Photovoltaic module defect detection system and method in daytime environment, and inspection system

Abstract

The invention provides a photovoltaic module defect detection system and method in a daytime environment, and an inspection system, and can be used in the technical field of photovoltaic power generation. The system comprises a high-voltage excitation module, an image acquisition module, an image processing module and a defect detection module, the high-voltage excitation module is used for inputting a coding current excitation signal based on code domain modulation to at least one photovoltaic module so as to excite the photovoltaic module to generate a time-varying electroluminescence signal; wherein the coding current excitation signal is an alternating current or a pseudo-random sequence current; the image acquisition module is used for continuously acquiring infrared images of the photovoltaic module to obtain an infrared image sequence; the image processing module is used for acquiring an electroluminescent image of the photovoltaic module based on the infrared image sequence; and the defect detection module is used for obtaining a defect detection result of the photovoltaic module based on the electroluminescent image. Daytime detection of hidden defects of the photovoltaic module can be realized, and convenience and timeliness of defect detection of the photovoltaic module in a daytime environment are greatly improved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, and in particular to a photovoltaic module defect detection system and method and inspection system under daytime conditions. Background Technology

[0002] Photovoltaic module defect inspection is a key link in ensuring the efficient and stable operation of photovoltaic power generation systems. Its role is to accurately identify potential defects inside the modules, such as hidden cracks, broken grids, poor soldering, black cores, and PID attenuation, through precision testing technologies (such as electroluminescence (EL) testing and photoluminescence (PL) testing).

[0003] However, traditional EL testing typically requires dark conditions to achieve a sufficiently high signal-to-noise ratio. In large photovoltaic power plants, this means necessitating daytime shading, module removal and placement in a dark room, or waiting for nighttime testing, resulting in stringent environmental requirements and low testing efficiency. PL testing, on the other hand, requires high-power laser light sources to excite the modules. In large-scale outdoor applications, this leads to bulky equipment, complex water-cooling systems, and significant operational inconvenience.

[0004] Therefore, there is currently a lack of a detection method that can efficiently detect defects in photovoltaic modules even in daytime environments. Summary of the Invention

[0005] This application provides a photovoltaic module defect detection system and method, and an inspection system for daytime environments, to solve the technical problem that existing photovoltaic module defect detection technologies cannot achieve efficient detection of photovoltaic module defects in daytime environments.

[0006] According to the first aspect disclosed in this application, this application provides a photovoltaic module defect detection system for daytime environments, comprising: A high-voltage excitation module is used to input a coded current excitation signal based on code domain modulation to at least one photovoltaic module to excite the photovoltaic module to generate a time-varying electroluminescent signal; wherein the coded current excitation signal is an alternating current or a pseudo-random sequence current; An image acquisition module is used to continuously acquire infrared images of the photovoltaic module and obtain an infrared image sequence; An image processing module is used to acquire an electroluminescent image of the photovoltaic module based on the infrared image sequence; A defect detection module is used to obtain defect detection results of the photovoltaic module based on the electroluminescent image.

[0007] In one feasible implementation, the high-voltage excitation module includes a programmable high-voltage DC power supply and a spread spectrum modulation control unit; The spread spectrum modulation control unit is used to control the programmable high voltage DC power supply so that the programmable high voltage DC power supply outputs the coded current excitation signal.

[0008] In one feasible implementation, the image acquisition module includes an infrared detection device for acquiring infrared images of the photovoltaic module.

[0009] In one feasible implementation, the detection lens of the infrared detection device is provided with a narrowband optical filter or a long-pass optical filter.

[0010] According to the second aspect disclosed in this application, this application provides a photovoltaic module defect inspection system for daytime environment, including the photovoltaic module defect detection system for daytime environment as described in any one of the first aspects, and a mobile platform; The image acquisition module of the photovoltaic module defect detection system under daytime conditions is mounted on the mobile platform.

[0011] According to a third aspect disclosed in this application, this application provides a method for detecting defects in photovoltaic modules under daytime conditions, based on a photovoltaic module defect detection system under daytime conditions as described in any one of the first aspects, comprising: A coded current excitation signal based on code domain modulation is input to at least one photovoltaic module to excite the photovoltaic module to generate a time-varying electroluminescent signal; wherein the coded current excitation signal is an alternating current or a pseudo-random sequence current; Infrared images of the photovoltaic module are continuously acquired to obtain an infrared image sequence; Based on the infrared image sequence, the electroluminescence image of the photovoltaic module is obtained; Based on the electroluminescent image, the defect detection results of the photovoltaic module are obtained.

