Hyperspectral online monitoring and intelligent control method and system for flexible OPV roll-to-roll production

Abstract

This application relates to the field of flexible organic solar cell technology. Specifically, it discloses a hyperspectral online monitoring and intelligent control method and system for roll-to-roll production of flexible OPVs. This method integrates snapshot hyperspectral imaging, drying speed quantification, and dual machine learning models to achieve intelligent control of the drying process during roll-to-roll production of flexible OPVs. Single-frame hyperspectral images are used to synchronously capture the wet / dry film region, locate the wet-dry transition boundary, and dynamically calculate the average drying time based on substrate speed. Hyperspectral image features are used to invert various thin film characteristic parameters. A dual database of "process parameters, image features, and device performance" is constructed, along with performance prediction and process inversion models, forming a "monitoring-prediction-control" closed-loop control system.

Description

Technical Field

[0001] The present application belongs to the technical field of flexible organic solar cells, and more specifically, relates to a method and system for hyperspectral online monitoring and intelligent control of flexible OPV roll-to-roll production. Background Art

[0002] Organic photovoltaics (OPVs) utilize organic semiconductor materials to achieve photoelectric conversion. Flexible OPVs, with their significant advantages such as lightweight, low cost, printability, and high flex resistance, are considered a promising future for photovoltaic commercialization and are expected to form a complementary market with rigid silicon solar cells. Over the past few years, innovations in organic photovoltaic materials and device structures have led to laboratory efficiencies exceeding 20% ​​for OPVs prepared on rigid substrates via spin coating. However, due to differences between industrial production and laboratory processes, device performance often degrades significantly during large-scale printing. Even with the best-performing material systems, large-scale fabrication using existing printing processes still suffers from significant efficiency losses. Furthermore, the lack of effective monitoring and optimization methods for large-scale film formation limits the commercialization of OPVs. As flexible OPVs move from the laboratory to industrialization, online monitoring and intelligent control of the roll-to-roll coating and printing process remain pressing technical challenges. Currently, the morphological evolution and phase separation processes during organic thin film preparation significantly impact the final device performance, but effective in situ characterization of these multi-dimensional properties, encompassing both spatial (thickness) and temporal (drying) characteristics, is lacking. Traditional processes typically rely on offline sampling and testing, such as post-coating sampling for thickness measurement, component analysis, and device performance testing. This post-analysis cannot meet the real-time quality control requirements of high-speed roll-to-roll production.

[0003] In recent years, both domestic and international research on online monitoring technologies for OPV manufacturing has begun, but overall, they are still in their infancy and lack mature and reliable solutions. Existing online monitoring equipment mostly focuses on a single parameter, such as using optical sensors to monitor film thickness. Currently, there is a lack of specialized software systems for flexible OPV manufacturing processes, making it impossible to directly correlate online monitoring data with device performance for process control and performance prediction. Adjustments to process parameters during production are primarily based on experience, lacking the support of data-driven intelligent optimization tools.

[0004] Hyperspectral imaging technology integrates the capabilities of spectral analysis and imaging measurement, demonstrating unique advantages in the field of materials testing. On the one hand, hyperspectral imaging captures a continuous spectrum at each pixel, revealing information such as the material's chemical composition and thickness. On the other hand, it also provides spatial resolution, revealing the distribution of properties across regions within a film. Existing technologies propose using visible-near-infrared hyperspectral imaging for online monitoring of polymer film production, extracting features highly correlated with the film's composition distribution and mechanical properties through multi-resolution multivariate image analysis. This suggests that hyperspectral imaging has the potential to be used for nondestructive testing of film composition and quality, as well as for predicting performance. However, in the field of organic photovoltaic device manufacturing, there are currently no public reports combining hyperspectral image analysis with process control and performance prediction.

