequence, the adjustment step size, and the corresponding production execution equipment.

[0099] In this example, production execution equipment refers to various types of equipment that can execute process parameter adjustment instructions during the pultrusion molding process of composite materials, including mold temperature control systems, traction machines, and glue injection equipment.

[0100] In this example, the production feedback data represents the real-time data of the pultrusion production process that is re-collected by the multi-source sensor acquisition module after the parameters of the production actuators have been adjusted according to the process parameter adjustment strategy.

[0101] In this example, the real-time production quality report is a visual report generated by the feedback control module after completing the quality evaluation based on the production feedback data. It includes information such as production quality status, defect status, process parameter adjustment process, and quality evaluation results.

[0102] The working principle and beneficial effects of the above technical solution are as follows: To effectively improve product yield and production efficiency while reducing production costs, real-time production data from multiple dimensions, including temperature, video, tension, dielectric, acoustic emission, and infrared thermal field, are simultaneously collected at key nodes in the pultrusion molding production process. This data is integrated to form unified real-time multi-source raw data, achieving blind-spot-free acquisition of multi-physics field data throughout the entire production process. This provides comprehensive, accurate, and real-time raw data support for subsequent quality assessment, defect prediction, and model building, overcoming the limitations of traditional single-data acquisition. Then, historical production data highly matching the current production process is precisely selected. The data is standardized based on production condition types, and training and validation sets are constructed. This effectively avoids interference from invalid data in model building, providing a standardized data source that fits the actual production scenario for model training and parameter adjustment. This ensures the relevance and accuracy of subsequent model building. Furthermore, based on the relevant physical properties of the production process... The knowledge-based physical model is trained and its parameters adjusted using training and validation sets. The resulting fusion model combines the support of physical theory with the data-driven nature of machine learning, accurately meeting the actual needs of pultrusion production. This improves the model's accuracy in production quality assessment and defect prediction, laying a reliable model foundation for subsequent assessment and prediction work. Further inputting real-time multi-source raw data into the fusion model enables real-time quantitative assessment of production quality and accurate prediction of product defects. Simultaneously, based on the assessment and prediction results, targeted process parameter adjustment strategies are generated, providing a clear and scientific basis for subsequent process control. This solves the problems of strong lag and high false negative rate in traditional detection methods, enabling early prediction and precise policy implementation of production anomalies. Finally, precise parameter adjustments are made to production actuators based on the process parameter adjustment strategies. Production feedback data after parameter adjustments is collected and quality evaluation is completed, generating and displaying a real-time production quality report, forming an "assessment- The complete closed-loop control of "control-verification" enables dynamic iterative optimization of process parameters, timely elimination of defects in the production process, avoidance of batch production of unqualified products, and visualization of quality reports facilitates real-time monitoring of production status by staff, while providing complete data reference for subsequent process optimization analysis.

[0103] Example 9: Based on Example 8, the real-time monitoring and closed-loop control method for the quality of composite material pultrusion molding, step 2 includes: Step 21: Obtain the field parameters in the production site, deduce several production conditions in the production site, and filter the sensor data, process parameter data and product quality test results corresponding to each production condition in the Internet big data. Step 22: Summarize the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. Step 23: Label the training dataset according to the defects in the working conditions, and divide the labeled training dataset into a training set and a validation set in a ratio of 7:3.

[0104] In this example, the field parameters refer to various parameters that define the basic production conditions in the pultrusion molding production site of composite materials, including core parameters such as product type, mold specifications, and raw material characteristics.

[0105] In this example, the sensor data represents multi-dimensional monitoring data collected by various multi-source sensors during the pultrusion molding process of composite materials, including real-time temperature information, real-time video information, and real-time tension information.

[0106] In this example, the process parameter data represents various process-related data used to control the production process during the pultrusion molding of composite materials, including parameters such as mold temperature, traction speed, injection pressure, and resin viscosity.

[0107] In this example, the product quality inspection results represent the product quality index test data and the product defect judgment results obtained after the composite material pultruded product has undergone professional testing methods such as non-destructive testing and destructive testing.

[0108] In this example, "condition defects" refers to various product defects that are likely to occur or actually occur under a specific production condition in the pultrusion molding of composite materials, such as poor fiber impregnation, internal porosity, delamination and debonding, and white filament defects.

