Carbon fiber prepreg production management and control system based on process parameter dynamic collaboration

Abstract

The invention discloses a carbon fiber prepreg production management and control system based on process parameter dynamic collaboration, and relates to the technical field of composite material production, the management and control system comprises a data acquisition module, a dynamic collaborative analysis module, a management and control instruction generation module, an execution module and a quality monitoring feedback module. According to the carbon fiber prepreg production management and control system based on process parameter dynamic collaboration, the health state of the sensor is monitored in real time, when the health degree is lower than a threshold value or parameter deviation exceeds the standard, a calibration instruction is generated by automatically combining the temperature and humidity of a workshop, the whole calibration process does not need to be stopped, and consumed time is controlled within a certain period of time; according to the scheme, the problem of economic loss caused by existing shutdown calibration can be avoided; secondly, by means of a multi-physical field and digital twinborn fusion model, the coupling relation of a temperature field, a flow field, a stress field and a chemical field is integrated, the parameter change trend is simulated, and when it is predicted that parameters such as permeability are lower than the standard, the coupling source can be clearly marked;

Description

Technical Field

[0001] This invention relates to the field of composite material production technology, and in particular to a carbon fiber prepreg production control system based on dynamic coordination of process parameters. Background Technology

[0002] Carbon fiber prepreg is a core intermediate material composed of carbon fiber and resin matrix. With its superior properties such as high specific strength, high specific modulus, and fatigue resistance, it has become a key basic material in strategic fields such as aerospace, high-end equipment, and new energy vehicles. Its production process requires uniformly impregnating carbon fiber bundles with resin, followed by rolling and shaping to form rolls of specific thickness and resin content. This involves multiple physicochemical processes, including resin rheology and fiber impregnation, requiring extremely high precision in the dynamic coordination of process parameters. Even minute temperature deviations or pressure fluctuations can lead to quality defects such as uneven resin content and insufficient impregnation in the prepreg, directly affecting the mechanical properties of the end product. As carbon fiber prepreg upgrades towards high purity, high stability, and large-scale production, the technical limitations of existing control systems are becoming increasingly apparent. Currently, the mainstream production control solutions in the industry are mainly divided into four categories, all of which have revealed significant shortcomings in actual implementation. (I) Single-Physical-Field Static Control Scheme: This scheme focuses on the temperature field as the core control object, collecting parameters solely through temperature sensors on the rolling rollers and hot plates. Adjustments are made in a single dimension based on preset fixed temperature thresholds. Specifically, if the temperature of the hot plate or hot roller falls below the lower threshold, the heating power is increased; if it rises above the upper threshold, the cooling fan is activated. Other parameters, such as rolling pressure and traction speed, are handled manually, without establishing logical relationships between parameters. Its drawbacks include: firstly, ignoring the coupling effects of multiple physical fields. In actual production, resin viscosity decreases as impregnation temperature increases, thus accelerating resin flow in the carbon fiber bundle. Adjusting only the temperature without matching the traction speed is problematic. There are several issues with impregnation. First, over- or under-impregnation can occur. For example, a batch of prepreg might meet the temperature requirements of the hot plate and hot rollers, but the impregnation of the prepreg is still insufficient. Tracing back reveals that the excessive traction speed caused the resin to stay in the hot plate and hot rollers for a shorter time, but the system failed to recognize this coupling relationship. Second, it is difficult to pinpoint the root cause of the deviation. When the prepreg thickness exceeds the standard, the system cannot distinguish whether it is due to insufficient rolling pressure, excessively low resin viscosity, or excessively slow traction speed. It can only rely on manual inspection, which seriously affects production efficiency. As for impregnation, it is necessary to collect parameters through temperature sensors on the rolling rollers and hot plate to detect whether the temperature meets the standard or whether there are cases where the temperature meets the standard but the impregnation is still insufficient.

[0003] (II) Single-Objective Weighted Control Scheme: This scheme takes quality compliance as the sole core objective and sets fixed optimization weights. Regardless of the type of production order or energy consumption quota restrictions, it prioritizes ensuring quality by increasing heating power. For example, to ensure that the impregnation of high-end prepreg meets the standards, even if the current energy consumption is close to the daily quota limit of the workshop, the heating power of the hot plate and hot roller will continue to be increased. Its disadvantages are: first, serious waste of resources. Controlling civilian-grade prepreg according to high-end standards will lead to quality overload, which will not only increase energy consumption, but may also reduce the activity of prepreg due to overheating; second, it cannot adapt to dynamic working conditions. When the workshop faces tight energy consumption quotas, the fixed weights cannot prioritize energy consumption control, which may lead to the situation of exceeding the energy consumption limit and limiting production, affecting order delivery. Similarly, for impregnation, the problem is solved by increasing the heating power of the hot plate and rolling roller, but overheating will reduce the activity of prepreg. Therefore, we propose a carbon fiber prepreg production control system based on dynamic coordination of process parameters. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a carbon fiber prepreg production control system based on dynamic coordination of process parameters. Through multi-module collaboration and innovative technologies, it solves the aforementioned defects of existing systems and achieves precise control of process parameters and coordinated optimization of quality and energy consumption.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A carbon fiber prepreg production control system based on dynamic coordination of process parameters. This control system includes: The data acquisition module has the function of full life cycle adaptive calibration and sensor health assessment linkage. It is used to acquire real-time process parameters in the production process of carbon fiber prepreg and dynamically correct the composite deviation of parameter measurement caused by long-term equipment aging and workshop environmental factors, including temperature and humidity changes. The real-time process parameters specifically include: resin flow rate, resin viscosity, and resin temperature of the resin supply system; temperature of the impregnation area of ​​the impregnation device, impregnation duration, and resin flow rate in the impregnation tank; pressure applied by the rolling roller of the rolling device, rotation speed of the rolling roller, and stress distribution in the rolling area; rotation speed of the traction roller of the traction device, fiber tension during traction, and temperature and heating duration of the hot plate and hot roller. The dynamic collaborative analysis module is connected to the data acquisition module via a data transmission link and incorporates a multiphysics and digital twin fusion analysis model. This model uses digital twin technology to construct a virtual production environment that is completely consistent with the actual production scenario, integrating the interrelationships of four physicochemical processes: temperature field, flow field, stress field, and chemical field. The temperature field refers to the temperature distribution in each equipment area, the flow field refers to the resin flow state, and the stress field refers to the force distribution generated by rolling and traction. The model can achieve early prediction of process parameter deviations and precise location of the root causes of deviations, rather than only performing post-event analysis after deviations occur. The control instruction generation module is connected to the dynamic collaborative analysis module and has reinforcement learning dynamic weight optimization logic. This logic can automatically adjust the optimization weights of the three objectives of quality compliance, energy consumption control, and equipment protection according to real-time production needs, and generate dynamic control instructions that simultaneously meet the requirements of multiple objectives. The real-time production needs include the prepreg quality grade required by different orders, the energy consumption limit set by the workshop, and the current operating load of the production equipment. The execution module, connected to the control command generation module, has the function of multi-device collaborative response and delay compensation. This function can synchronously drive related production equipment to perform adjustment operations, and compensate for the difference in response speed of different equipment after receiving the command. For example, it can compensate for the difference between the time required for the heating device to reach the target temperature and the time required for the traction machine to reach the target speed, so as to avoid the fluctuation of prepreg quality caused by asynchronous equipment adjustment. The quality monitoring and feedback module is connected to the execution module and the dynamic collaborative analysis module, respectively, and has the function of dynamic mapping of multi-dimensional raw material characteristics, production parameters and quality indicators. This function can not only adjust the quality inspection standards according to the batch characteristics of the resin, but also integrate the influence of carbon fiber characteristics on the quality standards. Among them, carbon fiber characteristics include carbon fiber monofilament diameter and carbon fiber surface treatment method, ultimately forming a cyclical control of parameter adjustment, quality inspection and feedback optimization.

[0006] Preferably, the lifecycle adaptive calibration and sensor health assessment linkage function of the data acquisition module is implemented through the following sub-units: A multi-dimensional calibration benchmark database subunit is used to store dynamic calibration benchmark data throughout the entire life cycle of production equipment. This benchmark data specifically includes: sensor calibration parameters under different ambient temperature conditions at the time of equipment delivery, covering an ambient temperature range of -10℃ to 40℃; optimal sensor operating parameters derived from the prepreg quality pass data of each production batch after completion; and emergency sensor calibration parameters under extreme production conditions, including sudden and significant changes in resin viscosity and sudden and rapid increases in the temperature of hot plates and hot rollers. This database differs from the static, single benchmark data used in existing systems and can adapt to calibration needs under different operating conditions. The deviation and health correlation analysis subunit is used to calculate in real time the deviation ratio between the process parameters collected by the current sensor and the baseline data under the same operating conditions in the benchmark database. At the same time, it monitors the operating status parameters of the sensor, including the parameter drift rate of the temperature sensor and the signal response time of the pressure sensor. By establishing a correspondence model between the parameter deviation ratio and the sensor health, the sensor status can be evaluated in real time. For example, when the parameter drift rate of the temperature sensor exceeds 0.5°C per month, the sensor health will drop to 70%, and a health warning will be triggered. The predictive calibration execution subunit automatically generates a sensor calibration command when the sensor health level falls below a set threshold or the parameter deviation ratio exceeds 3% for three consecutive minutes. The set threshold is 80%. Environmental factor compensation is incorporated during the calibration process. For example, when the workshop humidity is higher than 60%, the measurement calibration coefficient of the flow sensor is adjusted. After calibration, the sensor's parameter measurement accuracy is improved to within ±1%, and a prediction result of the sensor's remaining service life is output, with the prediction error controlled within 5%. This subunit overcomes the limitation of existing technologies that only calibrate parameter deviations and cannot predict sensor failures. The calibration and production collaboration subunit is used to automatically call the digital twin model in the dynamic collaborative analysis module when sensor calibration requires a brief interruption of the sensor's data acquisition. This model generates predicted values ​​for the corresponding process parameters during the interruption period, ensuring that process parameter data acquisition is not interrupted during production and avoiding data loss problems caused by sensor calibration in existing systems.