[0005] In summary, the large-scale printing and manufacturing of flexible OPVs requires new online monitoring and intelligent control methods. To address the problems of existing technologies, such as the single monitoring dimension, the lack of correlation between online detection data and device performance, and the inability to achieve real-time performance prediction and process adjustment, there is an urgent need to develop an intelligent process control system that integrates hyperspectral image analysis and machine learning prediction to achieve real-time quality monitoring of the flexible OPV film formation process and instant prediction of device performance, thereby improving production consistency and efficiency. Summary of the Invention

[0006] In order to solve the above technical problems, the present application is proposed. The present application provides a method and system for hyperspectral online monitoring and intelligent control of flexible OPV roll-to-roll production, which integrates snapshot hyperspectral imaging, drying speed quantification and dual machine learning models to realize intelligent control of the film drying process of flexible OPV roll-to-roll production. The wet film / dry film area is captured synchronously by a single-frame hyperspectral image, the wet-dry transition boundary is located, and the average drying time is dynamically calculated in combination with the substrate speed; the hyperspectral image features are used to realize the extraction of multiple thin film characteristic parameters; a dual database of "process parameters-image features-device performance" is constructed, and a performance prediction model and a process inversion model are established to form a "monitoring-prediction-control" closed-loop control system. The technical solution of this application is as follows:

[0007] According to one aspect of the present application, a method for hyperspectral online monitoring and intelligent control of flexible OPV roll-to-roll production is provided, including a training phase and a real-time monitoring phase:

[0008] The training phase includes:

[0009] Acquire a sample data set, wherein the sample data set includes multiple groups of sample data, each group of sample data includes process parameters, corresponding hyperspectral image features, and corresponding device performance parameters; the process parameters include coating speed, distance between coating head and substrate, solvent type, solution concentration, ink injection speed, and substrate temperature; the hyperspectral image features include spectral curve, average drying time, film thickness parameters, and donor-acceptor ratio parameters; the device performance parameters include photoelectric conversion efficiency PCE, fill factor FF, and open circuit voltage V OC , short-circuit current J SC ;

[0010] Establish mapping relationships between process parameters and hyperspectral image features, and between hyperspectral image features and device performance parameters, and construct dual databases, including: process parameter-image feature database, and image feature-performance parameter database;

[0011] Training process inversion model and performance prediction model based on dual databases;

[0012] The real-time monitoring stage includes:

[0013] Acquiring a hyperspectral image of the flexible OPV film during a drying process, wherein the hyperspectral image includes both a wet film area and a dry film area;

[0014] According to the characteristic absorption peaks of the donor and acceptor materials, a specific band is selected to extract the spectral curve of each pixel of the hyperspectral image;

[0015] Identify the wet-dry transition boundary based on the spectral curve and calculate the average drying time;

[0016] Calculate film thickness parameters and donor-acceptor ratio parameters based on spectral curves;

[0017] Based on the real-time hyperspectral image features, the performance prediction model is called to predict the device performance parameters. If the performance parameters deviate from the target values, the process inversion model is called to output the adjusted process parameters and control signals.

[0018] Furthermore, in the online monitoring and intelligent control method of the present application, the film thickness parameters include: thickness distribution, thickness average, and thickness variance.

[0019] Furthermore, in the online monitoring and intelligent control method of the present application, the donor-acceptor ratio parameters include ratio distribution, ratio mean, and ratio variance.

[0020] Furthermore, in the online monitoring and intelligent control method of the present application, the dual-database-based training of the process inversion model and the performance prediction model includes: training the process inversion model based on the process parameter-image feature database, and training the performance prediction model based on the image feature-performance parameter database.

[0021] Furthermore, in the online monitoring and intelligent control method of the present application, the calculation of film thickness parameters and donor-acceptor ratio parameters based on the spectral curve includes: using the Beer-Lambert law to calculate the film thickness; and using the spectral unmixing algorithm to calculate the donor-acceptor ratio.

[0022] Furthermore, in the online monitoring and intelligent control method of the present application, the calculation formula for the average drying time is as follows:

[0023]

[0024] where d is the distance between the wet-dry transition boundary and the coating head, and v is the substrate speed.

[0025] Furthermore, in the online monitoring and intelligent control method of the present application, if the PCE deviates from the target value by ±5%, the process inversion model is called to output the adjusted process parameters and control signals.

