A photovoltaic module production quality monitoring system based on digital twinning

Abstract

This invention discloses a photovoltaic module production quality monitoring system based on digital twins, belonging to the technical field of production quality monitoring systems. The system includes a dynamic monitoring interval division module for generating dynamic monitoring time intervals; a multi-source data fusion module for collecting equipment operation data and quality inspection data from each process stage within the dynamic monitoring time interval to generate fused data; a digital twin virtual mapping module for constructing a virtual model of the photovoltaic production line based on the fused data and outputting virtual mapping relationships; an intelligent quality control module for generating equipment control commands based on the virtual mapping relationships; a physical equipment execution module for executing equipment control commands and outputting equipment operating parameters; and a real-time feedback module for collecting equipment operating parameters and feeding them back to the dynamic monitoring interval division module. This invention solves the problems of transient feature dilution and insufficient sampling caused by traditional fixed time intervals by generating monitoring time intervals.

Description

Technical Field

[0001] This invention relates to the field of production quality monitoring system technology, and in particular to a photovoltaic module production quality monitoring system based on digital twins. Background Technology

[0002] In the global energy transition, the photovoltaic industry, as a core pillar of clean energy, is facing an urgent need to improve quality and efficiency. The production process of photovoltaic modules involves the collaboration of multiple processes, such as cell welding, lamination, and encapsulation, and its quality directly affects the efficiency and lifespan of the modules.

[0003] Existing photovoltaic module production quality monitoring systems typically use sensor networks deployed in various production processes to collect real-time equipment operating parameters such as laminator temperature, stringer current, and laying table positioning accuracy. These multi-source monitoring data are then uploaded to a central monitoring platform via industrial Ethernet. Through data analysis and visualization, the system enables centralized monitoring and alarm management of the production line equipment status.

[0004] However, existing photovoltaic module production quality monitoring systems generally use fixed time intervals to divide monitoring zones, which is difficult to adapt to the nonlinear continuous process with significant differences in reaction rates at different stages of photovoltaic module production. This results in the dilution of transient characteristics of key parameters in high-speed processes, while data sampling for low-speed processes is insufficient. Simultaneously, there is a lack of deep integration between equipment operation data and quality inspection data. Taking the photovoltaic lamination process as an example, process parameters such as temperature and pressure are separated from module defect data in independent systems, preventing virtual models from constructing a correlation mapping between process parameters and quality indicators, thus hindering further improvements in quality control accuracy. Therefore, there is an urgent need to provide a photovoltaic module production quality monitoring system based on digital twins to solve these problems. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of existing photovoltaic module production quality monitoring systems, which generally use fixed time intervals to divide monitoring intervals. These systems are difficult to adapt to the nonlinear continuous process with significant differences in reaction rates at different stages of photovoltaic module production, resulting in the dilution of transient characteristics of key parameters in high-speed processes and insufficient data sampling in low-speed processes. At the same time, there is a lack of deep integration between equipment operation data and quality inspection data. Taking the photovoltaic lamination process as an example, process parameters such as temperature and pressure and module defect data are placed in independent systems, which makes it impossible for the virtual model to build a correlation mapping between process parameters and quality indicators. This restricts the further improvement of quality control accuracy. The invention provides a photovoltaic module production quality monitoring system based on digital twins.

[0006] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: to provide a photovoltaic module production quality monitoring system based on digital twin, including a dynamic monitoring interval division module, a multi-source data fusion module, a digital twin virtual mapping module, an intelligent quality control module, a physical equipment execution module, and a real-time feedback module; The dynamic monitoring interval division module is used to generate dynamic monitoring time intervals in real time according to each process stage in the photovoltaic module production line. The multi-source data fusion module is used to collect equipment operation data and quality inspection data of each process stage within the dynamic monitoring time interval, and to generate fused data after processing the equipment operation data and the quality inspection data. The digital twin virtual mapping module constructs a virtual model of the photovoltaic production line based on the fused data and outputs virtual mapping relationships; The intelligent quality control module generates equipment control commands based on the virtual mapping relationship; The physical device execution module executes the device control commands and outputs the device operating parameters; The real-time feedback module collects the operating parameters of the equipment and feeds them back to the dynamic monitoring interval division module.

[0007] The present invention is further configured such that: each process stage in the photovoltaic production line includes a string bonding process, a lamination process, and a packaging process; a sensor network is deployed in each process stage to obtain real-time sensor data streams; the real-time sensor data streams include: in the string bonding process, a Hall current sensor is used to collect the welding current pulse waveform, and an infrared thermal imaging device is used to capture the temperature field distribution of the solder joint; in the lamination process, a high-temperature capacitive sensor is used to obtain the temperature gradient of the laminate in real time, and a vacuum pressure transmitter is used to monitor the pressure change curve of the chamber; in the packaging process, a photoelectric encoder is used to record the frame assembly displacement trajectory.

[0008] The present invention is further configured such that: the step of generating the dynamic monitoring time interval in the dynamic monitoring interval division module is as follows: S1. Obtain the real-time operating characteristics of each process stage through the preset process feature extraction unit, and generate process stage identifiers; S2. Call the preset rate matching unit according to the process stage identifier, and match the baseline reaction rate parameter of each process stage from the preset process knowledge base. S3. Based on the benchmark reaction rate parameter and the real-time sensor data stream, the real-time monitoring interval is calculated, and a dynamic monitoring time interval is generated according to the real-time monitoring interval. The dynamic monitoring time interval includes a millisecond-level monitoring interval and a second-level monitoring interval. The millisecond-level monitoring interval corresponds to the high-speed reaction process, and the second-level monitoring interval corresponds to the low-speed reaction process.

