Photovoltaic module defect identification and capacity detection system based on unmanned aerial vehicle inspection

Abstract

The present application belongs to the technical field of unmanned aerial vehicle autonomous inspection, and particularly relates to a photovoltaic module defect identification and capacity detection system based on unmanned aerial vehicle inspection. The adaptive module is used for sensor calibration and space-time data synchronization processing before inspection; the unmanned aerial vehicle path planning module is used for planning multiple unmanned aerial vehicle inspection paths based on photovoltaic power station array topology structure; the unmanned aerial vehicle inspection module is used for starting unmanned aerial vehicle cluster cooperative inspection and dynamically adjusting load and flight attitude; the light supplementing module is used for spectrum supplementing and image acquisition optimization under complex illumination environment; the edge identification module is used for photovoltaic module defect identification processing based on the optimized image at the edge of the unmanned aerial vehicle; and the capacity checking module is used for counting the actual number of photovoltaic modules in the inspection area based on the photovoltaic module image, and capacity calibration is performed in the dark.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) autonomous inspection technology, specifically relating to a photovoltaic module defect identification and capacity detection system based on UAV inspection. Background Technology

[0002] Drone inspection refers to a technical means of using drone platforms with autonomous flight or remote control capabilities, equipped with suitable sensing and data acquisition devices, to conduct aerial inspections and data collection of facilities and equipment in a target area according to a pre-planned path or a flight trajectory adjusted in real time. In the photovoltaic power station scenario, its core is to use drones to traverse the photovoltaic module array and obtain key data such as module appearance images and temperature distribution, so as to provide data support for subsequent module status assessment, defect identification and capacity verification.

[0003] In the process of intelligent identification and capacity testing of photovoltaic modules by drones, intelligent identification of defects usually involves the drone carrying only a single type of sensor to collect data from the modules. After the data is transmitted to the ground terminal, suspected defects are initially marked using basic image analysis algorithms. Then, manual verification is required to confirm the type and location of the defects one by one. Capacity testing, on the other hand, involves manually counting the number of photovoltaic modules in the visible light images taken by the drone, combining the known rated power parameters of a single module, manually calculating the theoretical power generation capacity of the corresponding area, and finally manually comparing the calculation results with the capacity data recorded by the power station system to complete the capacity verification process.

[0004] Generally, traditional intelligent defect identification of photovoltaic modules is carried out in two ways: one is manual ground inspection, in which maintenance personnel walk between photovoltaic arrays with portable testing tools to check for defects such as cracks and hot spots in each module; the other is fixed camera monitoring, in which static cameras are installed at key locations in the power station to continuously capture images of the modules and perform preliminary defect screening. Both methods have obvious drawbacks. Manual ground inspection is extremely inefficient, requiring a lot of manpower and time for large photovoltaic power stations, and is severely affected by personnel experience, physical strength, and inclement weather, making it easy to miss or misdetect. Fixed camera monitoring has problems such as fixed viewing angle, easy obstruction, and inability to cover all modules, making it difficult to capture defects outside the monitoring range. Traditional photovoltaic module capacity detection mainly relies on manually counting the number of modules on-site and then manually calculating the theoretical power generation capacity based on the rated power of each module. This method is not only time-consuming and labor-intensive, but also prone to errors in counting due to visual fatigue during manual counting. Moreover, the calculation results need to be manually entered into the system and compared with the data recorded in the power station, which cannot achieve real-time linkage and makes it difficult to quickly detect problems such as missing modules or configuration parameters that do not match the actual situation.

[0005] Based on this, the present invention provides a photovoltaic module defect identification and capacity detection system based on UAV inspection to solve the above-mentioned technical problems. Summary of the Invention

[0006] To overcome the problems in the prior art, this invention proposes a photovoltaic module defect identification and capacity detection system for unmanned aerial vehicle (UAV) inspection.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a photovoltaic module defect identification and capacity detection system based on drone inspection, comprising: The adapter module is used for sensor calibration and spatiotemporal data synchronization processing before inspection; The drone path planning module is used to plan inspection paths for multiple drones based on the topology of a photovoltaic power station array. The drone inspection module is used to initiate collaborative inspections of drone clusters and dynamically adjust the load and flight attitude. The supplementary lighting module is used for spectral supplementation and image acquisition optimization in complex lighting environments; An edge recognition module is used to identify defects in photovoltaic modules at the edge of the drone based on optimized images. The capacity verification module is used to count the actual number of photovoltaic modules in the inspection area based on the photovoltaic module image, and to perform capacity calibration in the absence of light.

[0008] Furthermore, it also includes a cloud-edge collaboration module, which is used for cleaning, indexing, and retrieving inspection data, including optimized images and photovoltaic module defect identification results.

[0009] Furthermore, it also includes an inversion module, which is used to trace the development process of defects in photovoltaic modules and predict the diffusion trend.

[0010] Furthermore, the adaptation module also includes a multi-source sensor calibration unit and a spatiotemporal data synchronization unit; The multi-source sensor calibration unit performs pixel-level coordinate calibration on the multimodal sensor using a laser calibration plate to achieve spatial dimension alignment of the multimodal sensor data. The multimodal sensor includes an infrared thermal imager and a visible light camera. The spatiotemporal data synchronization unit binds the image data collected by the multimodal sensors with the real-time GPS location data of the UAV through the timestamp synchronization protocol of the UAV flight control system, so as to realize the spatiotemporal correlation between inspection data and the physical location of photovoltaic modules.