[0026] According to another aspect of the present application, a flexible OPV roll-to-roll production hyperspectral online monitoring and intelligent control system is provided, comprising:

[0027] Data acquisition module: used to obtain sample data sets, which contain multiple sets of sample data. Each set of sample data includes process parameters, corresponding hyperspectral image features, and corresponding device performance parameters. The process parameters include coating speed, distance between coating head and substrate, solvent type, solution concentration, ink injection speed, and substrate temperature. The hyperspectral image features include spectral curve, average drying time, film thickness parameters, and donor-acceptor ratio parameters. The device performance parameters include photoelectric conversion efficiency PCE, fill factor FF, and open circuit voltage. V OC , short-circuit current J SC ;

[0028] Dual database module: used to establish mapping relationships between process parameters and hyperspectral image features, and between hyperspectral image features and device performance parameters, and to build dual databases, including: process parameter-image feature database, and image feature-performance parameter database;

[0029] Model training module: training process inversion model and performance prediction model based on dual databases;

[0030] Snapshot hyperspectral imaging device: used to obtain hyperspectral images of the flexible OPV film during the drying process, wherein the hyperspectral images include both the wet film area and the dry film area;

[0031] Data processing module: used to select specific bands based on the characteristic absorption peaks of donor and acceptor materials and calculate the spectral curve of the hyperspectral image at each pixel point;

[0032] Dynamic analysis module: used to identify the wet-dry transition boundary based on the spectral curve and calculate the average drying time;

[0033] Parameter extraction module: used to calculate film thickness parameters and donor-acceptor ratio parameters based on spectral curves;

[0034] Process control module: used to call the performance prediction model to predict device performance parameters based on the real-time hyperspectral image features. If the performance parameters deviate from the target values, the process inversion model is called to output the adjusted process parameters and control signals.

[0035] Furthermore, in the online monitoring and intelligent control system of the present application, the snapshot hyperspectral imaging device has an operating band of 450–950 nm, a spatial resolution of ≥50 μm, and an exposure time of ≥50 ms.

[0036] Furthermore, in the online monitoring and intelligent control system of the present application, the execution modes of the process control module include:

[0037] Semi-automatic mode: The adjusted process parameters are displayed on the interface, and manual confirmation is required to determine whether to issue a control signal;

[0038] Fully automatic mode: directly send control signals and execute the adjusted process parameters.

[0039] The beneficial effects of the technical solution of this application are as follows:

[0040] 1. Realize multi-dimensional monitoring: Simultaneously obtain multiple parameters such as film thickness and thickness uniformity, donor-acceptor ratio and uniformity, average drying time, etc., to comprehensively quantify film quality;

[0041] 2. Real-time prediction of performance parameters: By combining image features with a device performance database, machine learning models are used to achieve real-time performance prediction during the coating process, replacing offline sampling.

[0042] 3. Realize feedback adjustment of process parameters: When the predicted PCE deviates from the target value, the process parameters are automatically optimized by associating them with the image database. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0044] Figure 1 Schematic diagram of the deployment of a hyperspectral online monitoring and intelligent control system for flexible OPV roll-to-roll production according to an embodiment of the present application.

[0045] Figure 2 This is a flow chart of the hyperspectral online monitoring and intelligent control method for flexible OPV roll-to-roll production according to an embodiment of the present application.

[0046] Figure 3 Schematic diagram of the human-machine interface of the hyperspectral online monitoring and intelligent control method for flexible OPV roll-to-roll production according to an embodiment of the present application.

[0047] Figure 4 is a spectral absorption curve diagram of a pixel point in a hyperspectral image according to an embodiment of the present application.

[0048] Figure 5 This is a diagram of the model training results according to an embodiment of the present application. DETAILED DESCRIPTION