[0009] The present invention is further configured such that: the specific content of generating the dynamic monitoring time interval based on the real-time monitoring interval in step S3 is as follows: S31. Perform continuous preset time period scanning on the real-time monitoring interval of each process stage, extract the real-time monitoring intervals below the first preset threshold to form a first dataset, and assign a unique sub-index to each real-time monitoring interval in the first dataset. S32. The real-time monitoring intervals that are higher than the second preset threshold within a continuous preset time period are sorted and reorganized according to the timestamp sequence to generate multiple second datasets divided according to a preset period, and the missing positions in each second dataset are marked with a main index identifier. S33. Map the preset time period corresponding to the first dataset to a millisecond-level monitoring interval, and at the same time map the preset period of the second dataset to a second-level monitoring interval. According to the spatial correspondence between the sub-index and the main index, embed the millisecond-level monitoring interval into the empty position of the second-level monitoring interval. S34. Set a buffer zone at the beginning of the millisecond-level monitoring interval, and connect the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer zone to generate a dynamic monitoring time interval.

[0010] The present invention is further configured such that: the specific method for connecting the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer interval in step S34 is as follows: S341. Multiple preset transition point data are continuously distributed in the buffer zone, and the first and last transition point data in the buffer zone are respectively corrected to the real-time monitoring interval at the beginning of the millisecond-level monitoring interval and the real-time monitoring interval at the beginning of the second-level monitoring interval. S342. According to the preset interpolation algorithm in the transition unit, based on the corrected first transition point data and the last transition point data, interpolation calculation is performed on the remaining transition point data in the buffer to generate a transition point data sequence, and the millisecond-level monitoring interval and the second-level monitoring interval are connected through the transition point data sequence.

[0011] The present invention is further configured such that: the device operation data in the multi-source data fusion module is the real-time sensor data stream; The quality inspection data is collected through a pre-set online optical inspection system, specifically including: deploying a machine vision device to extract the solder strip alignment deviation vector in the subsequent process of the stringing process; setting a laser triangulation instrument to scan the EVA film thickness distribution at the exit end of the lamination process; and using a spectrophotometer to measure the transmittance attenuation coefficient of the photovoltaic module in the encapsulation process. The device operation data and the quality inspection data are timestamped and spatially registered through a preset edge computing gateway, and then windowed and recombined according to the dynamic monitoring time interval to generate fused data.

[0012] The present invention is further configured such that the construction process of the photovoltaic production line virtual model in the digital twin virtual mapping module is as follows: Q1. The fused data generated by windowed slicing and recombining is converted into a virtual entity attribute set, wherein the device operation data stream is mapped to a virtual device dynamic parameter cluster, the quality inspection data stream is mapped to a virtual quality index matrix, and the time-series coordinate axis is divided based on the dynamic monitoring time interval; Q2. Using the millisecond-level monitoring interval and the second-level monitoring interval on the time-series coordinate axis as time reference templates, high-speed reaction process model units and low-speed reaction process model units are synchronously generated in a preset virtual space. The virtual equipment dynamic parameter cluster drives the dynamic behavior simulation of the high-speed reaction process model unit, and the virtual quality index matrix is ​​injected into the quality performance evaluation of the low-speed reaction process model unit. Simultaneously, the transition point data sequence in the buffer zone is converted into spatiotemporal connection constraints. The high-speed reaction process model unit and the low-speed reaction process model unit are connected through the spatiotemporal connection constraints to generate a photovoltaic production line virtual model.

[0013] The present invention is further configured such that: the virtual mapping relationship in the intelligent quality control module includes mapping the dynamic behavior simulation results in the high-speed reaction process model unit to the real-time operating status of the stringing process in the physical photovoltaic module production line, mapping the quality performance evaluation results in the low-speed reaction process model unit to the quality inspection indicators of the lamination process and the encapsulation process in the physical photovoltaic module production line, and simultaneously mapping the process connection relationship reconstructed in the virtual space through the spatiotemporal connection constraints to the process transition state between the stringing process and the lamination process, and the process transition state between the lamination process and the encapsulation process in the physical photovoltaic module production line.

[0014] The present invention is further configured such that: the equipment control instructions of the intelligent quality control module include a string soldering process control instruction generated based on the deviation between the real-time operating status of the string soldering process mapped in the virtual mapping relationship and the preset standard operating status; a lamination process control instruction generated based on the deviation between the quality inspection index of the lamination process mapped in the virtual mapping relationship and the preset standard quality index; and a packaging process control instruction generated based on the deviation between the quality inspection index of the packaging process mapped in the virtual mapping relationship and the preset standard quality index.

[0015] The present invention is further configured such that the physical device execution module includes a serial soldering process execution unit, a lamination process execution unit, and a packaging process execution unit.

[0016] The beneficial effects of this invention are as follows: 1. This invention analyzes the reaction rate of each process stage in real time through a dynamic monitoring interval division module. Based on the dual scale of milliseconds and seconds, it dynamically generates monitoring time intervals to accurately match the transient characteristics of high-speed string soldering processes with the gradual changes of low-speed lamination / packaging processes, effectively solving the problems of transient feature dilution and insufficient sampling caused by traditional fixed time intervals. 2. This invention utilizes a buffer transition mechanism and an index mapping strategy to ensure data continuity during process switching, significantly improving the complete capture rate of key process parameters and providing high-fidelity input for virtual model construction; 3. This invention relies on an edge computing gateway to achieve spatiotemporal registration and windowed slicing reorganization of equipment operation data and quality inspection data, constructing a fused data stream; through virtual mapping relationships, it maps the simulation results of dynamic behavior of high-speed processes to the real-time operation status of string soldering, maps the quality evaluation results of low-speed processes to lamination / packaging quality indicators, and synchronously maps the process transition status between processes, completely solving the problem of missing correlation mapping caused by data silos in traditional systems, and significantly improving the accuracy of quality control and the efficiency of anomaly location. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart illustrating the steps for generating the dynamic monitoring time interval of the present invention. Figure 3 This is a flowchart illustrating the specific steps of step S3 of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0019] Please see Figure 1 - Figure 3 A photovoltaic module production quality monitoring system based on digital twins includes a dynamic monitoring interval division module, a multi-source data fusion module, a digital twin virtual mapping module, an intelligent quality control module, a physical equipment execution module, and a real-time feedback module. The dynamic monitoring interval division module is used to generate dynamic monitoring time intervals in real time according to each process stage in the photovoltaic module production line. The multi-source data fusion module is used to collect equipment operation data and quality inspection data of each process stage within the dynamic monitoring time interval, and generate fused data after processing the equipment operation data and quality inspection data; The digital twin virtual mapping module constructs a virtual model of a photovoltaic production line based on fused data and outputs virtual mapping relationships; The intelligent quality control module generates equipment control commands based on the virtual mapping relationship; The physical device execution module executes device control commands and outputs device operating parameters. The real-time feedback module collects equipment operating parameters and feeds them back to the dynamic monitoring interval division module.