[0007] Preferably, the multiphysics and digital twin fusion analysis model of the dynamic collaborative analysis module meets the following requirements: This model is based on four mathematical expressions that couple four physical and chemical interactions. These four mathematical expressions are: one describing the relationship between resin viscosity and temperature / time; another describing the relationship between fluid flow state and pressure, velocity, and viscosity; a third describing the relationship between heat transfer and temperature distribution; and a fourth describing the relationship between the stress state and deformation of a solid. Through digital twin technology, the physical structure of the production equipment, real-time collected process parameters, and workshop environmental parameters are mapped to a virtual production environment that corresponds 1:1 to the actual production scenario. The physical structure of the production equipment includes the specific dimensions of the hot plate and hot roller cavity, and the elastic characteristic parameters of the rolling roller. The workshop environmental parameters include workshop temperature and humidity. The model can achieve a visual simulation of the process parameter changes. The model has dual functions of deviation prediction and root cause tracing: it simulates the changing trends of various process parameters in the next 5 minutes through a virtual production environment, such as predicting that the resin viscosity will decrease by 2% due to the increase in temperature; when parameter deviations occur in actual production, the parameter inversion function of the digital twin model is used to trace the root cause of the deviation caused by the coupling of four physical and chemical effects. For example, when the prepreg thickness exceeds the standard, it is traced back to the combined effects of insufficient pressure applied by the rolling roller (or excessive spacing), high resin viscosity, and excessive rotation speed of the traction roller, rather than a single factor. This model includes a self-optimizing subunit that incorporates federated learning technology. This subunit can optimize the model by integrating anonymized production data from multiple factories in the same industry without sharing the original production data from each factory. For every 200 sets of cross-factory production data included, the model's prediction error for process parameters is reduced by 0.5% to 1%. This subunit addresses the problem that existing models rely solely on data from a single factory and have poor adaptability to different production scenarios, thus improving the consistency between the model's calculation results and actual production conditions. The model uses a four-dimensional interaction coefficient matrix method to calculate the correlation between different physicochemical effects, rather than a simple weighted calculation. The interaction coefficient between the flow field and the stress field is set to 1.15, which means that uneven resin flow will lead to an imbalance in the stress distribution in the rolling area. The specific calculation logic is described in words as follows: the correlation between a certain process parameter and the target quality index is equal to the influence weight of the physicochemical field corresponding to the parameter, the deviation ratio of the parameter from the benchmark value, and the product of the interaction coefficients between the four physicochemical fields. This calculation method is more in line with the complex dynamic interaction of multiple fields in actual production.

[0008] Preferably, the reinforcement learning dynamic weight optimization logic of the control instruction generation module is implemented through the following sub-units: The reinforcement learning training subunit uses maximizing the prepreg quality compliance rate, minimizing energy consumption in the production process, and minimizing equipment wear as the reward calculation criteria in three dimensions. It trains the intelligent computing unit using historical production data from over 1000 batches. This allows the intelligent computing unit to automatically adjust the optimization weights of the three dimensions based on real-time production conditions. For example, when receiving an order for aerospace-grade high-precision prepreg, the weight of the quality compliance target is increased to 0.8, while the weight of the energy consumption control target is reduced to 0.15. When workshop energy consumption has reached 90% of the set quota, the weight of the energy consumption control target is increased to 0.4, and the weight of the quality compliance target is adjusted to 0.55. This subunit differs from existing systems that use fixed weights, allowing for flexible adaptation to different operating conditions. The multi-scheme simulation and evaluation subunit is used to simulate the effects of each set of instructions after execution through a digital twin model in the dynamic collaborative analysis module, based on 3 to 5 sets of preliminary control instructions generated under dynamic weights. For example, it simulates the changes in the prepreg penetration degree, the increase in production energy consumption, and the change in the operating load of production equipment after adjusting the heating device power by 10%. At the same time, it calculates the comprehensive benefits of each scheme. The specific calculation logic is described in words as follows: the comprehensive benefit equals the benefit brought by the prepreg quality meeting the standard, minus the cost caused by the increase in production energy consumption, and then minus the cost caused by the increase in equipment wear and tear. The dynamic command optimization subunit is used to select the initial control command scheme with the highest overall benefit and adjust the time difference of command transmission according to the response characteristics of different devices. For example, it takes 5 seconds for the heating device to reach the target power after receiving the command and 3 seconds for the traction machine to reach the target speed after receiving the command. In this case, the command is sent to the heating device 2 seconds in advance so that the heating device and the traction machine can reach the target parameters synchronously. This subunit avoids the equipment adjustment delay problem caused by synchronous command transmission in the existing system. The extreme condition emergency subunit is used to automatically increase the weight of prepreg quality assurance to 0.95 when a sudden failure occurs during the production process, and generate emergency control instructions. Sudden failures include resin supply interruption due to resin supply blockage and sensor damage that prevents parameter acquisition. Emergency control instructions include pausing the traction machine operation and maintaining the current temperature of the hot plate and hot roller; at the same time, it calls the backup parameter acquisition channel, such as using infrared temperature measurement to replace the damaged temperature sensor. The fault response time of this subunit is controlled within 1 second, which can minimize the losses caused by the failure.

[0009] Preferably, the function of dynamically mapping multi-dimensional raw material characteristics, production parameters, and quality indicators in the quality monitoring and feedback module is implemented through the following sub-units: The multi-raw material characteristic acquisition subunit is used to acquire the characteristic parameters of resin and carbon fiber. At the same time, the wettability of carbon fiber and resin in carbon fiber is measured in real time through the built-in carbon fiber and resin permeability measurement device. The characteristic parameters of resin include the weight distribution range of resin molecules and the activity of resin wetting reaction. The characteristic parameters of carbon fiber include carbon fiber monofilament diameter, carbon fiber surface roughness, and sizing agent content on carbon fiber surface. The measurement deviation of carbon fiber monofilament diameter is controlled within ±1 micrometer, the measurement deviation of sizing agent content on carbon fiber surface is controlled within ±0.1%, and the measurement error of permeability is controlled within 0.5 MPa. The quality standard dynamic generation sub-unit is used to establish the correspondence rules between carbon fiber characteristics, resin characteristics and quality inspection standards. For example, when the diameter of a carbon fiber monofilament increases by 5 micrometers, the uniformity error standard of the resin content in the prepreg is relaxed by 0.2%, because the increased gap between carbon fibers requires more resin to fill, and overly strict standards will lead to production difficulties. When the surface roughness of the carbon fiber (expressed as Ra value) decreases by 0.2 micrometers, the temperature standard of the impregnation area is increased by 2°C. The quality and parameter traceability subunit is used to trace not only changes in process parameters during production when there are deviations in prepreg quality, but also changes in raw material characteristics. For example, if the sizing agent content on the surface of a batch of carbon fiber increases by 0.3%, it will cause resin to adhere to the carbon fiber. At the same time, it generates raw material adjustment suggestions, such as reducing the viscosity of the resin by 0.5 Pa·s to improve the resin adhesion. This subunit fills the gap in the existing technology that only traces production parameters and ignores the impact of changes in raw material characteristics on quality. The multi-batch data mining subunit is used to analyze more than 100 historical batches of raw material characteristic data, production parameter data, and quality inspection data to uncover implicit correlations between the data, such as the optimal matching relationship between the surface roughness of carbon fiber and the pressure applied by the rolling roller; and to update the corresponding rules of carbon fiber characteristics, resin characteristics, and quality inspection standards based on the mining results, so as to improve the accuracy of quality standard adaptation.