[0049] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0050] Example 1

[0051] Figure 1 A deployment method for the flexible OPV roll-to-roll production hyperspectral online monitoring and intelligent control system of this application is given, such as Figure 1As shown in the figure, a snapshot hyperspectral imaging device (operating in the visible-to-near-infrared range, with a resolution of 50 μm and an exposure time of 50 ms) is deployed downstream of the coating head and along the substrate's moving direction on a flexible OPV roll-to-roll coating line. The device captures the spectral characteristics of organic thin films in a single frame as the coated wet film moves with the substrate. By properly configuring the camera's exposure and acquisition speed, the image clearly captures the differences between the wet and dry film regions. The hyperspectral camera preferably uses an area array snapshot imaging method, acquiring a spectral image of the entire field of view in a single exposure, thus avoiding image distortion caused by motion. The resulting hyperspectral image is fed into the subsequent dynamic analysis, parameter extraction, and process control modules. Based on real-time image features, the device's characteristic parameters are extracted and device performance parameters are predicted. If the performance parameters deviate from the target values, the process parameters are adjusted and a control signal is output. The human-machine interface is used to display results such as film thickness heat maps and donor-acceptor ratio heat maps, allowing the operator to intuitively understand the current quality status of the coating. It also displays performance prediction results and adjusted process parameters. The operator can manually determine whether to issue a control signal (semi-automatic mode) or issue a control signal and execute the adjusted process parameters (fully automatic mode).

[0052] Figure 2 A flow chart of the hyperspectral online monitoring and intelligent control method for flexible OPV roll-to-roll production in this application is given. Figure 3 The hyperspectral image at a certain moment and its corresponding analysis results are given. The specific implementation process includes:

[0053] 1) Spectral curve extraction: The data processing module selects specific bands based on the characteristic absorption peaks of the donor and acceptor materials and calculates the spectral curve of each pixel of the hyperspectral image;

[0054] 2) Identifying wet-dry transition boundaries and calculating average drying time: The dynamic analysis module automatically identifies wet-dry transition boundaries in hyperspectral images. Because wet films contain unvolatile solvents, the absorption spectra of the donor and acceptor materials differ significantly from those of dry films. This is manifested by shifted absorption peaks, lower overall absorption values, and altered spectral shapes. This module achieves pixel-level classification using any of the following methods:

[0055] Spectral threshold method: For example, the absorption peak of the classic receptor material L8BO in solution is at 710nm. After drying, the absorption peak of the film will undergo a significant red shift to 790nm. The absorption intensity ratio of the characteristic bands 790nm and 710nm can be compared. If it is lower than 0.1, it is a wet film, higher than 1, it is a dry film, and between 0.1 and 1 is the intermediate state of the transition from wet film to dry film;

[0056] Machine learning segmentation method: train a classification model (such as SVM) to determine the wet / dry state based on the full-band spectral curve.

[0057] The recognition results generate binary wet / dry mask images (such as Figure 3 Wet / dry film analysis in ), where the wet film area is marked as 0 (black) and the dry film area is marked as 1 (white).

[0058] The dynamic analysis module determines the location of the wet-dry transition boundary. The boundary between the wet and dry films represents the film drying "front." Its downstream boundary (the region immediately adjacent to the dry film along the substrate's direction of motion) is defined as the drying front position d. This position corresponds to the critical point where post-coating drying is complete in roll-to-roll production. The average drying time t is calculated as follows:

[0059]

[0060] Where v is the real-time speed of the substrate conveyor (unit: m / s); d is the distance from the drying front to the coating head (converted to physical length from the image pixel coordinates). The camera is installed in a fixed position and the distance from the coating head is a constant L.

[0061] 3) Calculation of film thickness parameters and donor-acceptor ratio parameters: The parameter extraction module quantitatively analyzes the spectrum of each pixel in the hyperspectral image to invert the characteristic parameters of the film.

[0062] First, the Beer-Lambert law or machine learning spectrum fitting method is used to calculate the film thickness. For semi-transparent organic films, the absorption intensity of their hyperspectral absorption spectrum is directly related to the thickness. The specific principle is as follows:

[0063] Beer-Lambert Law: Absorption peak intensity at a specific wavelength Calculate thickness :

[0064]

[0065] in is the bulk absorption coefficient of the material.

[0066] The above method relies on a pre-built thickness-spectrum mapping model, which is obtained through experimental calibration:

[0067] 1. Prepare a set of thickness gradient film samples (e.g. 80 nm, 100 nm, 120 nm);

[0068] 2. Measure the hyperspectral absorption curve of the sample;

[0069] 3. Establish a quantitative relationship between spectral characteristics (such as the absorption intensity of the classic receptor material L8BO absorption peak at 790 nm) and thickness, which can be presented through a fitting curve.