[0011] Furthermore, the UAV path planning module includes a cluster path dynamic planning unit and a multi-UAV data relay unit; The cluster path dynamic planning unit is used to generate multi-machine parallel paths without overlapping inspection areas by combining the topology of the photovoltaic power station array, so as to realize the partitioned inspection of large photovoltaic power stations by multiple UAVs. The multi-drone data relay unit is used to establish a temporary two-way data link between adjacent inspection drones to realize the real-time relay transmission of image data during the inspection process.

[0012] Furthermore, the UAV inspection module includes an inspection load real-time monitoring unit and a UAV power adjustment unit; The real-time monitoring unit for inspection load is used to initiate collaborative inspection of the drone swarm. Multiple drones perform inspection tasks according to the planned path. The unit collects real-time weight data from the multimodal sensors and edge recognition modules mounted on the drones through miniature tension sensors, which is used to realize dynamic perception of load changes during the inspection process. The drone power regulation unit is used to adjust the motor output power and propeller speed in real time through the flight control system to achieve stable control of the drone's flight attitude when the load changes dynamically.

[0013] Furthermore, the supplementary lighting module includes an environmental spectral analysis unit and an adaptive supplementary lighting control unit; The environmental spectral analysis unit uses a miniature fiber optic spectrometer to detect the spectral composition and intensity of the current inspection environment light, which is used to identify key spectral bands of photovoltaic module features missing in the illumination. The adaptive supplemental lighting control unit uses a programmable LED array to supplement missing key spectral bands, optimizing the presentation of photovoltaic module texture and defect details in visible light images.

[0014] Furthermore, the edge recognition module includes an image preprocessing unit, a complex lighting adaptive processing unit, and a YOLOv8 model lightweight optimization unit; The image preprocessing unit is used to acquire infrared and visible light images, perform preprocessing, and fuse them to obtain a fused image; The complex illumination adaptive processing unit is used to correct illumination deviations in the fused image through image histogram equalization and dynamic grayscale threshold adjustment. The YOLOv8 model lightweight optimization unit compresses the YOLOv8 target detection model through channel pruning and INT8 quantization techniques, and identifies photovoltaic panel defects in the corrected fused image.

[0015] Furthermore, the cloud-edge collaboration module includes a multi-source data cleaning unit and a spatiotemporal index construction unit; The multi-source data cleaning unit is used to process the image data, defect identification results and GPS location data processed by the edge of the drone, and remove noise data and incomplete data caused by packet loss during the transmission of the drone, so as to achieve quality screening and purification of inspection data. The spatiotemporal index building unit is used to bind the purified inspection data with the geographical coordinates of the photovoltaic power station to build a multi-level spatiotemporal index, which is used to realize the rapid retrieval and access of inspection data in different regions and at different times.

[0016] Furthermore, the capacity verification module includes a component quantity statistics unit and a theoretical capacity calculation unit; The component quantity statistics unit is used for photovoltaic component images to realize the statistics of the number of photovoltaic components in the inspection area; The theoretical capacity calculation unit is used to calculate the theoretical power generation capacity of the inspection area based on the statistical data of the number of photovoltaic modules and the rated power of a single photovoltaic module, and to automatically compare it with the rated capacity recorded by the power station system.

[0017] Compared with the prior art, the present invention has the following technical effects: This invention achieves precise fusion of infrared thermal imaging and visible light data through multi-sensor synchronous calibration and spatiotemporal data binding, enabling more comprehensive and accurate capture of component defect characteristics and avoiding misjudgments and omissions caused by single data dimensions. Relying on drone swarm collaborative inspection and data relay transmission, it significantly improves the inspection coverage speed and data transmission stability of large photovoltaic power plants. Through real-time computation of a lightweight edge model and spectral supplementary lighting optimization under complex illumination, it achieves real-time defect identification and accuracy assurance, reducing reliance on traditional manual verification, lowering labor costs and human error. By reconstructing defect time-series data and simulating hot spot diffusion, the defect development process can be traced, providing forward-looking decision support for operation and maintenance. Dynamic load adaptation allows for real-time adjustment of drone flight attitude, ensuring stable data acquisition and data quality during load changes. Cloud-edge collaborative data cleaning and indexing improve the efficiency of multi-source inspection data management, enabling rapid data retrieval and retrieval. Automatic capacity verification automatically completes component counting and capacity calculation through algorithms and compares it with power plant system data in real time, completely solving the inefficiency and error-prone nature of traditional manual counting and comparison, and quickly detecting capacity deviations. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is a schematic diagram of the overall system structure of the present invention; Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solutions proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. Specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] Please see Figure 1 This invention proposes a photovoltaic module defect identification and capacity detection system based on UAV inspection, including an adaptation module, a UAV path planning module, a UAV inspection module, a supplementary lighting module, an edge recognition module, a cloud-edge collaboration module, an inversion module, and a capacity verification module; The adapter module is used for sensor calibration and spatiotemporal data synchronization processing before inspection; The drone path planning module is used to plan inspection paths for multiple drones based on the topology of a photovoltaic power plant array. The drone inspection module is used to initiate collaborative inspections of drone clusters and dynamically adjust the load and flight attitude. The supplementary lighting module is used for spectral supplementation and image acquisition optimization in complex lighting environments; The edge recognition module is used to identify photovoltaic module defects at the edge of the drone based on optimized images; The cloud-edge collaboration module is used for cleaning, indexing, and retrieving inspection data, which includes optimized images and photovoltaic module defect identification results. The inversion module is used to trace the development process of defects in photovoltaic modules; The capacity verification module is used to count the actual number of photovoltaic modules in the inspection area based on the photovoltaic module image, and to perform capacity calibration in the absence of light.