[0012] In one feasible implementation, the alternating current includes a periodically spaced forward current and a zero current; the infrared image sequence includes a set of luminescent images and a set of background images; the set of luminescent images includes multiple luminescent image frames acquired during the forward current cycle; and the set of background images includes multiple background image frames acquired during the zero current cycle. Based on the infrared image sequence, the electroluminescence image of the photovoltaic module is obtained, including: The set of luminescent images and the set of background images are respectively subjected to frame-by-frame image registration and averaging to obtain average luminescent images and average background images; The average luminescence image and the average background image are differentially processed to obtain an electroluminescent image.

[0013] In one feasible implementation, the pseudo-random sequence current includes a pseudo-randomly distributed forward current and a zero current, the infrared image sequence includes multiple infrared image frames acquired within the period of the pseudo-random sequence current, and based on the infrared image sequence, an electroluminescence image of the photovoltaic module is obtained, including: Extract the grayscale value sequence of each pixel in the infrared image frame across multiple consecutive infrared image frames; An electroluminescent image is obtained by performing cross-correlation calculations on the grayscale value sequence of each pixel on the infrared image frame and the modulation sequence of the pseudo-random sequence current.

[0014] In one feasible implementation, based on the electroluminescent image, the defect detection result of the photovoltaic module is obtained, including: The electroluminescent image is input into a pre-constructed defect classification and detection model to obtain the defect detection result output by the defect classification and detection model; wherein, the defect classification and detection model is trained based on a first deep learning model, and the defect detection result includes defect location, defect type, and damage degree; the first deep learning model is a convolutional neural network, a U-Net network, or a generative adversarial network.

[0015] In one feasible implementation, after acquiring the electroluminescence image of the photovoltaic module, the method further includes: The electroluminescent image is input into a pre-constructed image denoising and enhancement model to obtain an electroluminescent image after denoising processing by the image denoising and enhancement model; wherein, the image denoising and enhancement model is trained based on a second deep learning model; the second deep learning model is a convolutional neural network, U-Net or generative adversarial network, and the training data includes sample pairs superimposed with simulated strong light background noise.

[0016] In one feasible implementation, the method further includes: Monitor the loop response waveform of the encoded current excitation signal; wherein the loop response waveform is a voltage response waveform and / or a current response waveform; Based on the dynamic impedance spectrum or high-frequency noise characteristics of the loop response waveform, determine whether the photovoltaic module is abnormal. If the photovoltaic module malfunctions, the input of the coded current excitation signal to the photovoltaic module shall be stopped.

[0017] Compared with the prior art, this application has the following advantages: This application provides a photovoltaic module defect detection system and method for daytime environments, as well as an inspection system. This system enables direct electroluminescence imaging of photovoltaic modules under direct sunlight outdoors, eliminating the need for a darkroom or waiting until night. It obtains high-quality electroluminescence images even against strong sunlight, thus achieving daytime detection of hidden defects in photovoltaic modules. This significantly improves the convenience and timeliness of photovoltaic module defect detection in daytime environments. It is applicable to various scenarios such as photovoltaic module production quality inspection and photovoltaic power plant operation and maintenance inspection, and is of great significance for improving photovoltaic module quality control and power plant operation and maintenance levels, with broad application prospects. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] Figure 1 This application provides a schematic diagram of a photovoltaic module defect detection system for daytime environments. Figure 2 This application provides a schematic diagram of a photovoltaic module defect inspection system for daytime environments. Figure 3 A flowchart illustrating a method for detecting defects in photovoltaic modules under daytime conditions, provided as an embodiment of this application; Figure 4 A flowchart illustrating a method for acquiring an electroluminescent image provided in an embodiment of this application; Figure 5 A flowchart illustrating another method for acquiring electroluminescent images provided in an embodiment of this application; Figure 6 A flowchart illustrating an electrical fault detection method provided in an embodiment of this application; Figure 7 This is a schematic diagram of an electroluminescent image provided in an embodiment of this application.

[0020] Explanation of reference numerals in the attached figures: 100-Defect Detection System; 101 - High-voltage excitation module; 102 - Image acquisition module; 103 - Image Processing Module; 104 - Defect Detection Module; 200 - Photovoltaic modules; 300 - Mobile Platform.