[0070] Machine learning fitting: Prepare a series of thin film samples of known thickness, measure their hyperspectral absorption spectra, and use regression models (such as neural networks) for machine learning to establish a relationship between spectral characteristics (such as spectral curve shape and absorption intensity at specific wavelengths) and film thickness. When used, the thickness value is directly output based on the measured hyperspectral absorption curve.

[0071] Secondly, the donor-acceptor material ratio (i.e., the concentration distribution of the active layer components) is calculated based on the characteristic differences between the two materials in the spectrum. A typical organic photovoltaic active layer is composed of an electron donor material (such as a polymer donor PM6) and an electron acceptor material (such as a non-fullerene acceptor L8BO). Both have their own characteristic absorption peaks in the visible-near infrared band: for example, in the film, the absorption peak of the donor PM6 is around 614 nm, and the absorption peak of the acceptor L8BO is around 790 nm. By analyzing the spectral curve of each pixel in the hyperspectral image (such as Figure 4 As shown in the figure, calculate its fit with the pure donor and pure acceptor reference spectra, and then calculate the component ratio at that position. The specific implementation includes two methods:

[0072] 1. Linear spectral unmixing algorithm:

[0073] Assuming pixel spectrum Pure donor spectrum Pure receptor spectrum A linear combination of:

[0074]

[0075] Solving for coefficients using the least squares method and , then the donor-acceptor ratio = / .

[0076] 2. Machine Learning Mapping Method:

[0077] Based on training samples with known ratios (such as donor:acceptor = 1:0.5, 1:1.2, 1:1.5, etc.), the pixel spectra are input into a regression model (such as partial least squares PLS or neural network) and the ratio value is directly output.

[0078] After the above inversion calculation, the full-width hyperspectral image can be converted into a spatial distribution map of the donor-acceptor ratio (e.g. Figure 2 The component distribution in the matrix directly reflects the component uniformity and local deviation (such as low edge ratio).

[0079] In addition to thickness and composition ratio, this module can also extract other film quality indicators from hyperspectral data, such as thickness uniformity parameters, surface roughness indication (through scattered light intensity), etc. These parameters comprehensively reflect the film quality. Ultimately, the system will display the film thickness distribution map, composition distribution map and other results in the human-computer interface, such as Figure 3 The dry / wet film analysis results shown are an average drying time of 5.2 ± 1.2 s, an average wet film thickness of 20 ± 2 μm, an average dry film thickness of 105 ± 5 nm, a wet film donor-acceptor ratio (D / A) of 1:1.2 ± 0.05, and a dry film donor-acceptor ratio (D / A) of 1:1.2 ± 0.02. Performance predictions under current conditions are: V OC 14.1V, J SC 1.40mAcm 2 , FF 0.700, PCE 13.8%. Recommended optimization conditions: increase the distance between the coating head and the substrate, or increase the coating speed. 4) Dual database construction: During the production line commissioning phase, a mapping relationship is established through gradient experimental design, specifically including:

[0080] Experimental design: Thin film samples were prepared under orthogonal process parameter combinations (coating speed, slit height between the coating head and the substrate, solvent, solution concentration, ink injection speed, and substrate temperature).

[0081] Data collection:

[0082] 1. Record process parameters (coating speed, slit height between coating head and substrate, solvent, solution concentration, ink filling speed, substrate temperature, etc.);

[0083] 2. Hyperspectral analysis to extract features (average film thickness, thickness uniformity, donor-acceptor ratio, average drying time t);

[0084] 3. Device performance test (photoelectric conversion efficiency PCE, open circuit voltage V OC , short-circuit current J SC , filling factor FF, etc.);

[0085] Database structure:

[0086] {printing process parameters} → {hyperspectral image features};

[0087] {Hyperspectral image characteristics} → {Device performance parameters}.

[0088] 5) Model training: The model training module trains performance prediction models and process inversion models based on dual databases. Figure 5 The model training results are shown;

[0089] 1. Performance prediction model architecture:

[0090] Input layer: image feature vector;

[0091] Algorithm: Random forest regression + neural network assisted nonlinear fitting;

[0092] Output layer: PCE / V OC / J SC / FF predicted value.