[0022] In this embodiment, the adapter module is used to perform sensor calibration and spatiotemporal data synchronization processing before inspection, and it includes a multi-source sensor calibration unit and a spatiotemporal data synchronization unit.

[0023] The multi-source sensor calibration unit focuses on solving the spatial alignment problem between sensors. During drone inspections, the drone carries multimodal sensors, including an infrared thermal imager and a visible light camera. Due to the different installation positions and viewing angles of the multimodal sensors, the collected inspection data will be offset in the spatial dimension. To eliminate this offset, the multi-source sensor calibration unit is used to perform pixel-level coordinate calibration of the infrared thermal imager and visible light camera on the drone using a laser calibration plate, ensuring that the inspection data is accurately aligned in the spatial dimension.

[0024] The spatiotemporal data synchronization unit is used to achieve spatiotemporal correlation between inspection data and the actual location of photovoltaic modules. It binds each frame of image data collected by the multimodal sensor with the real-time GPS location information of the drone, ensuring that each image frame corresponds to specific physical coordinates and time points, thus achieving spatiotemporal correlation between inspection data and the actual location of photovoltaic modules. Specifically, relying on the timestamp synchronization protocol of the drone flight control system, during the drone's inspection mission, when the multimodal sensor collects each frame of image data, the flight control system synchronously generates a timestamp and captures the drone's real-time GPS location information at that moment, binding this location information to the corresponding frame of image data. In this way, each image carries its geographical location information at the moment it was captured, enabling subsequent analysis of the inspection data to clearly identify the specific location of each detected photovoltaic module defect or anomaly within the power plant, providing solid and reliable data support for the operation and maintenance, fault location, and performance evaluation of photovoltaic power plants.

[0025] In this embodiment, the UAV path planning module is used to plan inspection paths for multiple UAVs based on the topology of the photovoltaic power station array. It includes a cluster path dynamic planning unit and a multi-UAV data relay unit.

[0026] The cluster path dynamic planning unit is used to acquire topological information such as the component array distribution, area division, and obstacle locations of the target photovoltaic power station, and based on this, generates parallel paths for multiple drones without overlapping inspection areas. Specifically, firstly, based on GPS positioning information or known map information, the distribution of the target photovoltaic power station's component array is obtained, clarifying the specific location and arrangement of each component within the photovoltaic power station; simultaneously, different areas of the photovoltaic power station are divided; and the locations of obstacles, including buildings, trees, utility poles, and other objects that may affect drone flight, are identified. The component array distribution, area division, and obstacle locations of the target photovoltaic power station form topological information. Based on this topological information, parallel paths for multiple drones without overlapping inspection areas are generated. This planning method ensures that multiple drones can perform zoned inspections of large photovoltaic power stations, with each drone responsible for a specific area. This avoids blind spots, ensures that every part of the power station is inspected, prevents duplicate work, improves overall inspection efficiency, and shortens the inspection cycle.

[0027] The multi-drone data relay unit focuses on solving the data transmission challenges of UAVs during inspections. Due to the limited communication range of a single UAV, data transmission may be delayed or incomplete during the inspection of large photovoltaic power plants. To address this, the multi-drone data relay unit establishes temporary bidirectional data links by equipping adjacent inspection UAVs with millimeter-wave communication modules. Millimeter-wave communication offers advantages such as high bandwidth, fast transmission speed, and strong anti-interference capabilities, meeting the real-time relay transmission requirements of image data during inspections. During the inspection, once a UAV completes inspection of a certain area and collects a large amount of image data, it can transmit the data in real-time to adjacent UAVs via the temporarily established bidirectional data link. The adjacent UAVs then continue the relay transmission until the data is finally transmitted to the ground control center. This method effectively avoids the limitation of single-drone communication range, ensuring timely and accurate data transmission and providing strong support for subsequent data analysis and processing.

[0028] In this embodiment, the UAV inspection module is used to initiate collaborative inspection of the UAV cluster and dynamically adjust the load and flight attitude. It includes an inspection load real-time monitoring unit and a UAV power adjustment unit. The real-time load monitoring unit is used to initiate collaborative inspections by a swarm of drones. Multiple drones execute inspection tasks according to a planned path. Real-time weight data from the multimodal sensors and edge recognition modules mounted on the drones is collected via miniature tension sensors to dynamically perceive load changes during the inspection process. Specifically, during the initiation phase, the real-time load monitoring unit schedules multiple drones to execute inspection tasks in an orderly manner along pre-planned paths, ensuring that each drone clearly understands its inspection area and route, avoiding inspection confusion or omissions. During the inspection process, the miniature tension sensors mounted on the drones continuously collect real-time weight data from the multimodal sensors (such as high-definition cameras and infrared thermal imagers) and edge recognition modules. This data reflects load changes during the inspection. For example, when multimodal sensors collect different types of data or when the edge recognition module performs complex calculations, the weight may fluctuate slightly. By collecting this data in real time, dynamic perception of load changes is achieved, providing crucial information for subsequent power adjustments.