[0020] Specifically, the photovoltaic production line includes stringing, lamination, and encapsulation processes. A sensor network is deployed at each process stage to acquire real-time sensor data streams. These real-time data streams include: in the stringing process, Hall current sensors are used to collect welding current pulse waveforms, and infrared thermal imaging devices capture the temperature field distribution at the solder joints; in the lamination process, high-temperature capacitive sensors are used to acquire the laminate temperature gradient in real time, and vacuum pressure transmitters monitor the chamber pressure change curves; in the encapsulation process, photoelectric encoders are used to record the frame assembly displacement trajectory. The Hall current sensor, infrared thermal imaging device, high-temperature capacitive sensor, vacuum pressure transmitter, and photoelectric encoder constitute the sensor network.

[0021] The dynamic monitoring interval division module accurately generates millisecond and second-level monitoring intervals adapted to the real-time response rates of each process stage, effectively matching the monitoring needs of high-speed stringing processes and low-speed lamination and encapsulation processes. Relying on a sensor network consisting of Hall current sensors, infrared thermal imaging devices, high-temperature capacitive sensors, vacuum pressure transmitters, and photoelectric encoders deployed in each process, and combined with an edge computing gateway, the timestamps of equipment operation data and quality inspection data are aligned and spatially registered. High-precision fused data is generated through windowed slicing and recombination. The photovoltaic production line virtual model built based on the fused data maps the simulation results of the dynamic behavior of the high-speed reaction process to the real-time operating status of the stringing process through virtual mapping relationships, and maps the quality performance evaluation results of the low-speed reaction process to the quality inspection indicators of the lamination and encapsulation processes. It also synchronously maps the process transition states between processes, significantly improving the accuracy of virtual-real interaction. The intelligent quality control module generates equipment control commands based on this, which drive the physical equipment execution module to dynamically correct sensor parameters. The real-time feedback module forms a closed-loop optimization mechanism, ultimately achieving quality collaborative control and rapid response to anomalies across the entire photovoltaic module process chain.

[0022] One embodiment of the present invention is as follows: the steps for generating the dynamic monitoring time interval in the dynamic monitoring interval division module are as follows: S1. Obtain the real-time operating characteristics of each process stage through the preset process feature extraction unit, and generate process stage identifiers; The specific implementation of the process feature extraction unit is as follows: The process feature extraction unit extracts the current pulse frequency feature value by real-time analysis of the welding current pulse waveform collected by the Hall current sensor in the string welding process; it simultaneously processes the temperature field distribution of the solder joint captured by the infrared thermal imaging device to extract the temperature gradient dispersion feature value; for the lamination process, it analyzes the laminate temperature gradient data obtained by the high-temperature capacitive sensor to extract the temperature change rate feature value; it combines the chamber pressure change curve monitored by the vacuum pressure transmitter to extract the pressure fluctuation variance feature value; for the packaging process, it analyzes the frame assembly displacement trajectory recorded by the photoelectric encoder to extract the displacement trajectory linearity feature value; and it combines the above process feature values ​​into a multi-dimensional feature vector as the real-time running feature output of the process stage.

[0023] The steps for generating process stage identifiers are as follows: The current pulse frequency feature value and temperature gradient dispersion feature value extracted from the stringing process are input into a preset encoder, and a hexadecimal process code with the first digit "HX" is generated through feature mapping. The temperature change rate feature value and pressure fluctuation variance feature value extracted from the lamination process are normalized and then weighted and fused according to the feature weight table in a preset process knowledge base to generate a process code segment with the median "LY". The displacement trajectory linearity feature value extracted from the packaging process is converted into a standard process feature code, generating a process code segment with the last digit "FZ". The above three process codes are concatenated according to the process execution order, and a unique eight-digit process stage identifier is generated through a verification algorithm.

[0024] S2. Based on the process stage identifier, call the preset rate matching unit to match the baseline reaction rate parameters of each process stage from the preset process knowledge base. Preset Rate Matching Unit: The preset rate matching unit extracts the process type identifier and feature value check bits by parsing the process code segment (HX / LY / FZ) in the process stage identifier. Based on the process type identifier, it activates the corresponding rate matching channel and compares the feature value check bits with the feature value ranges in the preset process knowledge base step by step: First, in the string soldering process channel, it matches the joint check range of the current pulse frequency feature value and the temperature gradient dispersion feature value; second, in the lamination process channel, it matches the dynamic weight range of the temperature change rate feature value and the pressure fluctuation variance feature value; finally, in the packaging process channel, it matches the segmented threshold range of the displacement trajectory linearity feature value. Based on the center value of the range where the feature value is located, it extracts the baseline reaction rate parameter for that process from the process knowledge base and outputs millisecond-level or second-level baseline parameters, completing the accurate matching of multi-process baseline parameters; millisecond-level baseline parameters correspond to high-speed string soldering processes; second-level baseline parameters correspond to low-speed lamination / packaging processes.

[0025] Process Knowledge Base: The process knowledge base stores standard process parameter matrices for the three major processes of photovoltaic modules: the string welding process parameter matrix includes the frequency reference value of the welding current pulse waveform, the gradient dispersion reference value of the solder joint temperature field distribution, and its joint verification interval threshold; the lamination process parameter matrix includes the rate of change of the laminate temperature gradient, the fluctuation variance reference value of the chamber pressure change curve, and its dynamic weight matching coefficient; the encapsulation process parameter matrix includes the linearity reference value of the frame assembly displacement trajectory and its segmented calibration threshold; all parameters are classified and stored as three-dimensional parameter matrices according to process type (HX / LY / FZ). Each matrix dimension includes a reference value, a floating range, and a rate correlation factor. The rate correlation factor is used to map process feature values ​​to millisecond-level or second-level reference response rate parameters, forming a structured process database that supports rate matching unit calls.