[0021] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0022] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0023] Photovoltaic module defect inspection is a crucial step in ensuring the efficient and stable operation of photovoltaic power generation systems. Its role is to accurately identify potential defects within the modules, such as microcracks, broken grids, poor soldering, black cores, and PID degradation, through sophisticated testing technologies (such as electroluminescence (EL) testing and photoluminescence (PL) testing). This process not only prevents defective modules from causing problems like hot spot effects and power degradation during long-term outdoor operation, but also reduces subsequent maintenance costs, extends module lifespan, and increases the overall power generation of the system.

[0024] Electroluminescence (EL) testing is a non-destructive technique for detecting internal defects in photovoltaic modules. It involves forward biasing the module to excite it and then using an infrared camera to image the light, revealing hidden cracks, failures, and other defects. However, traditional EL testing typically requires darkness to achieve a sufficiently high signal-to-noise ratio. In large-scale photovoltaic power plants, this means either shading the modules during the day, disassembling them and placing them in a dark room, or waiting for nighttime illumination for testing, resulting in high environmental requirements and low efficiency. Photoluminescence (PL) testing is a non-contact technique that uses light to excite photovoltaic materials to generate fluorescence signals and analyzes the spectral characteristics to assess material defects and carrier recombination mechanisms. However, PL testing requires a high-power laser source to excite the module, leading to bulky equipment, complex water-cooling, and inconvenient operation for large-scale outdoor applications.

[0025] Therefore, there is currently a lack of a detection method that can efficiently detect defects in photovoltaic modules even in daytime environments.

[0026] To address the aforementioned technical issues, this application proposes a photovoltaic module defect detection system and method, as well as an inspection system, for daytime environments. This system enables daytime detection of hidden defects in photovoltaic modules, significantly improving the convenience and timeliness of defect detection in daytime environments.

[0027] The following detailed description, through specific embodiments, illustrates the technical solutions of the photovoltaic module defect detection system and method under daytime conditions, and the inspection system provided in this application. It should be noted that the following embodiments may exist independently or in combination; for identical or similar content, further description will not be repeated in different embodiments.

[0028] Figure 1 This application provides a schematic diagram of the structure of a photovoltaic module defect detection system 100 under daytime conditions, as shown in the embodiments of this application. Figure 1 In some embodiments, the photovoltaic module defect detection system 100 includes a high-voltage excitation module 101, an image acquisition module 102, an image processing module 103, and a defect detection module 104. The high-voltage excitation module 101 is used to input a coded current excitation signal based on code domain modulation to at least one photovoltaic module 200 to excite the photovoltaic module 200 to generate a time-varying electroluminescence signal. The coded current excitation signal is an alternating current or a pseudo-random sequence current. The image acquisition module 102 is used to continuously acquire infrared images of the photovoltaic module 200 to obtain an infrared image sequence. The image processing module 103 is used to obtain an electroluminescence image of the photovoltaic module 200 based on the infrared image sequence. The defect detection module 104 is used to obtain the defect detection result of the photovoltaic module 200 based on the electroluminescence image.

[0029] In this embodiment, during the detection process, the high-voltage excitation module 101 sends an coded current excitation signal to the photovoltaic module 200 through an innovative current modulation method. At the same time, the image acquisition module 102 acquires infrared images of the photovoltaic module 200, and the image processing module 103 processes the acquired infrared image sequence to extract the electroluminescence defect signal of the photovoltaic module 200 and generate a clear electroluminescence image. Finally, the electroluminescence image is used to detect defects such as microcracks and cell failures in the solar photovoltaic module 200.

[0030] In this design, the high-voltage excitation module 101 sends an coded current excitation signal to the photovoltaic module under test 200 as an alternating current or pseudo-random sequence current based on code domain modulation, thus replacing the traditional unipolar square wave phase-locked modulation with a novel excitation method. Unlike traditional unipolar on-off modulation, alternating current modulation includes alternating switching between forward excitation current and zero current (background state) within one cycle to avoid reverse voltage damage to the module. Pseudo-random sequence modulation generates a seemingly random pulse sequence with broadband spectral characteristics over a longer time window, but statistically, the number of occurrences of forward and reverse excitation is balanced. This excitation design effectively suppresses various low-frequency noises and ambient light drift, improves the system's ability to resist strong light interference, and allows for the extraction of electroluminescent images of photovoltaic module defects even in daytime environments. Simultaneously, by controlling the average input power, the risk of continuous module heating is reduced.