[0029] The UAV power regulation unit is used to adjust the motor output power and propeller speed in real time through the flight control system to achieve stable flight attitude control of the UAV when the load changes dynamically. Specifically, based on the load data fed back by the real-time inspection load monitoring unit, the flight control system adjusts the motor output power and propeller speed of the UAV in real time. When an increase in load is detected, the flight control system quickly increases the motor output power and propeller speed to provide stronger lift for the UAV to maintain a stable flight altitude and attitude; conversely, when the load decreases, the motor output power and propeller speed are reduced accordingly to prevent the UAV from becoming unstable due to excessive lift. Through this power regulation, stable flight attitude control of the UAV is achieved when the load changes dynamically, ensuring that the UAV maintains stable flight during the inspection process, avoiding flight attitude deviations caused by load fluctuations, and thus ensuring inspection accuracy.

[0030] In this embodiment, the supplementary lighting module is used for spectral supplementation and image acquisition optimization in complex lighting environments; it includes an environmental spectral analysis unit and an adaptive supplementary lighting control unit.

[0031] The environmental spectral analysis unit uses a miniature fiber optic spectrometer to detect the spectral composition and intensity of the current ambient light during inspection, identifying key spectral bands for photovoltaic module feature recognition that are missing from the illumination. Specifically, the unit uses the miniature fiber optic spectrometer to meticulously detect the current ambient light, capturing its spectral composition and clarifying the intensity distribution of different wavelengths. Through in-depth analysis of this data, missing spectral bands are identified, and these missing bands are precisely the key to photovoltaic module feature recognition. For example, certain wavelengths of light play a crucial role in highlighting the surface texture of photovoltaic modules and identifying subtle defects; if these bands are missing, the quality of the acquired images will be significantly reduced, affecting subsequent analysis and judgment.

[0032] The adaptive supplementary lighting control unit emits a supplementary light beam with a wavelength of 450nm-550nm through a programmable LED array to supplement missing key spectral bands and optimize the rendering effect of photovoltaic module texture and defect details in visible light images. Specifically, a programmable LED array is used as the supplementary light source. The programmable LED array has flexible and adjustable characteristics and can emit beams of specific wavelengths according to actual needs. In this embodiment, the LED array is controlled to emit a supplementary light beam with a wavelength range of 450nm-550nm. This wavelength band of light can precisely supplement the previously identified missing key spectral bands. When the supplementary light beam illuminates the photovoltaic module, it can effectively optimize the rendering effect of the photovoltaic module in the visible light image. The module texture, which was originally blurred due to lack of illumination, becomes clearly visible, and even subtle defects can be accurately captured, greatly improving the quality and detail richness of the visible light image.

[0033] In this embodiment, the edge recognition module is used to perform photovoltaic module defect recognition processing at the edge of the drone based on the optimized image. It includes an image preprocessing unit, a complex lighting adaptive processing unit, and a YOLOv8 model lightweight optimization unit.

[0034] The image preprocessing unit is used to acquire and preprocess infrared and visible light images. Specifically, a Gaussian filtering algorithm (σ=1.0) is used to remove noise interference from the infrared image, with a filtering window size of 3×3, effectively smoothing artifacts caused by uneven temperature distribution. A bilateral filtering algorithm (spatial sigma=5, grayscale sigma=20) is used for the illuminated visible light image to preserve edge details and avoid blurring of crack features after filtering. A weighted fusion algorithm is used to dynamically assign weights to the registered images according to regions: the infrared image weight for temperature anomaly (hot spot) regions is set to 0.7 to highlight temperature features; the visible light image weight for appearance inspection regions is set to 0.7 to highlight surface defect features, generating a fused image.

[0035] The complex illumination adaptive processing unit corrects image feature deviations under strong direct sunlight or shadow occlusion by image histogram equalization and dynamic grayscale threshold adjustment, thereby maintaining the recognition accuracy of photovoltaic module cracks and hot spot defects under complex illumination environments. Specifically, this includes: real-time analysis of the brightness histogram of the fused image, calculating the average brightness value L_avg ranging from 0 to 255; if L_avg < 50 (strong light backlight scene, loss of detail in dark areas), adjusting the gamma value to 0.8-0.9 to enhance dark details through power transformation; if L_avg > 200 (overcast low light scene, overall image is too bright), adjusting the gamma value to 1.1-1.2 to improve overall contrast; if 50 ≤ L_avg ≤ 200 (normal lighting), keeping the gamma value at 1.0; performing grayscale stretching on the adjusted image, mapping the pixel grayscale range from [L_min, L_max] to [0, 255] to amplify the difference between defect features and the background. The stretching formula is: G(x,y) = 255 × (F(x,y) - L_min) / (L_max - L_min) (F(x,y) is the adjusted pixel value, G(x,y) is the average brightness value, and L(x,y) is the average brightness value). (These are the stretched pixel values).

[0036] The YOLOv8 model lightweight optimization unit compresses the YOLOv8 target detection model through channel pruning and INT8 quantization techniques, and performs real-time identification and classification of photovoltaic panel defects (cracks, hot spots, damage, stains) in the fused image, outputting defect type, location coordinates and confidence level.