[0026] The process knowledge base referred to in this system is a structured database of parameters and rules used to store and manage physical models, historical experience, and optimization parameters for specific processes. Its design follows the principles of ensuring traceability and reproducibility.

[0027] Data Structure: Relational tables and parameter matrices are used for storage. Core tables include the process baseline parameter table, feature sensitivity coefficient table, and process context table. Each parameter is associated with its process stage, equipment model, material batch, and data source tags, such as "Simulation DOE", "Historical Statistics", and "Expert Experience".

[0028] Construction method: benchmark value ( Acquisition: derived from design values ​​for standard process conditions, validated historical production statistical averages, or optimization center points from calibration experiments; Rate correlation factor ( ) and base rate ( Calibration: Fitting and validation are performed using controlled experimental design (DOE) combined with multiple linear regression or physics-based simulation models such as finite element analysis. Each coefficient is accompanied by a confidence interval and the source experiment ID. Attenuation coefficient ( Determined by: Based on process characteristics, the nonlinear saturation range of the influence of characteristic deviations on the rate is analyzed, usually derived from process mechanism models or analysis of historical anomaly data with high confidence.

[0029] The specific form of the stored content: stored in the form of a parameter matrix.

[0030] S3. Based on the baseline reaction rate parameter and real-time sensor data stream, the real-time monitoring interval is calculated, and a dynamic monitoring time interval is generated according to the real-time monitoring interval. The formula for calculating the real-time monitoring interval is as follows: in, Real-time monitoring interval, unit: milliseconds; It is a preset benchmark value derived from the corresponding process in the process knowledge base, and its dimension is time; The number of process characteristic values, string welding process =2, lamination process =2, Packaging process =1; The deviation between the real-time feature value and the benchmark value, i.e. - Real-time feature values: generated by the process feature extraction unit, based on real-time data streams collected by the sensor network. Examples of real-time data stream features include current pulse frequency and temperature gradient dispersion features for the stringing process; temperature change rate and pressure fluctuation variance features for the lamination process; displacement trajectory linearity features for the packaging process; and rate correlation factors. Dimensions and real-time reaction rate Same, i.e., millimeters per second. The physical meaning can be understood as: the first The unit relative deviation of each characteristic value ( =1) Regarding the final reaction rate The contribution of this feature is measured in units of length / time, such as millimeters per second. Its value needs to be predefined in the process knowledge base and can be obtained through historical data, process simulation or experimental calibration. It represents the sensitivity weight of this feature to the process rate. This formula is a parameterized soft measurement model used for online estimation of the optimal monitoring sampling frequency to adapt to real-time process fluctuations. This measure achieves nonlinear suppression of large deviations in eigenvalues, ensuring real-time monitoring intervals for computation. It is both sensitive and stable. When the characteristic deviation... When the impact is small, it accumulates almost linearly, and the system shortens the monitoring interval. To increase the sampling frequency; when When the value is large, its impact is exponentially decayed, preventing the monitoring window from becoming inaccurate due to a single feature's instantaneous large fluctuation, thus ensuring the system's robustness under abnormal operating conditions.

[0031] The normalization bias is a dimensionless number, and its calculation formula is:

[0032] This allows feature values ​​from different sensors with different physical meanings, such as frequency, temperature, and pressure, to be weighted and fused at the same scale, i.e., relative rate of change, eliminating the influence of dimensions and improving the numerical stability and physical interpretability of the calculation. These are the feature reference values, which are derived from the reference values ​​stored in the corresponding process parameter matrix in the process knowledge base.

[0033] These feature values ​​are extracted into multi-dimensional feature vectors by real-time analysis of data from devices such as Hall current sensors and infrared thermal imaging devices. These multi-dimensional feature vectors include features such as current pulse frequency. The baseline values ​​originate from standard process parameter matrices in a pre-set process knowledge base. For example, the string soldering process parameter matrix stores the frequency baseline value of the welding current pulse waveform and the gradient dispersion baseline value of the solder joint temperature field distribution; the lamination process parameter matrix stores the rate of change baseline value of the laminate temperature gradient; and the packaging process parameter matrix stores the linearity baseline value of the frame assembly displacement trajectory. These baseline values ​​are categorized by process type (HX / LY / FZ) in the process knowledge base and include a floating range. The feature reference values ​​are stored in the process knowledge base parameter matrix; The dynamic attenuation coefficient is determined by the joint verification interval / segmentation threshold in the process knowledge base.

[0034] The formula for calculating the real-time monitoring interval is a nonlinear soft-sensor model designed for monitoring complex industrial processes, incorporating a process knowledge base. Its technical basis lies in: Soft measurement technology framework: using measurable auxiliary variables, i.e. eigenvalues, to estimate the dominant variable, i.e., the reaction rate, online.

[0035] Weighted fusion method with exponential decay penalty: through The method achieves adaptive and nonlinear weighting of eigenvalue deviations, with small deviations accumulating linearly and large deviations causing exponential decay, thus ensuring the stability and anti-interference capability of the rate estimate. This is the core innovation of the method.

[0036] Parametric models based on process knowledge: The baseline values, correlation factors, and attenuation coefficients in the formulas are all derived from a pre-built process knowledge base, ensuring the physical interpretability of the model and its specific adaptability to specific processes.

[0037] Example: In the photovoltaic module string welding process, the system collects the characteristic value of the welding current pulse frequency in real time ( =152Hz) and the characteristic value of the temperature gradient dispersion of the solder joint ( =0.85), obtain the baseline value from the process knowledge base ( =150Hz, =0.8) and parameters ( =0.6, =0.4, =3.2, =2.8, =120ms); Calculate the deviation =2Hz, =0.05, substituting into the formula, we get the real-time monitoring interval. =120+[0.6×2× +0.4×0.05× The time interval is approximately 120.019ms, from which a millisecond-level monitoring interval is generated.