[0093] 2. Process inversion model architecture:

[0094] Input layer: image feature vector;

[0095] Algorithms: Machine learning regression or optimization algorithms;

[0096] Output layer: adjusted process parameters.

[0097] 3. Training optimization:

[0098] Feature screening: Decision trees identify key features;

[0099] Anti-overfitting: k-fold cross validation, limiting tree depth to ≤10;

[0100] Accuracy: Test set R² ≥ 0.85, PCE prediction error ≤ ± 2%.

[0101] 6) Performance prediction and process control: The performance prediction model is used to predict device performance parameters. If the performance parameters deviate from the target values, the process inversion model is used to output the adjusted process parameters and control signals, including:

[0102] 1. Online prediction:

[0103] Real-time input of hyperspectral features → output of performance values ​​and 95% confidence intervals;

[0104] If the predicted PCE is lower than the target value by ±5%, an alarm is triggered.

[0105] 2. Parameter adjustment:

[0106] According to the target performance, the process parameter adjustment amount is calculated through the optimization algorithm.

[0107] Method: Solving constraint satisfaction problems (such as particle swarm optimization), minimizing the objective function ;

[0108] Output: coating speed adjustment (±20%), temperature correction value (±15°C).

[0109] As production continues, validation test results from new coating batches can be continuously added to the database, allowing the machine learning model to be updated regularly or in real time to maintain its adaptability to changes in material systems or processes. This self-learning capability makes the system increasingly intelligent, providing reliable performance prediction support for OPV production under different material formulations and environmental conditions.

[0110] The software system of the embodiment of the present application presents the analysis and prediction results in real time through the human-machine interface and realizes process feedback control. Figure 3 The interface layout shown includes:

[0111] Visualization display area: Hyperspectral image of the current batch (pseudo-color rendering); two-dimensional distribution diagram of wet / dry film distribution, film thickness and component ratio; key parameter values ​​(average thickness, drying time, predicted PCE, etc.) and historical trend curve.

[0112] Early warning and decision-making area: When the predicted performance (such as PCE) deviates from the preset target, a warning sign is highlighted; based on the process parameter-image feature library and machine learning model, an optimization plan is intelligently generated: if the drying time is too long, it is recommended to "increase the substrate temperature" or "reduce the ink injection speed"; if the film thickness is too low, it is recommended to "reduce the substrate movement speed."

[0113] Control execution mode: Semi-automatic mode displays the adjusted process parameters on the interface, and manual determination is made whether to issue a control signal; fully automatic mode directly sends a control signal to the production equipment to execute the adjusted process parameters.

[0114] The interface also provides historical data query and reporting functions. The historical data query function allows users to view film parameters and predicted performance trends for each batch (e.g., film thickness deviation within an hour, drying time variation). The report generation function outputs production quality reports (average performance predictions, abnormality records, and process adjustment measures).

[0115] The software system of this embodiment is adaptable to a variety of equipment environments. It can be run on on-site industrial computers / process control servers, or it can upload key data to the cloud for remote monitoring and big data analysis. Its modular software architecture allows for excellent scalability, allowing for the addition of analysis modules tailored to different material systems or emerging defect detection requirements.

[0116] In summary, the specific implementation method of the present application realizes closed-loop control of the entire production process of flexible organic solar cells through the collaboration of hardware acquisition and software analysis, including image acquisition, parameter extraction, performance prediction and feedback control. The system significantly improves the visualization and control accuracy of the production process, effectively enhances the performance consistency and production efficiency of large-area flexible OPV modules, and accelerates the transformation of laboratory technology into industrialization.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for online hyperspectral monitoring and intelligent control of flexible OPV roll-to-roll production, characterized in that: It includes training phase and real-time monitoring phase; The training phase includes: Acquire a sample data set, wherein the sample data set includes multiple groups of sample data, each group of sample data includes process parameters, corresponding hyperspectral image features, and corresponding device performance parameters; the process parameters include coating speed, distance between coating head and substrate, solvent type, solution concentration, ink injection speed, and substrate temperature; the hyperspectral image features include spectral curve, average drying time, film thickness parameters, and donor-acceptor ratio parameters; the device performance parameters include photoelectric conversion efficiency PCE, fill factor FF, and open circuit voltage V OC , short-circuit current J SC ; Establish mapping relationships between process parameters and hyperspectral image features, and between hyperspectral image features and device performance parameters, and construct dual databases, including: process parameter-image feature database, and image feature-performance parameter database; Training process inversion model and performance prediction model based on dual databases; The real-time monitoring stage includes: Acquiring a hyperspectral image of the flexible OPV film during a drying process, wherein the hyperspectral image includes both a wet film area and a dry film area; According to the characteristic absorption peaks of the donor and acceptor materials, a specific band is selected to extract the spectral curve of each pixel of the hyperspectral image; Identify the wet-dry transition boundary based on the spectral curve and calculate the average drying time; Calculate film thickness parameters and donor-acceptor ratio parameters based on spectral curves; Based on the real-time hyperspectral image features, the performance prediction model is called to predict the device performance parameters. If the performance parameters deviate from the target values, the process inversion model is called to output the adjusted process parameters and control signals.