[0031] The photovoltaic module defect detection system 100 of this application can directly perform electroluminescence imaging on photovoltaic modules 200 under direct sunlight outdoors, without the need to build a darkroom or wait until night. It can obtain high-quality electroluminescence images under strong sunlight, thereby realizing daytime detection of hidden defects in photovoltaic modules 200. This greatly improves the convenience and timeliness of photovoltaic module 200 defect detection in daytime environments. It is applicable to many scenarios such as photovoltaic module 200 production quality inspection and photovoltaic power station operation and maintenance inspection. It is of great significance for improving the quality control of photovoltaic modules 200 and the operation and maintenance level of power plants, and has broad application prospects.

[0032] In some embodiments, the high-voltage excitation module 101 includes a programmable high-voltage DC power supply and a spread spectrum modulation control unit; the spread spectrum modulation control unit is used to control the programmable high-voltage DC power supply so that the programmable high-voltage DC power supply outputs an coded current excitation signal.

[0033] The high-voltage excitation module 101 includes a programmable high-voltage DC power supply and its spread spectrum modulation control unit. During the detection process, the high-voltage excitation module 101 sends a pre-set coded current excitation signal to the photovoltaic module 200 under test under the control of the spread spectrum modulation control unit, thereby generating an electroluminescent effect.

[0034] Specifically, the programmable high-voltage DC power supply has a rated output voltage range of 1000V to 2000V and an output current of up to 20A, enabling it to provide forward current to the entire string of photovoltaic modules 200. This allows the defect detection system 100 to be directly applied to high-voltage scenarios such as centralized photovoltaic power plants, avoiding the hassle of disassembling and inspecting each module individually. The programmable high-voltage DC power supply is internally or externally connected to a spread spectrum modulation control unit. Control signals are sent from the system host to control the programmable high-voltage DC power supply to generate a specific waveform output current sequence. Specifically, the spread spectrum modulation control unit can output an alternating current waveform or a pre-set pseudo-random sequence current to apply a positive / zero excitation current to the photovoltaic module 200 under test in a periodic or quasi-periodic manner.

[0035] Specifically, the programmable high-voltage DC power supply also features overvoltage and overcurrent protection and insulation safety design to ensure safe and reliable high-voltage operation on site. The high-voltage excitation module 101 has a component safety protection mechanism: it sets safety thresholds for the output forward current amplitude and current ramp-up slope (di / dt), and monitors the port voltage in real time to ensure it remains within the unipolar range, preventing accidental conduction or overheating of the photovoltaic module 200 bypass diode due to reverse bias; furthermore, the high-voltage excitation module 101 can receive temperature or voltage feedback signals and adaptively adjust the duty cycle, symbol rate, or amplitude of the pseudo-random sequence to maintain the component within the safe operating area (SOA).

[0036] In some embodiments, the image acquisition module 102 includes an infrared detection device for acquiring infrared images of the photovoltaic module 200.

[0037] During the detection process, after the high-voltage excitation module 101 sends an coded current excitation signal to the photovoltaic module 200, the infrared detection device begins image acquisition. This infrared detection device is connected to the image processing module 103 via a data interface to achieve high-speed image acquisition. Specifically, the infrared image is a short-wave infrared (SWIR) image.

[0038] Specifically, infrared detection devices capable of covering the target electroluminescence spectrum can be used, including but not limited to indium gallium arsenide (InGaAs) infrared cameras, colloidal quantum dot (CQD) infrared cameras, mercury cadmium telluride (HgCdTe) (MCT), germanium-silicon (Ge-on-Si), or other devices capable of imaging in the 900–1700 nm range. Preferably, their peak response is located in the 1100–1200 nm range to match the EL peak of crystalline silicon modules. For different cell systems (such as heterojunction, perovskite / tandem, etc.), devices and filter sets matching their EL peaks can be selected accordingly.

[0039] For example, taking a high-sensitivity indium gallium arsenide (InGaAs) infrared camera and its matching dedicated detection lens as an example, the core of the indium gallium arsenide (InGaAs) infrared camera, as a photoelectric imaging device based on semiconductor materials, is the InGaAs absorption layer. This material is composed of a compound of indium (In), gallium (Ga) and arsenic (As), and has adjustable bandgap characteristics. It can efficiently absorb short-wave infrared (SWIR, wavelength range 900-1700nm) photons, so that the spectral response of the infrared camera can cover the electroluminescence band of photovoltaic silicon cells and is not sensitive to the visible light band.

[0040] In some embodiments, the detection lens of the infrared detection device is provided with a narrowband optical filter or a long-pass optical filter.