[0038] By using dynamic weighted attenuation calculation to accurately quantify the impact of process characteristic deviations on the reaction rate, adaptive control of millisecond-level response in high-speed processes is achieved, significantly improving the accuracy of virtual-real mapping and the speed of abnormal response, and strengthening the quality collaborative control capability of the entire process chain.

[0039] The dynamic monitoring time interval includes millisecond-level monitoring intervals and second-level monitoring intervals. The millisecond-level monitoring interval corresponds to high-speed response processes, and the second-level monitoring interval corresponds to low-speed response processes.

[0040] Specifically, the content of generating the dynamic monitoring time interval based on the real-time monitoring interval in step S3 is as follows: S31. Perform continuous preset time period scanning on the real-time monitoring interval of each process stage, extract the real-time monitoring intervals below the first preset threshold to form a first dataset, and assign a unique sub-index to each real-time monitoring interval in the first dataset. First preset threshold: High-speed process - string welding: The first preset threshold range is 10-50ms, and the matching welding pulse frequency is 100-500Hz; Low-speed process - lamination: The first preset threshold range is 200-500ms, adapting to temperature gradient change cycles; Low-speed process - packaging: The first preset threshold range is 150-400ms, corresponding to the displacement trajectory sampling interval; Preset time intervals: Stringing process: 80-120ms; Lamination process: 1.5-3.0s; Encapsulation process: 0.8-1.8s; S32. Sort and reorganize the real-time monitoring intervals that are higher than the second preset threshold within a continuous preset time period according to the timestamp sequence to generate multiple second datasets divided according to the preset period, and mark the main index identifier for the missing positions in each second dataset. Second preset threshold: High-speed process - string welding process: Second preset threshold range 60-100ms; Low-speed process - lamination process: Second preset threshold range 600ms-1.0s; Low-speed process - packaging process: second preset threshold range 500-700ms; Preset cycle times: Stringing process: 80-120ms; Lamination process: 1.5-3.0s; Encapsulation process: 0.8-1.8s; For the real-time monitoring interval between the second preset threshold and the first preset threshold, the following steps are performed: mark it as a transitional data point and assign a temporary index. When generating the second dataset, insert it as a boundary connection point at the beginning or end of the second dataset. Combined with the missing labeling mechanism of the main index, the data breakpoint is compensated by the spatiotemporal interpolation algorithm. The transition weight is calibrated by the rate correlation factor in the associated process knowledge base. Finally, a smooth transition connection between high-speed reaction data and low-speed reaction data under the preset period division is achieved.

[0041] S33. Map the preset time period corresponding to the first dataset to a millisecond-level monitoring interval, and at the same time map the preset period of the second dataset to a second-level monitoring interval. Based on the spatial correspondence between the sub-index and the main index, embed the millisecond-level monitoring interval into the vacant position of the second-level monitoring interval. S34. Set a buffer zone at the beginning of the millisecond-level monitoring interval, and connect the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer zone to generate a dynamic monitoring time interval.

[0042] Specifically, the method for connecting the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer in step S34 is as follows: S341. Multiple preset transition point data are continuously distributed in the buffer zone. The transition point data is the virtual real-time monitoring interval data. The first transition point data and the last transition point data in the buffer zone are corrected to the real-time monitoring interval data at the beginning of the millisecond-level monitoring interval and the real-time monitoring interval data at the beginning of the second-level monitoring interval, respectively. S342. According to the interpolation algorithm preset in the transition unit, based on the corrected first transition point data and the last transition point data, interpolate the remaining transition point data in the buffer to generate a transition point data sequence, and connect the millisecond-level monitoring interval and the second-level monitoring interval through the transition point data sequence.

[0043] Results: The process feature extraction unit accurately captures feature values ​​such as the frequency of string welding current pulses, solder joint temperature dispersion, lamination temperature change rate, chamber pressure variance, and packaging displacement linearity. Combined with the benchmark values ​​and rate correlation factors of the structured parameter matrix in the process knowledge base, it drives the adaptive division of millisecond-level and second-level monitoring intervals. Based on the preset time period and dual thresholds (i.e., the first preset threshold / second preset threshold scanning mechanism), high-speed response data and low-speed response data are classified into a first dataset and a second dataset. The first dataset has a sub-index, and the second dataset has a main index identifier. The transition between datasets is achieved through transitional data points and spatiotemporal interpolation algorithms. The transition point data sequence in the buffer zone connects the millisecond-level and second-level monitoring intervals. Combined with virtual mapping relationships and equipment control commands, a closed-loop optimization of the entire process chain is formed, significantly improving the accuracy of virtual-real interaction, anomaly response speed, and quality collaborative control capabilities in photovoltaic module production.

[0044] One embodiment of the present invention is: the device operation data in the multi-source data fusion module, i.e., the real-time sensor data stream; Quality inspection data is collected through a pre-set online optical inspection system, specifically including: deploying a machine vision device to extract the solder strip alignment deviation vector in the subsequent process of stringing; setting a laser triangulation instrument to scan the EVA film thickness distribution at the exit of the lamination process; and using a spectrophotometer to measure the transmittance attenuation coefficient of the photovoltaic module in the encapsulation process. Equipment operation data and quality inspection data are timestamped and spatially registered through a preset edge computing gateway, and then windowed and recombined according to the dynamic monitoring time interval to generate fused data.

[0045] EVA film thickness distribution refers to the two-dimensional or three-dimensional spatial thickness variation spectrum formed by scanning the laminated EVA film with a laser triangulation instrument during the photovoltaic module lamination process to obtain the thickness value data at various spatial positions on the module surface. This spectrum characterizes the uniformity of film curing and the integrity of coverage, directly affecting the light transmittance and encapsulation quality of the module, and is one of the core indicators of lamination process quality inspection data.

[0046] By using an edge computing gateway, the device operation data, i.e., real-time sensor data stream, is timestamped and spatially registered with the quality inspection data, which includes the solder strip alignment deviation vector, EVA film thickness distribution, and transmittance attenuation coefficient. The data is then recombined by windowing and reorganizing according to the dynamic monitoring time interval to generate high-precision fused data, which significantly improves the ability of virtual and real data to coordinate during the stringing, lamination, and encapsulation processes in photovoltaic module production.