2. The method according to claim 1, characterized in that The film thickness parameters include: thickness distribution, thickness average, and thickness variance.

3. The method according to claim 1, characterized in that The donor-acceptor ratio parameters include ratio distribution, ratio mean, and ratio variance.

4. The method according to claim 1, wherein The dual-database-based training of the process inversion model and the performance prediction model includes: training the process inversion model based on a process parameter-image feature database, and training the performance prediction model based on an image feature-performance parameter database.

5. The method according to claim 1, wherein The method of calculating the film thickness parameter and the donor-acceptor ratio parameter based on the spectral curve includes: calculating the film thickness using the Beer-Lambert law; and calculating the donor-acceptor ratio using a spectral unmixing algorithm.

6. The method according to claim 1, characterized in that The average drying time is calculated as follows: where d is the distance between the wet-dry transition boundary and the coating head, and v is the substrate speed.

7. The method according to claim 1 or 3, characterized in that If the PCE deviates from the target value by ±5%, the process inversion model is called to output the adjusted process parameters and control signals.

8. A flexible OPV roll-to-roll production hyperspectral online monitoring and intelligent control system, characterized by: include: Data acquisition module: used to obtain sample data sets, which contain multiple sets of sample data. Each set of sample data includes process parameters, corresponding hyperspectral image features, and corresponding device performance parameters. The process parameters include coating speed, distance between coating head and substrate, solvent type, solution concentration, ink injection speed, and substrate temperature. The hyperspectral image features include spectral curve, average drying time, film thickness parameters, and donor-acceptor ratio parameters. The device performance parameters include photoelectric conversion efficiency PCE, fill factor FF, and open circuit voltage. V OC , short-circuit current J SC ; Dual database module: used to establish mapping relationships between process parameters and hyperspectral image features, and between hyperspectral image features and device performance parameters, and to build dual databases, including: process parameter-image feature database, and image feature-performance parameter database; Model training module: training process inversion model and performance prediction model based on dual databases; Snapshot hyperspectral imaging device: used to obtain hyperspectral images of the flexible OPV film during the drying process, wherein the hyperspectral images include both the wet film area and the dry film area; Data processing module: used to select specific bands based on the characteristic absorption peaks of donor and acceptor materials and calculate the spectral curve of the hyperspectral image at each pixel point; Dynamic analysis module: used to identify the wet-dry transition boundary based on the spectral curve and calculate the average drying time; Parameter extraction module: used to calculate film thickness parameters and donor-acceptor ratio parameters based on spectral curves; Process control module: used to call the performance prediction model to predict device performance parameters based on the real-time hyperspectral image features. If the performance parameters deviate from the target values, the process inversion model is called to output the adjusted process parameters and control signals.

9. The system according to claim 8, characterized in that The snapshot hyperspectral imaging device has an operating band of 450–950 nm, a spatial resolution of ≥50 μm, and an exposure time of ≥50 ms.

10. The system according to claim 8, wherein: The execution modes of the process control module include: Semi-automatic mode: The adjusted process parameters are displayed on the interface, and manual confirmation is required to determine whether to issue a control signal; Fully automatic mode: Directly send control signals to execute the adjusted process parameters.

Images