[0010] Preferably, the deviation prediction function of the multiphysics and digital twin fusion analysis model is achieved through the following steps: Digital twin scenario initialization steps: Import the 3D structural model of the production equipment into the model, with the model size accuracy controlled within 0.1 mm; input the current workshop environment parameters and the characteristic parameters of the raw materials used in the current production, including workshop temperature and humidity, and raw material characteristic parameters including resin viscosity and carbon fiber monofilament diameter; finally, establish a virtual production environment that is completely consistent with the actual production scenario. Real-time data synchronization steps: The model receives real-time process parameters transmitted by the data acquisition module every 0.1 seconds, and updates the corresponding parameter values ​​in the virtual production environment according to the received parameters, such as updating the temperature values ​​of each area of ​​the hot plate and hot roller, and the pressure distribution status of the rolling roller. The four-dimensional coupled simulation steps involve calculating the effects of four physicochemical processes on the process parameters using a model. Specifically, these include: the effect of the temperature field on resin viscosity (decreases by 0.8 Pa·s for every 1°C increase in temperature); the effect of the flow field on resin impregnation velocity (shortening the impregnation duration by 0.5 minutes for every 0.1 m / s increase in resin flow velocity); the effect of the stress field on carbon fiber alignment (improving the orientation uniformity of carbon fibers by 2% for every 1 MPa increase in pressure applied by the rolling roller); and the effect of the chemical field on prepreg penetration (improving the penetration degree of prepreg by 1.2% for every 1 minute extension of the heating duration). The simulation also simulates the changing trends of each process parameter over the next 5 minutes. Deviation warning judgment steps: If the simulation results show that a certain process parameter will exceed the set benchmark range, such as predicting that the impregnation degree of the prepreg will be less than 95% after 3 minutes, the model immediately outputs deviation warning information and clearly marks the four physicochemical coupling root causes of the deviation, such as the temperature of the hot plate and the left side of the hot roller being 2°C lower than the benchmark value, the resin viscosity being too high, and the traction roller rotation speed being 0.2 m / min faster than the benchmark value; at the same time, the control command generation module is triggered in advance to generate adjustment commands, avoiding the prepreg scrap rate caused by adjusting only after the deviation occurs in the existing technology. The scrap rate of the existing system is usually above 2%, while the optimized system can control it to within 0.5%.

[0011] Preferably, the training and application process of the reinforcement learning dynamic weight optimization logic includes the following steps: Training dataset construction steps: Collect over 1000 batches of production data, covering process parameter data, quality inspection data, energy consumption statistics, and equipment load data under different production conditions. These different production conditions include production using high-viscosity resin, production using fine-diameter carbon fiber, and production with high energy consumption quotas set in the workshop. For each data set, label it with specific quality compliance rate, actual energy consumption value, and actual equipment load rate; for example, a quality compliance rate of 98%, actual energy consumption of 50 kWh / ton of prepreg, and actual equipment load rate of 85%. The training steps for the reinforcement learning agent are as follows: The reward calculation method is: Comprehensive reward = Prepreg quality compliance rate × Quality target weight + (1 - Actual energy consumption rate / Rated energy consumption rate) × Energy consumption target weight + (1 - Actual equipment load rate / Rated equipment load rate) × Equipment protection target weight; The initial weights of the quality target, energy consumption target, and equipment protection target are set to 0.5, 0.3, and 0.2, respectively; Iterative training is performed through the intelligent computing unit, with more than 10,000 iterations, so that the intelligent computing unit learns the optimal weight combination under different production conditions. For example, when producing high-precision prepreg orders, the weight of the quality target is adjusted to 0.7, the weight of the energy consumption target is adjusted to 0.2, and the weight of the equipment protection target is adjusted to 0.1. Real-time weight adjustment steps: During the production process, the system acquires current production demand information, energy consumption monitoring data, and equipment load data in real time. For example, the order requires aerospace-grade prepreg, with quality having the highest priority. Energy consumption monitoring data shows that current production energy consumption has reached 90% of the set quota. Equipment load data shows that the current operating load of the rolling mill has reached 88%. The intelligent computing unit automatically adjusts the weights of the three objectives based on the acquired information. For example, the weight of the quality objective is set to 0.7, the weight of the energy consumption objective is set to 0.25, and the weight of the equipment protection objective is set to 0.05. Based on the adjusted weights, control instructions are generated. Weight optimization feedback steps: After each production batch is completed, the actual quality inspection results, actual energy consumption statistics, and actual equipment operating status data of that batch are fed back to the intelligent computing unit to update the parameters of the training model; optimization is achieved through continuous feedback.

[0012] Preferably, the environmental compensation function of the predictive calibration execution subunit is specifically implemented in the following ways: The environmental parameter acquisition subunit is used to collect key parameters of the workshop environment in real time, including workshop ambient temperature, workshop ambient humidity, and workshop ambient air pressure. The measurement accuracy of workshop ambient temperature is controlled within ±0.5℃, the measurement accuracy of workshop ambient humidity is controlled within ±2% relative humidity, and the measurement accuracy of workshop ambient air pressure is controlled within ±1000 Pa. The environment and calibration coefficient mapping database subunit is used to store sensor calibration coefficients under different environmental parameter conditions. For example, when the relative humidity of the workshop environment increases by 10%, the measurement calibration coefficient of the flow sensor is reduced by 0.03 because the increase in humidity will affect the density of the resin fluid, thus affecting the flow measurement results. When the air pressure of the workshop environment decreases by 5000 Pa, the measurement calibration coefficient of the pressure sensor is increased by 0.02 because the decrease in air pressure will change the reference value of pressure measurement, which needs to be compensated by coefficient adjustment. Dynamic calibration calculation steps: When generating sensor calibration instructions, the corresponding calibration coefficient is retrieved from the mapping database based on the current environmental parameters obtained by the environmental parameter acquisition subunit; and the calibration target value of the sensor is corrected using this coefficient. For example, if the standard resin flow rate is 10 liters / minute, and the workshop ambient humidity is 60% relative humidity, the calibration target value is adjusted to 9.91 liters / minute. Calibration effect verification steps: After calibration, collect real-time measurement data from 5 sets of sensors and calculate the deviation ratio between each set of data and the corrected reference value; if the deviation ratio is ≤1%, the calibration is deemed qualified; if the deviation ratio is >1%, the environmental compensation coefficient is readjusted and the calibration operation is performed again; this function ensures the calibration accuracy of sensors under different environmental conditions and solves the calibration deviation problem caused by ignoring environmental influences in existing technologies. When environmental conditions change by 50%, the calibration error of existing systems is usually above 3%, while this optimized system can control it to within 1%.

[0013] Preferably, the integration and application of the carbon fiber and resin impregnation measurement device in carbon fiber includes the following: Measurement device integration method: An ultrasonic measurement device is installed at the material outlet of the impregnation device; Steps for constructing a model relating resin impregnation and mass in carbon fiber: For example, when the impregnation of carbon fiber and resin increases by 1 MPa, the tensile strength of the prepreg increases by 2 MPa. Quality standard adjustment steps: If the impregnation measurement value of a certain batch of carbon fiber and resin is lower than the set benchmark value, such as 25 MPa, the quality monitoring feedback module will automatically lower the tensile strength standard of the prepreg batch by 1 MPa to avoid the production process being difficult to control due to excessive pursuit of tensile strength; at the same time, the dynamic collaborative analysis module will be triggered to adjust the relevant process parameters, such as increasing the temperature of the impregnation area of ​​the impregnation device by 2°C to enhance the interfacial bonding effect between carbon fiber and resin. Data traceability steps: The carbon fiber and resin impregnation data, adjusted quality inspection standards, and corresponding process parameter adjustment records for each batch of production are uniformly included in the batch production file; when prepreg quality problems occur later, it is possible to quickly trace whether it is due to insufficient impregnation.

[0014] Preferably, the multi-device collaborative response and latency compensation function of the execution module is specifically implemented in the following ways: The equipment response delay database subunit is used to store the response characteristic data of each production equipment; for example, it takes 5 seconds for the heating device to reach the target temperature after receiving the temperature adjustment command, 3 seconds for the traction machine to reach the target speed after receiving the speed adjustment command, and 4 seconds for the hydraulic system of the rolling mill to reach the target pressure after receiving the pressure adjustment command. The delay compensation calculation subunit is used to determine the time difference of sending commands to each device based on the difference in response delay of related devices; for example, if the response time of the heating device is 2 seconds longer than that of the traction machine, the adjustment command is sent to the heating device 2 seconds in advance so that the heating device and the traction machine can reach the target parameters synchronously. The collaborative execution monitoring subunit is used to monitor the progress of each device in executing adjustment commands in real time. For example, the actual power of the heating device has increased to 80% of the target power, and the actual speed of the traction machine has decreased to 90% of the target speed. If the execution progress of a certain device exceeds the expected delay, for example, if the heating device has not reached the target power within 10 seconds, the execution speed of other related devices is immediately adjusted, for example, the adjustment rate of the traction machine speed is reduced, and the heating device is allowed to complete the adjustment. The emergency coordination subunit is used to immediately send emergency adjustment instructions to all related equipment when a certain production equipment suddenly fails. For example, if a rolling mill stops operating due to mechanical failure, the emergency adjustment instructions include stopping the resin delivery of the resin supply system and reducing the speed of the traction machine to 0. This subunit can avoid material waste in the event of a failure, reducing the amount of material waste during the failure period to 0.1 kg / min, while existing systems are usually above 0.5 kg / min.

[0015] (III) Beneficial Effects 1. By monitoring the health status of sensors in real time, such as the drift rate of temperature sensors, when the health level falls below a threshold or parameter deviations exceed the standard, a calibration command is automatically generated based on workshop temperature and humidity. The entire calibration process requires no downtime and the time consumption is controlled within a certain period. This solution not only avoids the economic losses caused by existing downtime calibration but also solves the dynamic deviation problem caused by environmental changes. Secondly, by leveraging a multiphysics and digital twin fusion model, the coupling relationships of temperature field, flow field, stress field, and chemical field are integrated, and the parameter change trend over a certain period of time is simulated through a 1:1 virtual scene. When parameters such as permeability are predicted to be below standard, the root causes of multi-field coupling, such as uneven temperature field and excessive flow velocity, can be clearly identified. Moreover, through reinforcement learning dynamic weight optimization logic, compared with the existing single-objective fixed weight control, it achieves flexible adaptation of multiple objectives. The weights of quality, energy consumption, and equipment protection are adjusted according to real-time production needs. In aerospace-grade orders, the weight of quality is increased to ensure permeability, in civilian-grade orders, the weight of energy consumption is increased to control costs, and when energy consumption quotas are tight, the equipment load is reduced first. This not only avoids the energy waste caused by the existing over-quality, but also solves the problem of dynamic operating condition adaptation.