[0047] One embodiment of the present invention is as follows: the construction process of the photovoltaic production line virtual model in the digital twin virtual mapping module is as follows: Q1. Convert the fused data generated by windowed slicing and reorganization into a set of virtual entity attributes, where the device operation data stream is mapped to a virtual device dynamic parameter cluster, the quality inspection data stream is mapped to a virtual quality index matrix, and the time-series coordinate axis is divided based on the dynamic monitoring time interval. The specific steps for mapping equipment operation data stream into a virtual equipment dynamic parameter cluster are as follows: First, analyze the timestamp sequence and spatial coordinates in the equipment operation data stream; extract the welding current pulse waveform feature points of the serial welding process and the temperature field distribution gradient vector of the solder joint captured by the infrared thermal imaging device; second, reorganize the welding current pulse waveform feature points and the temperature field distribution gradient vector of the solder joint according to the time-series coordinate axis to generate a current-temperature coupling feature cluster within the millisecond-level monitoring interval; third, align the temperature gradient change rate curve of the laminate and the chamber pressure change curve of the lamination process with their fluctuation phases and fuse them into a temperature-pressure dynamic response cluster; fourth, decompose the frame assembly displacement trajectory of the packaging process into displacement velocity vectors and acceleration feature values ​​to construct a displacement dynamic trajectory cluster; fifth, based on the starting and ending points of the second-level monitoring interval on the time-series coordinate axis, aggregate the current-temperature coupling feature cluster, the temperature-pressure dynamic response cluster, and the displacement dynamic trajectory cluster to generate a virtual equipment dynamic parameter cluster covering the entire process chain. The specific steps for mapping the quality inspection data stream into a virtual quality index matrix include: receiving the solder strip alignment deviation vector extracted by the machine vision device from the quality inspection data stream and calculating its Euclidean distance deviation from the standard solder strip position; simultaneously processing the EVA film thickness distribution data scanned by the laser triangulation instrument and extracting the thickness uniformity variance coefficient and edge attenuation gradient value; analyzing the photovoltaic module transmittance attenuation coefficient measured by the spectrophotometer and generating a transmittance attenuation heatmap by associating it with the module surface coordinates; windowing the Euclidean distance deviation, thickness uniformity variance coefficient, edge attenuation gradient value, and transmittance attenuation heatmap into millisecond-level and second-level monitoring intervals on the time-series coordinate axis; based on the timestamp synchronization of the sliced ​​data, mapping the solder strip alignment deviation to a string soldering quality index unit, mapping the film thickness feature value to a lamination quality index unit, and mapping the transmittance attenuation heatmap to a packaging quality index unit; stacking the quality index units in the virtual space according to the process execution order to generate a virtual quality index matrix with spatiotemporal markers. Q2. Using the millisecond-level and second-level monitoring intervals on the time-series coordinate axis as time reference templates, high-speed reaction process model units and low-speed reaction process model units are synchronously generated in a preset virtual space. The dynamic parameter cluster of virtual equipment drives the dynamic behavior simulation of the high-speed reaction process model unit, and the virtual quality index matrix is ​​injected into the quality performance evaluation of the low-speed reaction process model unit. Simultaneously, the transition point data sequence in the buffer zone is transformed into spatiotemporal connection constraints. The high-speed reaction process model unit and the low-speed reaction process model unit are connected through the spatiotemporal connection constraints to generate a virtual model of the photovoltaic production line.

[0048] The steps for generating the high-speed reaction process model unit are as follows: Using the millisecond-level monitoring interval on the time-series coordinate axis as the time reference template, the serial welding process spatial domain is defined in the preset virtual space. The current-temperature coupling feature cluster in the dynamic parameter cluster of the virtual equipment is decomposed into discrete current pulse energy field and temperature gradient vector field according to millisecond-level timestamps. Based on the current pulse energy field, the virtual welding heat source model is driven to generate a dynamic heat field distribution, and the temperature gradient vector field is simultaneously fused to construct a thermodynamic deformation field. The thermodynamic deformation field is used to back-map the weld pool morphology trajectory, and combined with the process constraints of the high-speed reaction process, the dynamic behavior simulation unit of the serial welding process is generated. Finally, all simulation units under millisecond-level timestamps are aggregated to form the high-speed reaction process model unit. The generation steps of the low-speed reaction process model unit are as follows: Using the second-level monitoring interval on the time-series coordinate axis as the time reference template, the joint spatial domain of the lamination process and the encapsulation process is divided in the preset virtual space. The lamination quality index unit in the virtual quality index matrix is ​​injected into the film rheology model to generate the curing stress distribution field. At the same time, the encapsulation quality index unit is input into the optical transmission model to generate the light decay gradient field. Through the coupling effect of the curing stress distribution field and the light decay gradient field, the lamination-encapsulation quality linkage evaluation algorithm is driven to calculate the correlation response between material deformation and optical performance. Based on the correlation response, the quality performance evaluation unit is output. Finally, the evaluation units within the second-level time window are integrated to form the low-speed reaction process model unit. The specific content of dynamic behavior simulation: The dynamic behavior simulation of the high-speed reaction process model unit driven by the dynamic parameter cluster of virtual equipment includes: analyzing the current pulse amplitude sequence and weld point temperature gradient vector in the current-temperature coupling feature cluster; inputting the current pulse amplitude sequence into the virtual resistance welding machine model to generate an arc energy pulse sequence; simultaneously importing the weld point temperature gradient vector into the heat conduction finite element model to calculate the heat diffusion field; solving the transient fluid dynamics equation of the weld point molten pool through the spatiotemporal superposition of the arc energy pulse sequence and the heat diffusion field; predicting the virtual weld strip deformation and thermal stress distribution based on the molten pool flow trajectory, and outputting a millisecond-level dynamic deformation simulation report for the serial welding process to achieve real-time prediction of welding quality defects; The specific content of the quality performance evaluation includes: injecting the virtual quality index matrix into the low-speed reaction process model unit for quality performance evaluation, specifically: extracting the film thickness uniformity variance coefficient from the lamination quality index unit, inputting it into the viscoelastic mechanical model to calculate the curing shrinkage stress of the film; simultaneously analyzing the transmittance attenuation thermogram from the encapsulation quality index unit, simulating the light path refraction loss on the component surface through a ray tracing algorithm; constructing a material-optics coupled response equation by correlating the curing shrinkage stress with the light refraction loss; solving this equation to obtain the quantitative relationship between the probability of lamination bubble defects and the encapsulation transmittance attenuation coefficient, outputting a lamination-encapsulation process quality linkage evaluation report, and achieving accurate prediction of the long-term reliability of the component.