[0037] Specifically, a photovoltaic panel defect dataset was first constructed, collecting 120,000 valid images covering four common defect types: 30,000 images of cracks, 40,000 images of hot spots, 30,000 images of damage, and 20,000 images of stains. Each image was meticulously annotated by professionals using specialized annotation tools, clearly marking not only the defect bounding boxes but also the corresponding category labels. This ensured the accuracy and consistency of the annotation results, laying a solid data foundation for subsequent model training.

[0038] Subsequently, the photovoltaic panel defect dataset was divided into training, validation, and test sets in a 7:2:1 ratio. The training set contained 84,000 images, the validation set contained 24,000 images, and the test set contained 12,000 images. For the training environment, an NVIDIA RTX 3090 GPU was selected, whose powerful computing performance provided strong support for model training and significantly shortened the training time.

[0039] YOLOv8 was used as the base object detection model, and the training parameters were set as follows: epochs were set to 150 to ensure that the model had sufficient time for learning and optimization; imgsz was set to 640 to ensure both detection accuracy and computational efficiency; the batch size was set to 16 to make reasonable use of GPU memory resources and improve training speed; the initial learning rate lr0 was set to 0.01 to provide a suitable step size for updating model parameters; at the same time, a cosine annealing learning rate decay strategy was adopted to dynamically adjust the learning rate during training, further optimizing the model training effect.

[0040] To better adapt the model to the limited computing power of edge devices, further optimization is needed. L1 regularization pruning is employed to remove redundant parameters, effectively reducing model complexity. Simultaneously, the model weight precision is quantized from 32-bit floating-point to 16-bit half-precision, significantly compressing the model size without substantially affecting performance, making it easier to deploy and run on edge devices. For edge deployment, the NVIDIA Jetson Nano is chosen as the deployment platform, and the optimized model is loaded onto this device. To cope with complex and variable lighting environments, a lighting compensation algorithm is introduced. The gamma value is adjusted (ranging from 0.8 to 1.2) to adjust the image lighting, enhancing the image's recognizability under different lighting conditions. In the defect detection stage, a confidence threshold of 0.7 is set. When the confidence level of the model detection result is greater than or equal to 0.7, it is considered a valid defect; if the confidence level is lower than this threshold, the result is discarded, thereby improving the accuracy and reliability of defect detection.

[0041] In this embodiment, the inversion module is used to trace the development process of photovoltaic module defects and predict the diffusion trend. It includes a defect time series data reconstruction unit and a hot spot diffusion simulation unit.

[0042] The defect time-series data reconstruction unit retrieves historical inspection data stored on the cloud platform and stitches together multiple defect images of the same photovoltaic module in a time sequence to trace the development process of defects from their initial occurrence to their current state. Specifically, the unit retrieves stored historical inspection data, which covers detailed information about the photovoltaic module at different time points, especially multiple defect images. Following strict time-series rules, it stitches together multiple defect images of the same photovoltaic module, integrating the originally scattered image information into a coherent whole. This allows for clear tracing of the defect's development from subtle initial signs to gradual morphological changes, and finally to its current complete state.

[0043] The hot spot diffusion simulation unit uses the finite element method (FEM) algorithm to numerically simulate the temperature field distribution in the hot spot region. Combined with the thermal conductivity parameters of the component materials, it is used to predict the diffusion trend of the hot spot over a future period. The FEM algorithm is employed to perform a detailed numerical simulation of the temperature field distribution in the hot spot region. The FEM algorithm divides the hot spot region into countless tiny units, and by calculating and simulating the temperature changes in each unit, a temperature field distribution model of the entire hot spot region is constructed. The thermal conductivity of the component materials is considered during the simulation. Different materials have different thermal conductivity, which directly affects the speed and manner of heat transfer within the component. By combining the thermal conductivity of the component materials, the simulation results can reflect the diffusion of the hot spot in the actual environment. Through analysis of the simulation data, the diffusion range and speed of the hot spot over a future period, as well as its potential impact on the surrounding area, are predicted. This provides a scientific basis for taking targeted maintenance measures in advance to prevent further hot spot diffusion that could lead to severe performance degradation or even damage to the component.

[0044] The hot spot diffusion simulation unit uses the finite element analysis algorithm to numerically simulate the temperature field distribution in the hot spot region. Combined with the thermal conductivity parameters of the component material, it is used to predict the diffusion trend of the hot spot over a future period of time. The specific implementation steps are as follows: Step S201: Establish a three-dimensional geometric model of the photovoltaic module and configure material parameters.

[0045] Based on the actual stacked structure of photovoltaic modules, a three-dimensional geometric model including the glass cover, encapsulating film (EVA / POE), cell layers, backsheet, and frame is constructed in a finite element analysis environment. Thermal properties such as thermal conductivity, density, and specific heat capacity are configured for each layer of materials in the three-dimensional geometric model. Among them, thermal conductivity is a key parameter that directly affects the rate and path of heat conduction inside the module.

[0046] Step S202: Set the hot spot heat source model and boundary conditions.

[0047] Based on the hotspot location and morphology information obtained from the defect time-series data reconstruction unit, the geometric contour and heat source intensity of the hotspot region are defined in the three-dimensional geometric model. The heat source intensity is quantified and set according to the power loss density of the hotspot region. At the same time, according to the actual operating environment of the photovoltaic module, convective heat transfer boundary conditions between the module surface and the air are set, and radiative heat transfer boundaries are selectively added according to environmental conditions to simulate the heat dissipation mechanism under real operating conditions.