[0016] 2. Leveraging a multi-device collaborative response and delay compensation scheme, the system stores the response characteristics of each device and compensates for delay differences by sending instructions in advance, thereby improving the synchronization rate of the devices and reducing the proportion of quality fluctuations caused by asynchronous adjustments. Simultaneously, in the event of a sudden equipment failure, the system can immediately drive related devices to coordinate emergency responses; for example, stopping resin supply and reducing traction speed when the rolling mill fails, thus reducing raw material waste. Furthermore, through a multi-dimensional dynamic mapping scheme of raw material characteristics, production parameters, and quality indicators, the system adjusts standards based on resin batch characteristics such as molecular weight distribution and flowability, and optimizes parameters based on carbon fiber characteristics such as monofilament diameter and surface roughness. For example, increasing the carbon fiber diameter relaxes the resin content uniformity standard, while decreasing the surface roughness increases the temperature. This solves the problem of poor adaptability of existing quality standards and provides more complete traceability of quality deviations. Attached Figure Description

[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0018] Figure 1 This is an architecture diagram of an embodiment of the present invention. Detailed Implementation

[0019] This application provides a carbon fiber prepreg production control system based on dynamic coordination of process parameters, solving the economic losses caused by downtime calibration in existing technologies. It also addresses the dynamic deviation problem caused by environmental changes. By monitoring the health status of sensors in real time, such as the drift rate of temperature sensors, when the health status falls below a threshold or parameter deviation exceeds the standard, it automatically generates calibration instructions based on workshop temperature and humidity. The entire calibration process requires no downtime and is controlled within a certain timeframe. Secondly, by leveraging a multi-physics and digital twin fusion model, it integrates the coupling relationships of temperature field, flow field, stress field, and chemical field, using a 1:1 virtual scene model. By predicting parameter trends over a certain period, when parameters such as permeability are predicted to fall below the standard, the root causes of multi-field coupling, such as uneven temperature field and excessive flow velocity, can be clearly identified. Moreover, through reinforcement learning dynamic weight optimization logic, compared with the existing single-objective fixed weight control, it achieves flexible adaptation to multiple objectives. The weights of quality, energy consumption, and equipment protection are adjusted according to real-time production needs. For aerospace-grade orders, the weight of quality is increased to ensure permeability, while for civilian-grade orders, the weight of energy consumption is increased to control costs. When energy consumption quotas are tight, equipment load is reduced first. This not only avoids the energy waste caused by the existing over-quality, but also solves the problem of dynamic operating condition adaptation.

[0020] Example: Figure 1As shown, the technical solution in this application embodiment addresses the economic losses caused by downtime calibration in the prior art, and also solves the dynamic deviation problem caused by environmental changes. The overall approach is as follows: Carbon fiber prepreg is a key material in aerospace and high-end equipment fields. Its production requires resin impregnation and rolling shaping, involving four types of physical and chemical processes: temperature, flow, stress, and chemical reaction, referred to as the four fields. These four processes affect each other. For example, an increase in temperature will reduce the viscosity of the resin, thereby accelerating the resin flow rate. If only a single process is controlled, it is easy to cause quality defects such as uneven resin content and insufficient impregnation in the prepreg.

[0021] Existing control systems cannot solve problems such as missing four-field coupling analysis, insufficient dynamic deviation calibration, poor fixed weight adaptation, neglect of raw material coordination, and asynchronous equipment response: Traditional systems only monitor temperature and cannot identify coupling deviations caused by excessively fast traction speeds leading to insufficient impregnation time; quarterly shutdowns are required to calibrate sensors, resulting in a large amount of waste due to drift; a fixed mass weight of 0.8 and the production of civilian-grade prepregs according to aerospace-grade standards lead to excessive energy consumption; only resin characteristics are considered, ignoring the impact of carbon fiber diameter changes on resin demand; independent equipment adjustments and asynchronous responses between heating devices and traction machines result in thickness deviations.

[0022] To address the problems existing in the prior art, this invention provides a carbon fiber prepreg production control system based on dynamic coordination of process parameters. This management system uses precise data acquisition, four-field collaborative analysis, dynamic command generation, synchronous execution, and quality feedback as its logical chain, with five modules achieving data interaction via industrial Ethernet. Details are as follows: 1. Data Acquisition Module: The data acquisition module enables integrated adaptive calibration and sensor health assessment throughout the entire lifecycle. This module primarily dynamically corrects combined measurement deviations caused by equipment aging and environmental changes, strictly adhering to a four-step process: precise sensor deployment, real-time health monitoring, dynamic calibration, and production collaboration. (1) Sensors shall be deployed according to the five major nodes of the entire production process, and the specific requirements are as follows: An electromagnetic flow meter is installed at the outlet of the resin conveying equipment to collect the resin flow rate; a rotational viscometer is installed downstream of the flow meter to collect the resin viscosity; and a platinum resistance temperature sensor is installed next to the viscometer to collect the resin temperature. Multiple platinum resistance temperature sensors are evenly arranged around the inner wall of the impregnation tank, with spacing controlled at approximately 50 cm to ensure regional temperature uniformity ≤ ±1℃. A laser velocimeter is installed at the outlet of the impregnation tank to collect resin flow velocity. An encoder is installed at the end of the drive roller shaft of the impregnation tank conveyor belt to calculate the conveyor belt speed based on the roller diameter, thereby accumulating the impregnation duration. Pressure sensors are installed at the bearing seats at both ends of the upper and lower rollers to collect the pressure applied by the rollers. A speed sensor is installed at the end of the roller drive motor shaft to collect the roller rotation speed. Distributed pressure plates are attached to the surface of the rollers every 10 cm along the width direction to collect the stress distribution in the rolling area. A speed sensor is installed at the end of the traction roller drive motor shaft to collect the traction speed. Tension sensors are installed at both ends of the traction tension roller to collect fiber tension. Three platinum resistance temperature sensors are evenly arranged on the hot plate and hot roller. Encoders are installed at the ends of the hot plate and hot roller shafts to accumulate the heating duration.

[0023] (2) Real-time sensor health assessment method: Establish a mapping model between operating parameters and health status, and determine the health status by real-time monitoring of key sensor performance indicators. The specific implementation is as follows: Temperature sensors monitor monthly drift rate, statistically analyzing the deviation between sensor measurements and a standard constant temperature bath each month. A monthly drift ≤0.5℃ is considered healthy; for every 0.1℃ exceeding this drift, the health level decreases by 10%. Pressure sensors monitor response time; by applying a step pressure through a standard pressure source, a response time ≤0.3 seconds is considered healthy; for every 0.05 seconds exceeding this, the health level decreases by 10%. Flow sensors monitor repeatability error; by continuously collecting measurements at the same stable flow rate 10 times, a repeatability error ≤0.5% is considered healthy; for every 0.1% exceeding this, the health level decreases by 10%. The deviation ratio calculation calculates the deviation ratio between the current collected value and the benchmark value under the same working condition in real time, that is, deviation ratio = (current value - benchmark value) / benchmark value × 100%. If the deviation ratio exceeds 3% for 3 consecutive minutes or the health level is lower than 80%, the dynamic calibration process is triggered immediately.

[0024] (3) Dynamic calibration execution process: The calibration process needs to incorporate environmental compensation to ensure calibration accuracy under different environmental conditions. The specific steps are as follows: Real-time temperature, humidity, and air pressure are obtained through the workshop environmental monitoring station; The preset environment and calibration coefficient mapping table can be consulted. Specifically, the calibration coefficient for the flow sensor is 1.00 when the humidity is 40% to 50%, 0.98 when it is 50% to 60%, and 0.97 when it is 60% to 70%; the calibration coefficient for the pressure sensor is 1.00 when the air pressure is 95 to 100 kPa, 1.02 when it is 90 to 95 kPa, and 1.04 when it is 85 to 90 kPa. This can be preset and modified in advance. The calibration target is corrected based on the environmental factor. For example, if the standard resin flow rate is 10 L / min and the humidity is 65%, the corrected calibration target value is 10 × 0.97 = 9.7 L / min. Send calibration commands to the sensor, such as zero-point calibration commands for flow sensors. After calibration, collect three sets of data continuously and calculate the deviation ratio from the corrected target value. If the deviation is ≤1%, the calibration is deemed qualified. Otherwise, readjust the environmental coefficient and calibrate again. The calibration time for each calibration should be controlled within 1 minute.

[0025] (4) Production collaboration guarantee: If the calibration requires a brief interruption of the acquisition of a certain parameter, the digital twin model of the dynamic collaborative analysis module will be automatically called to predict the parameter value of the missing period based on the current working condition parameters such as resin temperature and traction speed. At the same time, the parameter fluctuation threshold of the associated equipment will be relaxed, such as the traction speed fluctuation threshold will be relaxed from ±0.1m / min to ±0.2m / min to avoid false triggering of quality alarm; The entire lifecycle is achieved through a multi-dimensional benchmark library consisting of factory benchmarks, batch optimal benchmarks, and extreme operating condition benchmarks, covering the entire stage from equipment commissioning to aging; adaptive calibration is achieved through environmental compensation and real-time health triggering, which is different from traditional fixed-cycle calibration; health assessment linkage is reflected in the automatic triggering of calibration when the health status is abnormal, and the calibration process and production work together to ensure data continuity.