[0041] Among them, a narrow-band or long-pass optical filter is configured in front of the detection lens of the infrared detection device to filter out the background of sunlight to the greatest extent.

[0042] Specifically, narrowband optical filters allow only a specific narrow band of light to pass through (the passband width is typically less than 5% of the center wavelength), while effectively blocking light of other bands. Since electroluminescence signals are relatively weak and easily interfered with by ambient light and light reflected from the battery itself, narrowband filters significantly improve the signal-to-noise ratio by allowing only the target band (e.g., center wavelength ±20nm) of light to pass through while suppressing other wavelengths. This enables the defect detection system 100 to clearly capture differences in luminescence intensity caused by micro-defects in the battery (such as microcracks or broken grids) or material inhomogeneities.

[0043] Specifically, long-pass optical filters allow light with wavelengths longer than a set value to pass through while blocking light with wavelengths shorter than that value. During daytime inspections in strong sunlight, short-wavelength components of sunlight (such as ultraviolet and visible light) can mask the weak electroluminescence signal from the battery. Long-pass filters (e.g., with a cutoff wavelength of 700nm) effectively block these short-wavelength interferences, retaining only the long-wavelength emission emitted by the battery itself (such as in the near-infrared region), thereby improving the sensitivity of defect detection.

[0044] In some embodiments, the high-voltage excitation module 101, the image acquisition module 102, the image processing module 103, and the defect detection module 104 are uniformly controlled by the system host.

[0045] According to one embodiment of this application, the system host is equipped with a Global Navigation Satellite System (GNSS) timing module or a Precision Time Protocol (PTP) module; the system host uses the unified time reference provided by the GNSS timing module or PTP module to perform microsecond-level hard synchronization triggering of the current output time of the high-voltage excitation module and the exposure time of the image acquisition module.

[0046] According to one embodiment of this application, in order to enable synchronous operation between the various modules and ensure the precise alignment of the current waveform of the high-voltage excitation module 101 with the exposure window of the image acquisition module 102, the system adopts a dual synchronization strategy of "hardware synchronization + delayed triggering".

[0047] Specifically, hard synchronization uses GPS / PTP or LoRa to send a unified time reference or trigger signal.

[0048] Throughout the detection process, all modules achieve precise coordination through a hardware synchronization scheme, ensuring a high degree of synchronization between the coded current excitation signal and the signal acquisition process of the image acquisition module 102. The system host sends trigger commands to the image acquisition module 102 via a high-speed digital interface, ensuring that it begins continuous image acquisition at a predetermined frame rate. Simultaneously, it controls the high-voltage excitation module 101 to output a current sequence according to a phase-synchronized modulation rhythm via a communication interface. This fully hardware synchronization scheme avoids the limitations of traditional soft synchronization, reduces system hardware interfaces and wiring requirements, and makes the collaborative work of multiple modules more precise and reliable, providing a foundation for remote and wireless control of the system.

[0049] To ensure high-precision synchronization among all modules within the system, the system host employs a high-precision clock and uses a GPS hard synchronization scheme to timestamp-align the image acquisition module 102 and the high-voltage excitation module 101. In a single-module system, since all modules are coordinated and controlled by the same main control software, nanosecond-level synchronization accuracy can be achieved, meeting the requirements of phase-locked loop (PLL) or related calculations. In multi-module expansion scenarios, multiple modules or even multiple detection systems are connected via a network. To ensure synchronization, a dual-channel synchronization mechanism is introduced: on the one hand, the Precision Time Protocol (PTP / IEEE 1588) is used to synchronize time between the system hosts, obtaining a sub-millisecond-level global synchronization clock; on the other hand, the high-precision synchronization signal provided by GPS hard synchronization ensures accurate synchronization between spatially dispersed devices. This GPS hard synchronization scheme achieves time synchronization by directly receiving GPS signals, avoiding the effects of network latency and calculation errors, providing more reliable and accurate synchronization, and is particularly suitable for collaborative operation of large-scale devices. Compared with traditional phase-locked amplifier analog reference signals, the GPS hard synchronization scheme simplifies system hardware, reduces costs, and improves system reliability and scalability, meeting the needs of large-scale deployments.