[0048] Step S203: Perform finite element mesh generation and local refinement.

[0049] The constructed 3D geometric model is meshed using finite element methods, and the model is discretized using tetrahedral or hexahedral elements. Local mesh refinement is performed on the hot spot region and its adjacent areas to ensure sufficient computational accuracy in regions with drastic temperature gradient changes, while controlling the overall mesh count to balance computational efficiency.

[0050] Step S204: Solving the transient heat conduction equation and calculating the temperature field.

[0051] Based on the finite element method, the transient heat conduction control equations are solved, and the temperature field distribution of the hot spot region in the time domain is numerically calculated. By setting the time step and convergence criterion, the temperature change of each tiny unit in the model at each time node is iteratively solved, and the temperature field distribution sequence of the hot spot region evolving from the current state to the future is gradually constructed.

[0052] Step S205: Hot spot diffusion trend analysis and prediction.

[0053] Post-processing analysis is performed on the obtained temperature field time-series data to extract key characteristic parameters such as the hot spot center temperature, the boundary of the heat-affected zone, and the temperature gradient distribution. By analyzing the rate of change of the expansion radius of the heat-affected zone over time, the diffusion rate and range of the hot spot are quantified. Based on the diffusion patterns within the simulated time period, the temperature field distribution of the hot spot at specific future time points (e.g., 1 hour, 2 hours, 4 hours) is extrapolated and predicted to assess the potential impact of hot spot diffusion on surrounding solar cells and encapsulation materials.

[0054] Step S206: Output the prediction results and maintain decision support.

[0055] The predicted hotspot propagation trend is output in a visual format, including a dynamic evolution cloud map of the temperature field, curves showing changes in key characteristic parameters, and a trend map of the expansion of the thermally affected area. Combined with preset temperature thresholds and material tolerance limits, graded early warning information and maintenance recommendations are generated. This provides quantitative evidence for maintenance personnel to take targeted maintenance measures in advance, such as removing obstructions and replacing faulty components, preventing further hotspot propagation that could lead to severe performance degradation or even permanent damage to components.

[0056] In this embodiment, the cloud-edge collaboration module is used for cleaning, indexing, and retrieving inspection data, which includes optimized images and photovoltaic module defect identification results. The cloud-edge collaboration module includes a multi-source data cleaning unit and a spatiotemporal index construction unit.

[0057] The multi-source data cleaning unit processes image data, defect identification results, and GPS location data processed at the UAV edge. It uses the Isolation Forest algorithm to remove noise data and incomplete data caused by packet loss during UAV transmission, thus achieving quality screening and purification of inspection data. Specifically, during UAV inspection missions, the complex and variable wireless transmission environment inevitably generates noisy data, and packet loss may result in incomplete data. Using this poor-quality data directly for subsequent analysis will affect the accuracy and reliability of the results. Therefore, the Isolation Forest algorithm is introduced to screen image data, defect identification results, and GPS location data. The Isolation Forest algorithm can quickly identify outliers in the data, removing abnormal data caused by noise interference or packet loss, thereby achieving quality screening and purification of inspection data.

[0058] The spatiotemporal index construction unit uses Geohash encoding to bind the purified inspection data with the geographical coordinates of the photovoltaic power station, constructing a multi-level spatiotemporal index for rapid retrieval and access to inspection data from different regions and times. Specifically, the spatiotemporal index construction unit focuses on solving the problem of rapid retrieval and access to inspection data. In photovoltaic power stations, inspection data is often closely related to geographical location and time period. To quickly locate inspection data from different regions and times, Geohash encoding is used, allowing the purified inspection data to be tightly bound to the geographical coordinates of the photovoltaic power station. Geohash encoding can convert two-dimensional geographical coordinates into string format. Based on this, a multi-level spatiotemporal index structure is constructed to retrieve the required inspection data.

[0059] In this embodiment, the capacity verification module is used to count the actual number of photovoltaic modules in the inspection area based on the photovoltaic module image, and to perform capacity calibration in the absence of sunlight. The capacity verification module includes a module quantity counting unit and a theoretical capacity calculation unit.

[0060] The component quantity counting unit is used for photovoltaic component images to count the number of photovoltaic components within the inspection area. Specifically, based on GPS positioning information, the photovoltaic power station is divided into a 50m×50m rectangular inspection area, with each area assigned a unique location code (e.g., "P01-P20"). A photovoltaic component recognition model is built using the YOLOv8 model to identify the photovoltaic components in each area. The non-maximum suppression (NMS) algorithm (IOU threshold 0.5) is used to remove duplicate markers and count the number of components in each area.

[0061] The theoretical capacity calculation unit is used to calculate the theoretical power generation capacity of the inspection area based on the statistical data of the number of photovoltaic modules and the rated power of a single photovoltaic module, and to automatically compare it with the rated capacity recorded by the power station system.

[0062] Specifically, the rated power P0 of a single photovoltaic module is obtained from the photovoltaic power station archive database. In this embodiment, P0 = 280W, and the photovoltaic module model is JKM325M-72HL4. The theoretical power generation capacity of the photovoltaic power station is calculated as P_theo = N × P0. In this embodiment, N = 4000 modules, and P_theo = 4000 × 280W = 1120kW.