[0049] Results: By accurately mapping the equipment operation data stream to a virtual equipment dynamic parameter cluster, which includes a current-temperature coupling feature cluster, a temperature-pressure dynamic response cluster, and a displacement dynamic trajectory cluster, the quality inspection data stream is converted into a virtual quality index matrix containing string soldering, lamination, and encapsulation quality index units. The monitoring intervals are divided into millisecond and second levels based on the time-series coordinate axis. High-speed reaction process model units are generated synchronously in the virtual space to achieve real-time prediction of welding quality defects through dynamic behavior simulation, and low-speed reaction process model units to achieve accurate prediction of long-term component reliability through quality performance evaluation. Multi-process models are connected using spatiotemporal connection constraints, ultimately constructing a high-fidelity photovoltaic production line virtual model, significantly improving the accuracy of virtual-real mapping and the collaborative quality control capability of the entire process chain.

[0050] One embodiment of the present invention is as follows: the virtual mapping relationship in the intelligent quality control module includes mapping the dynamic behavior simulation results in the high-speed reaction process model unit to the real-time operating status of the stringing process in the physical photovoltaic module production line, mapping the quality performance evaluation results in the low-speed reaction process model unit to the quality inspection indicators of the lamination process and the encapsulation process in the physical photovoltaic module production line, and mapping the process connection relationship reconstructed in the virtual space through the spatiotemporal connection constraints to the process transition state between the stringing process and the lamination process, and the process transition state between the lamination process and the encapsulation process in the physical photovoltaic module production line.

[0051] Specifically, the equipment control instructions of the intelligent quality control module include string soldering process control instructions generated based on the deviation between the real-time operating status of the string soldering process mapped in the virtual mapping relationship and the preset standard operating status, dynamically adjusting the current pulse waveform parameters of the Hall current sensor; lamination process control instructions generated based on the deviation between the quality inspection index of the lamination process mapped in the virtual mapping relationship and the preset standard quality index, dynamically correcting the temperature gradient setpoint of the high-temperature capacitive sensor and the chamber pressure threshold of the vacuum pressure transmitter; and packaging process control instructions generated based on the deviation between the quality inspection index of the packaging process mapped in the virtual mapping relationship and the preset standard quality index, dynamically calibrating the displacement trajectory control parameters of the photoelectric encoder.

[0052] This embodiment uses a virtual mapping relationship to accurately map the simulation results of the dynamic behavior of the high-speed reaction process and the quality performance evaluation results of the low-speed reaction process to the real-time operating status of the stringing process, the quality inspection indicators of the lamination and packaging processes, and the process transition status between processes on the physical production line. Based on the deviation, it generates control commands to dynamically adjust the current pulse waveform parameters of the Hall current sensor, the temperature gradient setpoint of the high-temperature capacitive sensor, the chamber pressure threshold of the vacuum pressure transmitter, and the displacement trajectory control parameters of the photoelectric encoder, thereby achieving collaborative optimization of the entire process chain and significantly improving the accuracy of quality monitoring and the speed of abnormal response.

[0053] Specifically, the physical device execution module includes a serial soldering process execution unit, a lamination process execution unit, and a packaging process execution unit.

[0054] Preferably, the real-time feedback module specifically includes: capturing the equipment operating parameters output by the physical device execution module through a high-speed data acquisition interface. The equipment operating parameters include the adjusted Hall current sensor current pulse waveform parameters, the corrected temperature gradient setpoint of the high-temperature capacitive sensor, the updated chamber pressure threshold of the vacuum pressure transmitter, and the calibrated displacement trajectory control parameters of the photoelectric encoder. The parameters are classified and reorganized according to the process type (series soldering / lamination / packaging), the reorganized data stream is converted into a process feature value format, and transmitted to the process feature extraction unit of the dynamic monitoring interval division module through the feedback bus. This drives the iterative update of the real-time monitoring interval and the adaptive optimization of the dynamic monitoring time interval, forming a closed-loop control loop.

[0055] The above are merely embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A photovoltaic module production quality monitoring system based on digital twins, characterized in that: It includes a dynamic monitoring interval division module, a multi-source data fusion module, a digital twin virtual mapping module, an intelligent quality control module, a physical equipment execution module, and a real-time feedback module; The dynamic monitoring interval division module is used to generate dynamic monitoring time intervals in real time according to each process stage in the photovoltaic module production line. The multi-source data fusion module is used to collect equipment operation data and quality inspection data of each process stage within the dynamic monitoring time interval, and to generate fused data after processing the equipment operation data and the quality inspection data. The digital twin virtual mapping module constructs a virtual model of the photovoltaic production line based on the fused data and outputs virtual mapping relationships; The intelligent quality control module generates equipment control commands based on the virtual mapping relationship; The physical device execution module executes the device control commands and outputs the device operating parameters; The real-time feedback module collects the operating parameters of the equipment and feeds them back to the dynamic monitoring interval division module.

2. The photovoltaic module production quality monitoring system based on digital twin according to claim 1, characterized in that: The photovoltaic production line includes string bonding, lamination, and encapsulation processes. Real-time sensor data streams are acquired by deploying a sensor network at each process stage. These real-time sensor data streams include: in the string bonding process, Hall current sensors are used to collect welding current pulse waveforms, and infrared thermal imaging devices are used to capture the temperature field distribution of the solder joints; in the lamination process, high-temperature capacitive sensors are used to acquire the temperature gradient of the laminate in real time, and vacuum pressure transmitters are used to monitor the pressure change curve of the chamber; and in the encapsulation process, photoelectric encoders are used to record the frame assembly displacement trajectory.