[0026] 2. Dynamic Collaborative Analysis Module: The dynamic collaborative analysis module features a multi-physics field and digital twin fusion model. This module is the core for achieving early prediction and root cause location of process parameter deviations. The implementation process follows a four-step workflow: digital twin scenario construction, four-field coupling quantization, deviation prediction and root cause tracing, and model self-optimization, ensuring that the multi-field coupling relationship is calculable and traceable. The specific steps are as follows: (1) Specifications for constructing digital twin scenarios: To construct a virtual scenario that is 1:1 with the actual production line, it is necessary to clarify the modeling accuracy and parameter input requirements, as follows: Three-dimensional models of the hot plate, hot roller, rolling mill, and traction machine were created using SolidWorks software; after importing them into the Unity3D engine, physical properties were added to each piece of equipment. Input the workshop environmental parameters and raw material characteristic parameters. The environmental parameters are temperature, humidity, and air pressure. The raw material characteristic parameters are resin molecular weight distribution, flowability, and viscosity at 25°C. The carbon fiber diameter, surface roughness, and sizing agent content are also input. A real-time data connection between the virtual scene and the actual sensors is established through the OPCUA protocol. The data update frequency is set to 0.1 seconds / time to ensure that the virtual scene is synchronized with the actual production.

[0027] (2) The four-field coupling quantitative calculation method transforms the coupling relationship between the temperature field, flow field, stress field, and chemical field into quantitative calculation logic. The specific formulas and parameter values ​​are as follows: The temperature field is coupled with the chemical field, and the relationship between permeability and temperature and time is described based on the Kamal equation: permeability = 90 + 1.2 × (actual temperature - 200) / 5 + 1.2 × (actual time - 10) / 1, where 200℃ and 10 minutes are the baseline conditions. For every 5℃ increase in temperature or every 1 minute increase in time, the permeability decreases by 1.2%. The temperature field is coupled with the flow field. Based on the principles of fluid mechanics, the resin viscosity is calculated as 200 - 0.8 × (actual temperature - 25°C), where 25°C is the baseline. For every 1°C increase in temperature, the viscosity decreases by 0.8 Pa·s. The resin flow velocity is calculated as 0.5 - 0.1 × (actual viscosity - 200) / 0.8. For every 0.8 Pa·s increase in viscosity, the flow velocity decreases by 0.1 m / s. The flow field and stress field are coupled. For every 0.1 m / s decrease in resin flow velocity, the impregnation time is extended by 0.5 minutes, which in turn leads to an increase in resin accumulation in the compaction area. The compaction pressure needs to be increased by 0.05 MPa to ensure that the thickness meets the standard. The stress field is coupled with the product shape. Based on Hooke's law, the fiber orientation degree = 95 + 0.2 × (actual rolling pressure - 1.5), where 1.5 MPa is the benchmark. For every 0.1 MPa increase in pressure, the fiber orientation degree increases by 0.2%, which in turn affects the tensile strength of the prepreg.

[0028] (3) Deviation prediction and root cause tracing process: Based on four-field coupled quantization logic, the parameter change trend is predicted in advance and the root cause of the deviation is located. The specific steps are as follows: The four-field coupling calculation model is used to simulate the changes of various process parameters in the next 5 minutes. For example, if the current hot plate and hot roller temperature is 202℃ (3℃ lower than the baseline) and the traction speed is 5.2m / min (0.2m / min higher than the baseline), then after 3 minutes, the resin viscosity = 200 + 0.8 × 3 = 202.4 Pa·s, the flow rate = 0.5 - 0.1 × (202.4 - 200) / 0.8 = 0.47 m / s, and the permeability = 90 + 1.2 × (202 - 200) / 5 + 1.2 × (10 - 0.2 × 3) / 1 = 94.2%, which is lower than the baseline of 95%. By tracing the source of the deviation through parameter inversion, we deduced from the result of low permeability that low permeability is caused by short impregnation time, short impregnation time is caused by high flow rate, high flow rate is caused by high viscosity, high viscosity is caused by low temperature, and high traction speed further shortens the impregnation time. Finally, we found that the deviation was caused by the combination of uneven temperature field and excessive flow field velocity. If the predicted parameters exceed the baseline range, an early warning message is immediately sent to the control instruction generation module, clearly indicating the parameters and magnitude that need to be adjusted, such as increasing the heating power of the hot plate and hot roller by 5% and reducing the traction speed by 0.2m / min.

[0029] (4) Model self-optimization mechanism: Federated learning technology is introduced to improve the model's generalization ability. The specific implementation is as follows: A data alliance was established with three factories in the same industry. Using a federated learning framework such as FedAvg, anonymized four-field data from each factory were integrated without sharing the original data. For example, temperature field distribution data from factory A, flow field data from factory B, and stress field data from factory C. Every 200 sets of cross-factory data were included, the coupling coefficients of the four fields were refitted. For example, the coefficients of temperature field and chemical field were optimized from 1.2 to 1.22, and the coefficients of flow field and stress field were optimized from 1.15 to 1.17, which improved the model fit from 0.92 to 0.96 and reduced the prediction error. For each optimization of the coefficients, 10 batches of new production data are selected to verify the model accuracy. If the prediction error is ≤2%, the optimization is confirmed to be effective; otherwise, it is reverted to the previous version of the coefficients.

[0030] Multiphysics is achieved through the quantitative coupling of temperature field, flow field, stress field, and chemical field; digital twin fusion is achieved through 1:1 scene construction and real-time data synchronization; deviation prediction and root cause localization are achieved through coupled simulation and parameter inversion, solving the shortcomings of traditional systems that only perform post-event analysis.

[0031] 3. Control command generation module: The control instruction generation module has reinforcement learning dynamic weight optimization logic. This module dynamically adjusts the weights of quality, energy consumption, and equipment protection according to real-time production needs. (1) Training dataset construction specifications: collect production data covering different working conditions to provide diverse samples for reinforcement learning. Specific requirements are as follows: 1,200 batches of production data were collected, covering 6 typical operating conditions: aerospace-grade prepreg with high viscosity resin, civil-grade prepreg with low viscosity resin, summer power rationing with high energy consumption quota, winter normal power supply with low energy consumption quota, new equipment commissioning with low load, and old equipment with high load. Each batch is labeled with quality compliance rate, actual energy consumption, equipment load rate, order type, energy consumption quota, and equipment aging degree. Among them, the quality compliance rate is ≥96% for aviation grade and ≥93% for civil grade; the equipment load rate is ≤90% for key equipment; and the equipment aging degree is ≤1 year for new equipment and ≥3 years for old equipment. The batches are divided into training set (960 batches) and validation set (240 batches) in an 8:2 ratio. The training set is used for agent learning, and the validation set is used to evaluate the weight optimization effect.

[0032] (2) Reinforcement learning agent training process: Deep Q-Network (DQN) is used to construct the agent, and the network structure and training parameters are defined as follows: The input layer consists of 6 operating condition features: order quality level, energy consumption quota utilization rate, equipment load rate, resin viscosity, carbon fiber diameter, and ambient temperature. The hidden layer consists of 2 layers with 64 neurons in each layer. The output layer consists of 3 weights: quality weight, energy consumption weight, and equipment protection weight. Comprehensive benefit = Quality compliance rate × Quality weight + (1 - Actual energy consumption / Rated energy consumption) × Energy consumption weight + (1 - Equipment load rate / Rated load rate) × Equipment protection weight, where the rated energy consumption is 50 kWh / ton and the rated load rate is 90%; The learning rate was set to 0.001, the experience replay pool capacity was set to 10,000, the batch size was set to 32, and the number of iterations was set to 15,000; the initial weights were set to quality 0.5, energy consumption 0.3, and device protection 0.2. Every 1000 iterations, the validation set is used to evaluate the agent weight optimization effect. If the comprehensive benefit of aviation-grade orders is ≥0.65 and that of civil-grade orders is ≥0.60, the training is considered successful; otherwise, the learning rate is adjusted and training is repeated. The final training results are as follows: aviation-grade orders: quality weight 0.7, energy consumption weight 0.2, equipment protection weight 0.1; summer power rationing conditions: energy consumption weight 0.4, quality weight 0.5, equipment protection weight 0.1; old equipment conditions: equipment protection weight 0.3, quality weight 0.5, energy consumption weight 0.2.