[0063] The actual power output P_act of the inverter (model Sungrow SG1250HV) is acquired in real time via the edge 4G communication module. The data acquisition interval is 5 minutes, and the capacity deviation rate is calculated. The deviation rate calculation formula is as follows: δ= (|P_theo - P_act| / P_theo)×100%; Where δ is the capacity deviation rate, P_theo is the theoretical power generation capacity (1120kW), and P_act is the actual power generation. The real-time acquisition value in this embodiment is 1050kW. Judgment Logic: The preset threshold δ_0 in this embodiment is δ=(|1120-1050| / 1120)×100%=6.25%>5%. If δ≤5%, the capacity is judged to be normal; if δ>5%, the capacity is judged to be abnormal. Simultaneously trigger on-site audible and visual alarms (drones equipped with buzzers and LED warning lights, alarm frequency 1 time / second) and mobile SMS notifications to maintenance personnel, sent via Alibaba Cloud SMS API, with the content "Abnormal area: P05-P08, deviation rate 6.25%". Combined with the component statistics of each area, the possible deviation area is located. In the specific embodiment, the defective component ratio of the P05-P08 area is 8%, which is higher than the 2% of other areas, and is judged to be the main source of deviation.

[0064] Based on the same inventive concept, embodiments of the present invention also provide a method for defect identification and capacity testing of photovoltaic modules based on UAV inspection. The solution provided by this method is similar to the solution described in the above system; therefore, the specific limitations in one or more system embodiments provided below can be found in the limitations of the UAV-based photovoltaic module defect identification and capacity testing system described above, and will not be repeated here.

[0065] Please see Figure 2 The method for defect identification and capacity testing of photovoltaic modules based on the above system using drone inspection includes the following steps: S1. Perform sensor calibration and spatiotemporal data synchronization processing before inspection.

[0066] A laser calibration board is used to perform pixel-level coordinate calibration on the infrared thermal imager and visible light camera carried by the UAV to ensure that the spatial dimensions of the data collected by the two types of devices are aligned. Meanwhile, by using the high-precision timestamp synchronization protocol of the UAV flight control system, the image data subsequently collected by the sensors is bound to the real-time GPS location data of the UAV, realizing the spatiotemporal correlation between inspection data and the physical location of photovoltaic modules, laying the foundation for subsequent multi-sensor data fusion.

[0067] S2. Planning multi-UAV inspection paths based on the photovoltaic power station array topology.

[0068] The system acquires topological information such as the distribution of the target photovoltaic power station's component arrays, area divisions, and obstacle locations. Based on this information, it generates parallel paths for multiple drones without overlapping inspection areas, ensuring that multiple drones can perform efficient zonal inspections of large photovoltaic power stations, avoiding blind spots and repetitive work, and improving overall inspection efficiency. S3. Initiate collaborative inspection of the drone swarm and dynamically adjust the load and flight attitude.

[0069] Multiple drones perform inspection tasks according to the planned path. During the process, they collect real-time weight data of the drone's mounted equipment (including infrared / visible light acquisition equipment and computing modules) through miniature tension sensors to sense load changes during the inspection. The flight control system adjusts the motor output power and propeller speed in real time to maintain the stability of the UAV's flight attitude when the load changes dynamically, thus avoiding a decrease in image acquisition accuracy due to load fluctuations.

[0070] S4. Perform spectral supplementary lighting in complex lighting environments.

[0071] The spectral composition and intensity of the ambient light during the current inspection are detected by a miniature fiber optic spectrometer, which identifies the missing spectral bands in the illumination that are crucial for identifying defects in photovoltaic modules. Subsequently, the programmable LED array is controlled to emit supplementary light beams with wavelengths of 450nm-550nm to fill in the missing key spectral bands, optimize the presentation of photovoltaic module texture and defect details in the visible light image, and simultaneously obtain the visible light image and infrared thermal imaging image after supplementary lighting.

[0072] S5. Perform real-time defect identification and processing at the edge of the drone.

[0073] The YOLOv8 target detection model is compressed using channel pruning and INT8 quantization techniques. The compressed model is then loaded into the UAV edge computing module to perform real-time calculations on the acquired image data. During this process, image feature deviations under strong direct sunlight or shadow occlusion are corrected by image histogram equalization and dynamic grayscale threshold adjustment algorithms, enabling real-time identification of defects such as cracks and hot spots in photovoltaic modules.

[0074] S6. Upload the inspection data to the cloud and perform cleaning and indexing.

[0075] The image data, defect identification results, and GPS location data processed at the edge of the drone are synchronously uploaded to the cloud platform. The isolated forest algorithm is used to remove noise data generated during transmission and incomplete data caused by packet loss, thereby completing the quality screening and purification of inspection data. Then, the purified inspection data is bound to the geographical coordinates of the photovoltaic power station through Geohash encoding to build a multi-level spatiotemporal index, which facilitates the rapid retrieval and access of inspection data from different regions and times.

[0076] S7. Trace the development process of defects.

[0077] Historical inspection data of the target photovoltaic module is retrieved from the cloud platform, and multiple defect images of the module are stitched together in time sequence to trace the development process of the defect from its initial occurrence to its current state. Meanwhile, the finite element analysis algorithm is used to numerically simulate the temperature field distribution in the hot spot region, and combined with the thermal conductivity parameters of the photovoltaic module material, the diffusion trend of the hot spot in the future is predicted.

[0078] S8. Count the number of photovoltaic modules and complete the power generation capacity verification.