[0109] In this example, defect labeling refers to the operation of adding corresponding feature labels such as defect type and defect severity to the relevant historical production data in the training dataset based on the defects corresponding to each production condition.

[0110] The working principle and beneficial effects of the above technical solution are as follows: To significantly improve the relevance, effectiveness, and accuracy of model training, the actual production conditions are first derived based on the on-site parameters of the production site. This allows for precise identification of various data that match the current production site. From the big data, sensor data, process parameter data, and product quality inspection results corresponding to each condition are selectively filtered, effectively eliminating invalid data irrelevant to the current production. This avoids interference from irrelevant data in subsequent model construction, ensuring that the raw data for subsequent data processing highly matches the actual production site, thus improving the relevance and effectiveness of the data. Finally, various relevant data under the same production condition are uniformly aggregated. The system generates corresponding historical production data, enabling the classification, integration, and systematic organization of various types of production data. The constructed training dataset is both complete and specific to different operating conditions. It also clearly identifies the defects corresponding to each production condition, establishing a direct correlation between production data and actual defects in production. This provides a clear and accurate basis for subsequent defect labeling, allowing the training dataset to reflect actual production defect patterns. Finally, defect labeling is performed on the training dataset based on operating condition defects, giving the data clear quality characteristics and defect information. This provides a clear label basis for the fusion model to learn defect patterns under different operating conditions, helping the model accurately identify defect features. The training and validation sets are divided in a fixed 7:3 ratio, ensuring a scientific fit between the two datasets. This guarantees sufficient labeled data for the fusion model's training while effectively verifying the model's training performance through the validation set. This standardization of data processing provides standardized and reasonable dataset support for subsequent model training and parameter adjustment.

[0111] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A real-time monitoring and closed-loop control system for the quality of the pultrusion of composite materials, characterized in that, include: The multi-source sensor acquisition module is used to collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal and real-time infrared thermal field information during the production process, and to construct real-time multi-source raw data; The modeling preprocessing module is used to filter several historical production data related to the production process and construct training and validation sets by combining the production condition type corresponding to each historical production data. The model building module is used to build a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model. The evaluation and prediction module is used to input the real-time multi-source raw data into the fusion model to perform production quality evaluation and defect prediction, and generate process parameter adjustment strategies. The feedback control module is used to adjust the parameters of the production execution equipment using the process parameter adjustment strategy to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

2. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 1, characterized in that, The multi-source sensing acquisition module includes: A first temperature sensor is installed at the mold inlet, a second temperature sensor is installed in the middle section of the mold, and a third temperature sensor is installed at the mold outlet. The first temperature sensor, the second temperature sensor, and the third temperature sensor are used to monitor the mold temperature distribution and temperature change information during the product curing process in real time, so as to obtain real-time temperature information during the production process. A first vision sensor is installed at the mold exit, a second vision sensor is installed behind the traction machine, and a third vision sensor is installed behind the preforming stage. The first vision sensor, the second vision sensor, and the third vision sensor are used to monitor the cross-sectional shape after preforming and the surface state after curing in real time to obtain real-time video information during the production process. A tension sensor installed at the traction clamp of the traction machine is used to collect tension data in real time during the product traction process, so as to obtain real-time tension information during the production process. A dielectric sensor embedded in the preforming stage is used to collect real-time dielectric signals during the production process; An acoustic emission probe is installed at the traction clamp of the traction machine to collect real-time acoustic emission signals during the production process; Infrared thermal imagers are installed at the production site to collect real-time infrared thermal field information during the production process. The data integration unit is used to integrate all real-time data to construct the real-time generated raw data in the production process.

3. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 2, characterized in that, Also includes: The acquisition setting unit is used to adjust the acquisition frequency of the first temperature sensor, the second temperature sensor and the third temperature sensor to 1 second / time; Set the sampling frequency of the tension sensor to 1 second / time; The sampling frequency of the dielectric sensor is set to 1 second / time; The acquisition frequency of the infrared thermal imager is set to 50 sets of thermal field distribution maps per second.

4. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 1, characterized in that, The modeling preprocessing module includes: The data filtering unit is used to acquire field parameters in the production site, deduce several production conditions in the production site, and filter sensor data, process parameter data and product quality test results corresponding to each production condition from Internet big data. The data aggregation unit is used to aggregate the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. The dataset partitioning unit is used to label the training dataset according to the defects in the working conditions, and to divide the labeled training dataset into a training set and a validation set in a ratio of 7:

3.

5. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 1, characterized in that, The model building module includes: The reasoning building unit is used to build a physical model based on the on-site parameters in the production site, the principles of curing reaction kinetics, heat and mass transfer laws, and the mechanical properties of materials. The model training unit is used to train the physical model using the training set to obtain several parameters of the physical model to be adjusted, to determine the parameter value corresponding to each parameter to be adjusted using the validation set, and to adjust the parameters of the physical model to obtain an adjusted model. The functional verification unit is used to construct functional test data using the training set and the verification set, and to perform functional verification on the adjusted model. When the defect prediction accuracy of the fusion model is ≥95% and the false alarm rate is ≤2.5%, training is stopped to obtain the fusion model.

6. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 1, characterized in that, The evaluation and prediction module includes: The data processing unit is used to mine the nonlinear relationships of the real-time multi-source raw data using gradient boosting trees, perform feature enhancement on the real-time multi-source raw data using support vector machines, and input the nonlinear relationships and the enhanced real-time multi-source raw data into the fusion model. The quality assessment unit is used to recreate the finished product in the production process in real time in the fusion model, analyze the internal curing, surface flatness and cross-sectional dimensions of the finished product, and generate quality assessment results. The defect prediction unit is used to analyze abnormal changes in the production process using the fusion model, analyze the defects and their severity corresponding to the abnormal changes, and generate defect prediction results. The strategy production unit is used to determine the process parameters that need to be adjusted in the production process based on the quality assessment results and the defect prediction results, and to generate a process parameter adjustment strategy in combination with the production execution equipment corresponding to each process parameter that needs to be adjusted.

7. The real-time monitoring and closed-loop control system for composite material pultrusion molding quality as described in claim 1, characterized in that, The feedback control module includes: The sequence determination unit is used to identify the influence weight of each process parameter to be adjusted in the process parameter adjustment strategy on the production process, and determine the parameter adjustment sequence of this parameter adjustment work according to the order of production weight from high to low. The step size determination unit is used to obtain the parameter deviation value and parameter limit threshold corresponding to each of the production actuators, and determine the parameter adjustment step size corresponding to each of the production actuators in the current adjustment work by combining the parameter adjustment amount corresponding to the process parameter adjustment strategy. The monitoring unit is used to dynamically and iteratively adjust each of the production execution devices according to the parameter adjustment sequence and the parameter adjustment step size, and at the same time obtain the production feedback parameters of the current adjustment work, and determine the real-time production defects of the production process based on the production feedback parameters. The continuous monitoring unit is used to end the current adjustment work when the real-time production defect disappears, determine that the production site has entered a high-quality production state, and at the same time use the production feedback data to analyze and display the real-time production quality report of the production process.

8. A method for real-time monitoring and closed-loop control of composite material pultrusion molding quality, characterized in that, include: Step 1: Collect real-time temperature information, real-time video information, real-time tension information, real-time dielectric signal, real-time acoustic emission signal, and real-time infrared thermal field information during the production process to construct real-time multi-source raw data; Step 2: Filter several historical production data related to the production process, and construct a training set and a validation set by combining the production condition type corresponding to each historical production data. Step 3: Construct a physical model based on the relevant physical knowledge of the production process, train the physical model using the training set, and adjust the parameters of the physical model using the validation set to generate a fusion model; Step 4: Input the real-time multi-source raw data into the fusion model for production quality assessment and defect prediction, and generate process parameter adjustment strategies; Step 5: Use the process parameter adjustment strategy to adjust the parameters of the production execution equipment to obtain production feedback data, evaluate the quality of the production feedback data, obtain a real-time production quality report, and display it.

9. The method for real-time monitoring and closed-loop control of composite material pultrusion molding quality as described in claim 8, characterized in that, Step 2 includes: Step 21: Obtain the field parameters in the production site, deduce several production conditions in the production site, and filter the sensor data, process parameter data and product quality test results corresponding to each production condition in the Internet big data. Step 22: Summarize the sensor data, process parameter data and product quality inspection results corresponding to the same production condition to obtain historical production data corresponding to each production condition, construct the training dataset of the production site, and determine the condition defects corresponding to each production condition. Step 23: Label the training dataset according to the defects in the working conditions, and divide the labeled training dataset into a training set and a validation set in a ratio of 7:3.

Images