[0033] (3) Multi-scheme simulation and optimal selection method: Based on the optimal weight of the current working condition, multiple sets of instructions are generated and filtered. The specific steps are as follows: For the current deviation, such as a 1.5% decrease in penetration, 3 to 5 sets of adjustment instructions are generated based on the optimal weight. For example, Option 1 increases heating power by 10% and decreases traction speed by 5%, Option 2 increases heating power by 8% and decreases traction speed by 6%, and Option 3 increases heating power by 12% and decreases traction speed by 4%. The parameter adjustment values ​​of each scheme were input into the digital twin model to simulate the changes in quality, energy consumption, and equipment within 3 minutes after execution. Scheme 1 achieved the penetration standard but energy consumption increased by 5.2% and equipment load increased by 8.5%; Scheme 2 achieved the penetration standard but energy consumption increased by only 4.1% and equipment load increased by 6.2%; Scheme 3 exceeded the penetration standard but energy consumption increased by 6.3% and equipment load increased by 9.8%. The comprehensive returns of each option are calculated using the reward function: Option 1 = 0.962 × 0.7 + (1 - 5.2%) × 0.2 + (1 - 8.5%) × 0.1 = 0.6734 + 0.1896 + 0.0915 = 0.9545; Option 2 = 0.958 × 0.7 + (1 - 4.1%) × 0.2 + (1 - 6.2%) × 0.1 = 0.6706 + 0.1918 + 0.0938 = 0.9562; Option 3 = 0.965 × 0.7 + (1 - 6.3%) × 0.2 + (1 - 9.8%) × 0.1 = 0.6755 + 0.1874 + 0.0902 = 0.9531. Option 2, which has the highest comprehensive return, is selected.

[0034] (4) Command optimization and emergency handling process; optimize command sending time to ensure equipment synchronization, and clarify the emergency handling mechanism, as follows: The response time of the associated device is retrieved from the device response delay database of the execution module. The heating device needs 5 seconds to reach the target power, and the traction machine needs 3 seconds to reach the target speed. Delay difference = heating device response time - traction machine response time = 5 - 3 = 2 seconds; send an 8% power increase command to the heating device 2 seconds in advance, and send a 6% speed decrease command to the traction machine 3 seconds later to ensure that both reach the target parameters synchronously; When resin supply blockage occurs (flow rate drops by 90%), sensor failure, or roller malfunction, the quality weight is automatically increased to 0.95, an emergency command is generated, the traction machine is stopped, the heating power to the hot plate and hot roller is turned off, and the protective gas supply to the hot plate and hot roller is maintained. If the temperature sensor fails, the infrared thermometer on the top of the hot plate and hot roller is immediately activated. The emergency command response time is controlled within 1 second, and an audible and visual alarm is sent to the central control system. Reinforcement learning is achieved through iterative training of the training dataset and DQN agent; dynamic weights are achieved through weight adjustment adapted to operating conditions; multi-objective instruction generation is achieved through scheme simulation and comprehensive benefit selection, taking into account the functional definitions of quality, energy consumption, and equipment protection, and solving the adaptation defects of traditional fixed weights.

[0035] 4. Execution Module: The execution module features multi-device collaborative response and latency compensation. This module prevents quality fluctuations caused by asynchronous device adjustments, following a four-step process: building a device response characteristic database, calculating latency compensation, monitoring collaborative execution, and coordinating fault emergencies, ensuring synchronized device adjustments. (1) Equipment response characteristic database construction specifications: measure and store the response characteristics of key equipment to provide data support for delay compensation. Specific requirements are as follows: The time taken for each device to reach the target parameter from receiving the command was measured using a high-precision timer. The heating device required 5±0.5 seconds, the traction machine required 3±0.3 seconds, the hydraulic system of the rolling roller required 4±0.4 seconds, and the resin supply pump required 2±0.2 seconds. The response time was repeatedly measured under different equipment loads. The response time of the rolling mill was 4 seconds when the load was 80% and 4.5 seconds when the load was 90%. The response time of the heating device was 4.5 seconds when the load was 80% and 5.5 seconds when the load was 90%. The measured data were stored according to equipment type and load rate to form an equipment response delay database, which was updated once a quarter.

[0036] (2) Delay compensation calculation and command sending process; adjust the command sending time according to the response time difference of the associated devices. The specific steps are as follows: Based on the scheme generated by the control instruction generation module, the equipment that needs to be adjusted in coordination is determined. For example, when adjusting the immersion temperature and traction speed, the associated equipment is the heating device and the traction machine; when adjusting the compaction pressure and resin flow rate, the associated equipment is the compaction machine and the resin supply pump. Using the slowest-responding device as a benchmark, calculate the delay difference of other devices. For example, if the heating device responds in 5 seconds and the traction machine responds in 3 seconds, the delay difference is 5-3=2 seconds; if the rolling mill responds in 4 seconds and the resin supply pump responds in 2 seconds, the delay difference is 4-2=2 seconds. Commands are sent to slow-responding devices with a time difference between advance and delay. For example, the heating device receives the command to increase power by 8% 2 seconds in advance, and the traction machine receives the command to decrease speed by 6% 3 seconds later. The rolling mill receives the command to increase pressure by 0.1MPa 2 seconds in advance, and the resin supply pump receives the command to increase flow rate by 0.2L / min 2 seconds later. Command transmission is achieved through industrial Ethernet to ensure that the command transmission delay is ≤1ms.

[0037] (3) Collaborative execution monitoring and dynamic adjustment method; real-time monitoring of equipment execution progress and dynamic adjustment of execution speed to ensure synchronization, as detailed below: Real-time operating parameters are collected through the equipment controller, the heating device collects the actual power (difference from the target power), the traction machine collects the actual speed (difference from the target speed), and the compactor collects the actual pressure (difference from the target pressure). If a device is lagging behind in its execution progress, such as a heating device only reaching 80% of the target power within 10 seconds when it should normally reach 100% in 5 seconds, it is considered an execution delay. Immediately send adjustment commands to other related equipment to reduce the execution speed, such as reducing the adjustment rate of the traction machine speed, and wait for the heating device to complete the adjustment; after all equipment has reached the target parameters, restore the normal adjustment rate.

[0038] (4) Fault emergency coordination process: In the event of equipment failure, rapid linkage response is achieved to reduce raw material waste. The specific steps are as follows: Equipment malfunctions are identified using vibration sensors, current sensors, and flow sensors. Specifically, the vibration sensor detects whether the vibration value of the roller mill bearing exceeds 0.5g; the current sensor detects whether the motor current exceeds the rated value by 120%; and the flow sensor detects whether the resin flow rate drops suddenly by 90%. Three sets of instructions are generated: the resin supply pump stops running, i.e., the raw material supply is cut off; the traction machine speed drops to 0, i.e., the material conveying stops; the hot plate and hot roller maintain the current temperature to prevent the impregnation of the prepreg already in the furnace from decreasing. If the temperature sensor fails, it automatically switches to the infrared thermometer on the hot plate and hot roller as a backup channel to monitor the furnace temperature in real time; if the pressure sensor fails, the backup pressure sensor next to the rolling mill is activated. Record the fault type, fault location, scope of impact, emergency measures, and handling results in the production log to facilitate subsequent analysis and optimization, such as the bearing failure of the rolling mill. Collaborative response is achieved by synchronously sending instructions from associated devices; delay compensation is achieved by adjusting the response time difference and instruction sending time; fault emergency response is achieved by device linkage and backup channel invocation, solving the asynchronous problem of independent adjustment of traditional devices.

[0039] 5. Quality monitoring and feedback module: The quality monitoring and feedback module employs a dynamic mapping of multi-dimensional raw material characteristics, production parameters, and quality indicators. This module performs cyclical quality optimization and features a four-step process: multi-raw material characteristic data collection, dynamic quality standard setting, quality traceability and adjustment, and data mining and rule updating, ensuring accurate matching between raw material characteristics and production parameters. (1) Specifications for collecting multiple raw material characteristics: Collect key characteristics of resin and carbon fiber, and clarify measurement methods and accuracy requirements, as follows: Molecular weight distribution was measured using gel permeation chromatography (GPC); flowability was measured using differential scanning calorimetry (DSC); viscosity versus temperature curves were measured using a rotational viscometer, for example, 200 to 250 Pa·s at 25°C, with one point collected every 5°C; each batch of resin was sampled three times, and the average value was taken as the characteristic value of that batch. The diameter of a single filament was measured using a laser diameter gauge; the surface roughness was measured using an atomic force microscope (AFM); the surface wetting agent content was measured using the Soxhlet extraction method; five samples were taken from each batch of carbon fiber, and the average value was taken as the characteristic value of that batch. A miniature monofilament pull-out tester is installed at the outlet of the impregnation device. It automatically clamps the carbon fiber monofilament and pulls it out of the resin matrix at a speed of 1 mm / min. The maximum pull-out force is recorded. The impregnation performance is calculated as: maximum pull-out force / (monofilament diameter × resin wrapping length), with a range of 25 to 30 MPa. One sample is collected every 5 minutes, and the measurement time is ≤30 seconds, which does not affect continuous production.

[0040] (2) Dynamic quality standard setting method, which establishes quality standard association rules based on raw material characteristics, as follows: The carbon fiber monofilament diameter is ≤1.5% when it is 7 to 9 μm and ≤1.8% when it is 9 to 12 μm. The increase in diameter leads to an increase in the gap between fibers, which requires more resin to fill. The resin molecular weight distribution is ≤1.5% when it is 2.5 to 2.7 and ≤1.8% when it is 2.7 to 3.0. That is, the wide distribution leads to poor resin flowability. When the resin fluidity is 0.8 to 0.85, the temperature should be 205 to 210℃; when it is 0.85 to 0.9, the temperature should be 200 to 205℃. Low activity requires higher temperature to accelerate wetting. When the carbon fiber surface roughness Ra is 0.8 to 1.0 μm, the temperature should be 200 to 205℃; when it is 0.6 to 0.8 μm, the temperature should be 205 to 210℃. Low roughness requires higher temperature to strengthen interfacial bonding. Tensile strength standards: ≥2.0GPa when impregnation is 28 to 30MPa, ≥1.8GPa when impregnation is 25 to 28MPa, and ≥1.6GPa when impregnation is <25MPa. If the bonding strength is low, the strength standard should be reduced to avoid scrap. For every 50 batches of prepreg produced, the rationality of the standard is evaluated based on the quality data. If more than 5% of the batches fail to meet a certain standard, the standard is adjusted. For example, when the carbon fiber diameter is 12μm, the resin content uniformity error standard is relaxed from 1.8% to 2.0%.