[0079] Based on the photovoltaic module image with edge recognition, the Hough transform line detection algorithm is used to locate the module border and count the actual number of photovoltaic modules in the inspection area. Based on the statistical data of the number of components and the rated power of a single component, the theoretical power generation capacity of the inspection area is calculated. The theoretical power generation capacity is automatically compared with the rated capacity recorded by the power station system to identify potential component configuration deviations or missing components.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A photovoltaic module defect identification and capacity detection system based on unmanned aerial vehicle (UAV) inspection, characterized in that, include: The adapter module is used for sensor calibration and spatiotemporal data synchronization processing before inspection; The drone path planning module is used to plan inspection paths for multiple drones based on the topology of a photovoltaic power station array. The drone inspection module is used to initiate collaborative inspections of drone clusters and dynamically adjust the load and flight attitude. The supplementary lighting module is used for spectral supplementation and image acquisition optimization in complex lighting environments; An edge recognition module is used to identify defects in photovoltaic modules at the edge of the drone based on optimized images. The capacity verification module is used to count the actual number of photovoltaic modules in the inspection area based on the photovoltaic module image, and to perform capacity calibration in the absence of light.

2. The photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, It also includes a cloud-edge collaboration module, which is used for cleaning, indexing and retrieving inspection data, including optimized images and photovoltaic module defect identification results.

3. The photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 2, characterized in that, It also includes an inversion module, which is used to trace the development process of defects in photovoltaic modules.

4. The photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The adapter module also includes a multi-source sensor calibration unit and a spatiotemporal data synchronization unit; The multi-source sensor calibration unit performs pixel-level coordinate calibration on the multimodal sensor using a laser calibration plate to achieve spatial dimension alignment of the multimodal sensor data. The multimodal sensor includes an infrared thermal imager and a visible light camera. The spatiotemporal data synchronization unit binds the image data collected by the multimodal sensors with the real-time GPS location data of the UAV through the timestamp synchronization protocol of the UAV flight control system, so as to realize the spatiotemporal correlation between inspection data and the physical location of photovoltaic modules.

5. A photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The UAV path planning module includes a cluster path dynamic planning unit and a multi-UAV data relay unit; The cluster path dynamic planning unit is used to generate multi-machine parallel paths without overlapping inspection areas by combining the topology of the photovoltaic power station array, so as to realize the partitioned inspection of large photovoltaic power stations by multiple UAVs. The multi-drone data relay unit is used to establish a temporary two-way data link between adjacent inspection drones to realize the real-time relay transmission of image data during the inspection process.

6. The photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The UAV inspection module includes an inspection load real-time monitoring unit and a UAV power adjustment unit. The real-time monitoring unit for inspection load is used to initiate collaborative inspection of the drone swarm. Multiple drones perform inspection tasks according to the planned path. The unit collects real-time weight data from the multimodal sensors and edge recognition modules mounted on the drones through miniature tension sensors, enabling dynamic perception of load changes during the inspection process. The drone power regulation unit is used to adjust the motor output power and propeller speed in real time through the flight control system to achieve stable control of the drone's flight attitude when the load changes dynamically.

7. A photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The supplemental lighting module includes an environmental spectrum analysis unit and an adaptive supplemental lighting control unit; The environmental spectral analysis unit uses a miniature fiber optic spectrometer to detect the spectral composition and intensity of the current inspection environment light, which is used to identify key spectral bands of photovoltaic module features missing in the illumination. The adaptive supplemental lighting control unit uses a programmable LED array to supplement missing key spectral bands, optimizing the presentation of photovoltaic module texture and defect details in visible light images.

8. A photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The edge recognition module includes an image preprocessing unit, a complex lighting adaptive processing unit, and a YOLOv8 model lightweight optimization unit. The image preprocessing unit is used to acquire infrared and visible light images, perform preprocessing, and fuse them to obtain a fused image; The complex illumination adaptive processing unit is used to correct illumination deviations in the fused image through image histogram equalization and dynamic grayscale threshold adjustment. The YOLOv8 model lightweight optimization unit compresses the YOLOv8 target detection model through channel pruning and INT8 quantization techniques, and identifies photovoltaic panel defects in the corrected fused image.

9. A photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 2, characterized in that, The cloud-edge collaboration module includes a multi-source data cleaning unit and a spatiotemporal index construction unit; The multi-source data cleaning unit is used to process the image data, defect identification results and GPS location data processed by the edge of the drone, and remove noise data and incomplete data caused by packet loss during the transmission of the drone, so as to achieve quality screening and purification of inspection data. The spatiotemporal index building unit is used to bind the purified inspection data with the geographical coordinates of the photovoltaic power station to build a multi-level spatiotemporal index, which is used to realize the rapid retrieval and access of inspection data in different regions and at different times.

10. A photovoltaic module defect identification and capacity detection system based on UAV inspection according to claim 1, characterized in that, The capacity verification module includes a component quantity statistics unit and a theoretical capacity calculation unit; The component quantity statistics unit is used for photovoltaic module images to realize the statistics of the number of photovoltaic modules in the inspection area; The theoretical capacity calculation unit is used to calculate the theoretical power generation capacity of the inspection area based on the statistical data of the number of photovoltaic modules and the rated power of a single photovoltaic module, and to automatically compare it with the rated capacity recorded by the power station system.

Images