3. The photovoltaic module production quality monitoring system based on digital twin according to claim 2, characterized in that: The steps for generating the dynamic monitoring time interval in the dynamic monitoring interval division module are as follows: S1. Obtain the real-time operating characteristics of each process stage through the preset process feature extraction unit, and generate process stage identifiers; S2. Call the preset rate matching unit according to the process stage identifier, and match the baseline reaction rate parameter of each process stage from the preset process knowledge base. S3. Based on the benchmark reaction rate parameter and the real-time sensor data stream, the real-time monitoring interval is calculated, and a dynamic monitoring time interval is generated according to the real-time monitoring interval. The dynamic monitoring time interval includes a millisecond-level monitoring interval and a second-level monitoring interval. The millisecond-level monitoring interval corresponds to the high-speed reaction process, and the second-level monitoring interval corresponds to the low-speed reaction process.

4. The photovoltaic module production quality monitoring system based on digital twin according to claim 3, characterized in that: The specific content of generating the dynamic monitoring time interval based on the real-time monitoring interval in step S3 is as follows: S31. Perform continuous preset time period scanning on the real-time monitoring interval of each process stage, extract the real-time monitoring intervals below the first preset threshold to form a first dataset, and assign a unique sub-index to each real-time monitoring interval in the first dataset. S32. The real-time monitoring intervals that are higher than the second preset threshold within a continuous preset time period are sorted and reorganized according to the timestamp sequence to generate multiple second datasets divided according to a preset period, and the missing positions in each second dataset are marked with a main index identifier. S33. Map the preset time period corresponding to the first dataset to a millisecond-level monitoring interval, and at the same time map the preset period of the second dataset to a second-level monitoring interval. According to the spatial correspondence between the sub-index and the main index, embed the millisecond-level monitoring interval into the empty position of the second-level monitoring interval. S34. Set a buffer zone at the beginning of the millisecond-level monitoring interval, and connect the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer zone to generate a dynamic monitoring time interval.

5. A photovoltaic module production quality monitoring system based on digital twins according to claim 4, characterized in that: The specific method for connecting the millisecond-level monitoring interval and the second-level monitoring interval through the transition unit in the buffer zone in step S34 is as follows: S341. Multiple preset transition point data are continuously distributed in the buffer zone, and the first and last transition point data in the buffer zone are respectively corrected to the real-time monitoring interval at the beginning of the millisecond-level monitoring interval and the real-time monitoring interval at the beginning of the second-level monitoring interval. S342. According to the preset interpolation algorithm in the transition unit, based on the corrected first transition point data and the last transition point data, interpolation calculation is performed on the remaining transition point data in the buffer to generate a transition point data sequence, and the millisecond-level monitoring interval and the second-level monitoring interval are connected through the transition point data sequence.

6. A photovoltaic module production quality monitoring system based on digital twins according to claim 5, characterized in that: The device operation data in the multi-source data fusion module is the real-time sensor data stream. The quality inspection data is collected through a pre-set online optical inspection system, specifically including: deploying a machine vision device to extract the solder strip alignment deviation vector in the subsequent process of the stringing process; setting a laser triangulation instrument to scan the EVA film thickness distribution at the exit end of the lamination process; and using a spectrophotometer to measure the transmittance attenuation coefficient of the photovoltaic module in the encapsulation process. The device operation data and the quality inspection data are timestamped and spatially registered through a preset edge computing gateway, and then windowed and recombined according to the dynamic monitoring time interval to generate fused data.

7. A photovoltaic module production quality monitoring system based on digital twins according to claim 6, characterized in that: The construction process of the photovoltaic production line virtual model in the digital twin virtual mapping module is as follows: Q1. The fused data generated by windowed slicing and recombining is converted into a virtual entity attribute set, wherein the device operation data stream is mapped to a virtual device dynamic parameter cluster, the quality inspection data stream is mapped to a virtual quality index matrix, and the time-series coordinate axis is divided based on the dynamic monitoring time interval; Q2. Using the millisecond-level monitoring interval and the second-level monitoring interval on the time-series coordinate axis as time reference templates, high-speed reaction process model units and low-speed reaction process model units are synchronously generated in a preset virtual space. The virtual equipment dynamic parameter cluster drives the dynamic behavior simulation of the high-speed reaction process model unit, and the virtual quality index matrix is ​​injected into the quality performance evaluation of the low-speed reaction process model unit. Simultaneously, the transition point data sequence in the buffer zone is converted into spatiotemporal connection constraints. The high-speed reaction process model unit and the low-speed reaction process model unit are connected through the spatiotemporal connection constraints to generate a photovoltaic production line virtual model.

8. A photovoltaic module production quality monitoring system based on digital twins according to claim 7, characterized in that: The virtual mapping relationship in the intelligent quality control module includes mapping the dynamic behavior simulation results in the high-speed reaction process model unit to the real-time operating status of the stringing process in the physical photovoltaic module production line, mapping the quality performance evaluation results in the low-speed reaction process model unit to the quality inspection indicators of the lamination and encapsulation processes in the physical photovoltaic module production line, and mapping the process connection relationship reconstructed in the virtual space through the spatiotemporal connection constraints to the process transition state between the stringing process and the lamination process, and the process transition state between the lamination process and the encapsulation process in the physical photovoltaic module production line.

9. A photovoltaic module production quality monitoring system based on digital twins according to claim 8, characterized in that: The equipment control commands of the intelligent quality control module include: a string soldering process control command generated based on the deviation between the real-time operating status of the string soldering process mapped in the virtual mapping relationship and the preset standard operating status; a lamination process control command generated based on the deviation between the quality inspection index of the lamination process mapped in the virtual mapping relationship and the preset standard quality index; and a packaging process control command generated based on the deviation between the quality inspection index of the packaging process mapped in the virtual mapping relationship and the preset standard quality index.

10. A photovoltaic module production quality monitoring system based on digital twins according to claim 9, characterized in that: The physical device execution module includes a string soldering process execution unit, a lamination process execution unit, and a packaging process execution unit.

Images