[0041] (3) Quality traceability and parameter adjustment process: When there is a quality deviation, trace the root cause and generate adjustment suggestions. The specific steps are as follows: The quality of prepreg is monitored by online testing equipment, such as resin content analyzers, thickness gauges, and tensile testing machines. If the resin content uniformity error exceeds 1.8%, the impregnation is less than 93%, or the tensile strength is less than 1.8 GPa, quality traceability is triggered. Retrieve production parameters for the period of deviation and compare them with the parameter range of historical qualified batches to identify parameter deviations, such as a traction speed increase of 0.2 m / min; simultaneously compare the raw material characteristics of the current batch with those of historical qualified batches to identify characteristic changes, such as a decrease in carbon fiber impregnating agent content from 0.8% to 0.5%; By combining parameter deviations and characteristic changes, the core root cause can be determined. For example, a decrease in the sizing agent content may lead to insufficient resin adhesion on the carbon fiber surface, which in turn may cause excessive resin content uniformity. Based on the root cause, production parameters and raw material adjustment suggestions are generated. When the wetting agent content decreases by 0.3%, it is recommended to increase the resin viscosity by 0.5 Pa·s and increase the rolling pressure by 0.1 MPa. When the resin fluidity decreases by 0.05, it is recommended to increase the impregnation temperature by 3°C and extend the impregnation time by 1 minute. The adjustment suggestions are sent to the control instruction generation module to optimize subsequent production parameters.

[0042] (4) Data mining and rule update methods: mining implicit associations in historical data and optimizing quality standard mapping rules, as detailed below: Import raw material characteristic data from 100 historical batches, including resin molecular weight distribution, carbon fiber diameter, etc.; as well as production parameter data and quality data; Pearson correlation coefficient analysis revealed that the combination of carbon fiber surface roughness of 0.8 μm and rolling pressure of 1.5 MPa resulted in the best resin content uniformity (correlation coefficient 0.92); the combination of resin molecular weight distribution of 2.8, traction speed of 5 m / min, and impregnation temperature of 205℃ resulted in the highest impregnation (correlation coefficient 0.88). The dynamic quality standard mapping rules are optimized based on the mining results. For example, the rolling pressure standard corresponding to a carbon fiber roughness of 0.8μm is reduced from 1.4 to 1.6MPa to 1.5 to 1.55MPa. The updated rules need to be verified through 10 batches of production. If the quality pass rate increases by ≥2%, the update is confirmed to be effective.

[0043] Multi-dimensional raw material characteristics are achieved through the full-characteristic collection of resin and carbon fiber; dynamic mapping is achieved through the association rules between raw material characteristics and quality standards; quality closed loop is achieved through traceability, adjustment, and rule updates, which meets the functional definition of adapting to raw material characteristics and optimizing quality, and solves the adaptation defects of traditional methods that only focus on resin characteristics.

[0044] 6 sub-units: 6.1 The adaptive calibration and environmental compensation subunit of the data acquisition module is detailed below: A multi-dimensional calibration benchmark database stores three types of benchmark data throughout the entire lifecycle of the equipment. The factory benchmark consists of calibration parameters for the equipment at various temperature points from -10℃ to 40℃, such as a flow sensor outputting 0.02V at zero point at 25℃ and a pressure sensor outputting 4.00V at 1MPa. The batch benchmark is the optimal measurement value of the sensor for each batch after it has passed quality inspection, such as a flow sensor measurement value of 10.2L / min when the resin content uniformity is 1.5%. The extreme operating condition benchmark is the emergency calibration value of the sensor when simulating scenarios such as a 30% change in resin viscosity or a sudden 10℃ increase in the temperature of the hot plate or hot roller, such as a temporary calibration coefficient of 1.05 for the viscosity sensor. The three types of benchmark data are stored in categories according to equipment number and usage time, and can be retrieved according to operating conditions. After collecting the workshop temperature, humidity, and air pressure, a preset environmental and calibration coefficient mapping table is invoked. The flow sensor coefficient is 1.00 when the humidity is 40% to 50%, 0.98 when it is 50% to 60%, and 0.97 when it is 60% to 70%. The pressure sensor coefficient is 1.00 when the air pressure is 95 to 100 kPa and 1.02 when it is 90 to 95 kPa. The calibration target value is corrected by multiplying the standard value by the environmental coefficient. For example, the standard flow rate of 10 L / min is corrected to 9.7 L / min when the humidity is 65%. After calibration, three sets of data are collected for verification. If the deviation is ≤1%, it is considered qualified, which solves the defect of traditional methods that ignore the influence of the environment.

[0045] 6.2 The deviation prediction and root cause tracing sub-unit of the dynamic collaborative analysis module is as follows: A multi-physics coupling model is constructed based on the wetting dynamics equation, fluid dynamics equation, heat conduction equation, and solid mechanics equation. It integrates the interaction relationships of these four fields: a coefficient of 1.2 for the temperature field and chemical field (meaning a 1°C increase in temperature accelerates the resin wetting reaction rate by 20%), and a coefficient of 1.15 for the flow field and stress field (meaning uneven flow velocity leads to a 15% deviation in stress distribution). This forms a four-dimensional interaction coefficient matrix. The correlation degree is calculated using the formula: correlation degree = physical field weight × deviation ratio × interaction coefficient product. For example, the correlation degree between resin temperature and impregnation is 0.42 × 5% × 1.2 × 1.15 = 0.029. The deviation prediction steps are as follows: First, initialize the virtual scene and import the equipment's 3D model and operating parameters; second, synchronize real-time data and update the virtual parameters every 0.1 seconds; third, perform four-field coupling simulation to calculate the impact of temperature, flow rate, and pressure on quality; fourth, issue warnings and identify root causes, such as predicting that when the temperature of the hot plate and hot roller is 3°C lower, the permeability will drop to 94.2% after 3 minutes, and the root cause is identified as uneven temperature field and fast traction speed.

[0046] 6.3 The reinforcement learning training subunit of the control instruction generation module is as follows: A dataset of 1200 batches covering 6 types of working conditions was constructed. The reward function was designed as: comprehensive benefit = quality × quality weight + energy consumption × energy consumption weight + equipment × equipment weight. After 15000 iterations, the agent learned that the quality weight of aviation-grade orders was 0.7 and the energy consumption weight of summer power rationing was 0.4. Based on the equipment response time, such as 5 seconds for heating and 3 seconds for traction, calculate the delay difference of 2 seconds, and send the instruction to the heating device 2 seconds in advance to ensure synchronous compliance; when simulating multiple schemes, use digital twins to quantify quality, energy consumption, and equipment impact, and select the scheme with the highest overall benefit.

[0047] 6.4 The permeability measurement subunit of the quality monitoring feedback module is as follows: A miniature monofilament pull-out tester is integrated at the outlet of the impregnation device. It automatically clamps the carbon fiber monofilament and pulls it out of the resin matrix at a speed of 1 mm / min to calculate the impregnation. The test is conducted once every 5 minutes, and the measurement time is ≤30 seconds, which does not affect production.

[0048] The tensile strength is ≥1.8GPa when the impregnation pressure is 25 to 28MPa, and drops to ≥1.6GPa when it is <25MPa; at the same time, the impregnation temperature standard is adjusted according to the resin flowability and carbon fiber roughness.

[0049] 6.5 The emergency coordination subunit of the execution module is as follows: The equipment response delay database stores response times such as 5 seconds for the heating device and 3 seconds for the traction machine; delay compensation calculates the response difference of related equipment and adjusts the command sending time; collaborative execution monitoring collects equipment progress in real time and adjusts the execution speed when there is a lag; emergency collaboration sends commands to stop resin supply, reduce traction speed, and infiltrate and heat preservation within 1 second in case of failure; infrared thermometer is activated when temperature sensor fails, reducing raw material waste from 0.5 kg / min to 0.1 kg / min. These are the technical features of fault emergency collaboration and reduced raw material waste.

[0050] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A carbon fiber prepreg production management and control system based on dynamic synergy of process parameters, characterized in that, The management and control system comprises: A data acquisition module is configured to acquire real-time process parameters of a full process of carbon fiber prepreg production and dynamically correct parameter measurement composite deviation caused by equipment aging and environmental interference factors, wherein the real-time process parameters include operation parameters of each production link such as resin film supply, carbon fiber dry cloth supply, impregnation, rolling and traction; A dynamic collaborative analysis module is configured to acquire the acquired real-time process parameters, and is provided with a multi-physical field and digital twin fusion analysis model, the model constructs a virtual environment corresponding to an actual production scene through digital twin technology, integrates four-dimensional coupling relationship of temperature field, flow field, stress field and chemical field in the production process, and performs prediction and root positioning of process parameter deviation; A management and control instruction generation module is connected with the dynamic collaborative analysis module, automatically adjusts multi-objective optimization weights of quality standard, energy consumption control and equipment protection according to real-time production requirements, and generates dynamic management and control instructions meeting multi-objective collaboration, wherein the real-time production requirements include order quality level, energy consumption quota and equipment load state; An execution module is configured to accept the dynamic management and control instructions, synchronously drive associated production equipment to perform adjustment operation, and compensate for response delay difference of different equipment; A quality monitoring feedback module is connected with the execution module and the dynamic collaborative analysis module, adjusts quality detection standards according to resin batch characteristics, and fuses influence of carbon fiber characteristics on the quality standards.

2. The system of claim 1, wherein, The data acquisition module comprises: A multi-dimensional calibration reference database subunit is configured to store dynamic calibration reference data of the equipment throughout the life cycle; wherein the reference data includes sensor calibration parameters under different temperature conditions in the conventional environmental temperature interval of industrial production when the equipment is shipped, sensor optimal working parameters inversely deduced based on the quality qualified data of the prepreg batch after each production batch is completed, and sensor emergency calibration parameters under extreme production conditions; A deviation and health degree correlation analysis subunit is configured to real-time calculate deviation proportion of the current acquisition parameters and the same condition reference data, simultaneously monitor sensor operation state parameters, real-time evaluate the sensor state through a corresponding model of the deviation proportion and the health degree, and trigger a warning when the sensor health degree is lower than a set threshold or the deviation proportion exceeds a limited range; A predictive calibration execution subunit is configured to generate sensor calibration instructions when the sensor health degree is lower than a preset threshold or the parameter deviation proportion exceeds a preset proportion for a continuous preset time length; the calibration process incorporates environmental factor compensation, and outputs a remaining service life prediction result of the sensor; A calibration and production collaboration subunit is configured to call a multi-physical field and digital twin fusion analysis model in the dynamic collaborative analysis module when sensor calibration needs to temporarily interrupt data acquisition of the sensor, and generate predicted values of corresponding process parameters in the interruption period.

3. The system of claim 2, wherein, The environmental compensation of the predictive calibration execution subunit comprises: An environmental parameter acquisition subunit is configured to real-time acquire key parameters of the workshop environment, the key parameters including temperature, humidity and air pressure, and each parameter measurement accuracy is controlled within a preset range; An environment and calibration coefficient mapping database subunit is configured to store sensor calibration coefficients under different environment parameters, and to call corresponding calibration coefficients to compensate for the influence of the environment on sensor measurement when the environment parameters change; In the dynamic calibration, the calibration coefficient corresponding to the current environment is called to generate a calibration instruction to correct the sensor calibration target value; after calibration, a preset number of sensor real-time measurement data are collected to calculate the deviation ratio from the corrected reference value; if the deviation ratio is within the limited range, the calibration is determined to be qualified, otherwise the environmental compensation coefficient is adjusted again and calibration is performed again.

4. The system of claim 1, wherein, The multi-physical field and digital twin fusion analysis model of the dynamic collaborative analysis module is based on the description of resin infiltration reaction, fluid flow, heat transfer, and solid stress relationship to build a four-dimensional coupled calculation model; through digital twin technology, the physical structure parameters of the production equipment, the real-time collected process parameters, and the workshop environment parameters are mapped to a virtual production environment corresponding to the actual production scene 1:1, and the visualization simulation of the process parameter change process is performed; The dynamic collaborative analysis module simulates the trend of changes in each process parameter within a preset time in the future through the virtual production environment, and when the actual production deviates from the parameters, the digital twin model is used for parameter inversion to trace the root cause of the deviation caused by the coupling of the four physical and chemical actions; The dynamic collaborative analysis module includes a model self-optimization subunit that optimizes the model by integrating cross-factory anonymous production data through federated learning technology.

5. The system of claim 1, wherein, The control instruction generation module includes: The reinforcement learning training subunit maximizes the quality standard rate, minimizes energy consumption, and minimizes equipment wear and tear as multi-dimensional reward criteria, and trains the intelligent computing unit through historical production data to automatically adjust the multi-objective optimization weight according to real-time working conditions; The multi-scheme simulation evaluation subunit is used to generate multiple sets of preliminary control instructions under dynamic weights, simulate the execution effect of each instruction through the multi-physical field and digital twin fusion analysis model in the dynamic collaborative analysis module, and calculate the comprehensive benefits of each scheme; The dynamic instruction optimization subunit is used to select the preliminary control instruction scheme with the highest comprehensive benefits and adjust the instruction sending time difference according to the response characteristics of different equipment to synchronize the associated equipment to reach the target parameters; The extreme working condition emergency subunit is used to automatically increase the pre-impregnated material quality guarantee weight to a preset value when a sudden failure occurs in the production process, generate emergency control instructions, and call the backup parameter acquisition channel.

6. The management system according to claim 1, wherein The quality monitoring feedback module includes: The multi-raw material characteristic acquisition subunit is used to obtain resin characteristic parameters and carbon fiber characteristic parameters, and to obtain real-time permeability data of carbon fiber and resin through the built-in carbon fiber and resin permeability measuring device in carbon fiber; The quality standard dynamic generation subunit is configured to establish a corresponding rule of carbon fiber characteristics, resin characteristics, and quality detection standards, and to adjust the quality standard according to the change of raw material characteristics. If the resin permeability is lower than the benchmark value, the prepreg infiltration performance standard is automatically adjusted, and the process parameter adjustment is triggered to enhance the interfacial bonding effect. The resin characteristic parameters include resin molecular weight distribution range and resin viscosity, and the carbon fiber characteristic parameters include carbon fiber single filament diameter, carbon fiber surface roughness, and carbon fiber surface sizing agent content. The quality and parameter traceability subunit is configured to trace the production parameter change and the raw material characteristic change when the prepreg quality deviates, and to generate a raw material adjustment suggestion. The multi-batch data mining subunit is configured to analyze the historical raw material, parameter, and quality data, to mine the implicit correlation between the data, and to update the corresponding rule of carbon fiber characteristics, resin characteristics, and quality detection standards.

7. The management system according to claim 6, wherein The integration and use mode of the carbon fiber and resin in carbon fiber permeability measuring device includes the following contents: A miniaturized carbon fiber and resin in carbon fiber permeability measuring device is arranged at the material outlet position of the impregnation device. The device uses an ultrasonic sensor testing method to collect data from the production line in real time and complete the measurement of the resin permeability in the carbon fiber. If the measurement value of a batch is lower than the preset benchmark value, the quality monitoring feedback module automatically reduces the prepreg standard of the batch, and triggers the dynamic collaborative analysis module to adjust the related process parameters. Each batch of data, the adjusted quality standard, and the process parameter adjustment record are included in the batch production file.

8. The management system according to claim 4, wherein, The deviation prediction process of the multi-physical field and digital twin fusion analysis model is as follows: The three-dimensional structure model of the production equipment is imported into the model, the current workshop environment parameters and raw material characteristic parameters are input, and a virtual production environment corresponding to the actual production scene is established. The model receives real-time process parameters transmitted by the data acquisition module every preset time interval, and updates the corresponding parameter values in the virtual production environment. The influence of the four physical and chemical actions on the process parameters is calculated by the model, the trend of the change of each process parameter in the future is simulated, and the trend of the change of each process parameter in the future is simulated. If the simulation result shows that a process parameter will exceed the preset benchmark range, the model immediately outputs a deviation warning information, and marks the root cause of the deviation caused by the coupling of the four physical and chemical actions, and triggers the control instruction generation module to generate an adjustment instruction.

9. The management system of claim 5, wherein, The steps of the multi-objective optimization weight in the control instruction generation module are as follows: The production data of historical production batches are collected, the data cover process parameter data, quality detection data, energy consumption statistical data, and equipment load data under different production conditions, and each group of data is labeled with quality compliance rate, actual energy consumption value, and equipment actual load rate. The reward calculation method is comprehensive income = prepreg quality compliance rate × quality target weight + (1-actual energy consumption rate / rated energy consumption rate) × energy consumption target weight + (1-equipment actual load rate / equipment rated load rate) × equipment protection target weight, and the intelligent calculation unit is iteratively trained for sufficient times to make the intelligent calculation unit learn the optimal weight combination under different conditions. In the production process, the system acquires production demand information, energy consumption monitoring data and equipment load data in real time, and the intelligent computing unit automatically adjusts the three-dimensional target weight according to the acquired information to generate control instructions based on the adjusted weight; After each production batch is completed, the actual quality, energy consumption and equipment data of the batch are fed back to the intelligent computing unit.

10. The management system of claim 1, wherein, The execution module comprises: A device response delay database subunit for storing response characteristic data of each production device; A delay compensation calculation subunit for determining the transmission time difference of each device instruction according to the response delay difference of the associated devices to ensure that the associated devices synchronously reach the target parameters; A collaborative execution monitoring subunit for monitoring the progress of each device in executing the adjustment instructions in real time, and if the execution progress of a certain device exceeds the expected delay, the execution speed of other associated devices is immediately adjusted to wait for the completion of the adjustment of the device; An emergency coordination subunit for sending emergency adjustment instructions to all associated devices immediately when a certain production device suddenly fails.

Images