[0050] Optionally, in large-scale photovoltaic power plants, there are often hundreds or thousands of photovoltaic module defect detection systems that need to be inspected. To improve efficiency, multiple sets of the photovoltaic module defect detection system 100 of this embodiment can be used in parallel. The photovoltaic module defect detection system 100 supports a "multi-machine parallel operation" mode, that is, when multiple detection systems are running simultaneously in the same photovoltaic power plant area, the high-voltage excitation module 101 of each system is configured to output pairwise orthogonal pseudo-random sequence codes; that is, the spread spectrum modulation control unit is configured with multiple mutually orthogonal pseudo-random sequence code groups, and different photovoltaic module defect detection systems are assigned different orthogonal pseudo-random sequence codes to suppress signal crosstalk from other detection systems when the image processing module performs cross-correlation calculations; for the cluster inspection needs of large-scale photovoltaic power plants, orthogonal code division (CDMA) technology is adopted. Code domain spread spectrum modulation has spectral characteristics similar to white noise. Different devices are assigned orthogonal modulation sequences. Even if multiple photovoltaic module defect detection systems 100 excite and capture images of adjacent strings, the image processing module 103, during demodulation / despreading, utilizes the zero cross-correlation characteristic of orthogonal codes to suppress signal crosstalk from other detection systems. It only performs correlation operations on the sequence codes assigned to it, effectively suppressing signal crosstalk between devices, supporting multi-machine collaboration, and significantly improving the operation and maintenance efficiency of large-scale power plants.

[0051] Specifically, considering the physical response time (RiseTime, e.g., 2ms) of high-power power supplies (e.g., 30kW level) during current transitions, the coordination control unit will forcibly delay by a preset time window Δt (e.g., 3ms) after sending the current command to ensure that the exposure window of the infrared detection device falls entirely within the stable plateau period of the current waveform. The image acquisition module 102 is configured to use gated exposure or global shutter mode to sample within the "stable characteristic window" of the coded current. After the current stabilizes and enters the "flat-top region," the image acquisition module 102 is triggered to continuously acquire SWIR images at a fixed frame rate (e.g., 50fps). This employs a "delay trigger" strategy to address the waveform distortion problem caused by the current settling time (RiseTime) of the high-power power supply. Finally, GNSS / PTP timestamps or hardware trigger signals are used to ensure that each frame of the image strictly corresponds to the current state of the coded current excitation signal.

[0052] For example, several photovoltaic module defect detection systems 100 are placed in different areas of a photovoltaic array, simultaneously applying current and imaging multiple strings of modules. To avoid mutual interference and ensure that each system independently acquires high-quality data, synchronization and coordination of multiple systems are required. A GPS hard synchronization scheme is well-suited for expansion. The system hosts of each system are precisely synchronized via GPS hard synchronization signals, ensuring clock uniformity for each device. Through GPS hard synchronization, the clock accuracy of all systems can reach the nanosecond level, ensuring that detection systems in different areas can accurately trigger at the same time, avoiding the impact of network latency and synchronization errors.

[0053] In this scheme, each system host is interconnected via wired or wireless network, with one host selected as the master controller, or distributed clock synchronization is achieved using network protocols. Through precise time synchronization, each system host can simultaneously trigger its controlled high-voltage excitation module 101 and image acquisition module 102 to start working according to a predetermined schedule. When using the photovoltaic module defect detection system 100 in a daytime environment with pseudo-random modulation, orthogonal sequence codes can also be assigned to different systems. The assigned orthogonal sequence codes are mathematically orthogonal to each other, with an inner product of zero. The orthogonal codes can be distinguished from each other during demodulation, thereby preventing signal crosstalk. After the orthogonal sequence codes are assigned, the high-voltage excitation module 101 sends an coded current excitation signal to the photovoltaic module 200 according to the assigned orthogonal sequence codes to begin the detection operation.

[0054] When multiple image acquisition modules 102 need to cover the same area, they can exchange real-time status via wireless communication to ensure perfect alignment of their exposure and code-domain-based spread spectrum modulation. Through these methods, the system can be expanded into a multi-node collaborative detection network, enabling comprehensive defect detection scanning of large-scale photovoltaic arrays in a short time. This multi-module solution has good scalability, allowing for flexible addition or reduction of equipment as needed to meet the detection requirements of large-scale photovoltaic power plants.

[0055] Specific on-site deployment and parameter settings: Equipment Installation: Before testing begins, connect the positive and negative terminals of the 200 strings of photovoltaic modules to the output of the high-voltage excitation module 101 via cables. Align the image acquisition module 102 (mounted on the mobile platform 300 or tripod) with the module surface and adjust the lens field of view to cover the entire module or multiple modules.

[0056] Communication connection: The image acquisition module 102 is connected to the system host via a low-latency wireless link (for UAV scenarios) or a wired interface (for ground scenarios). The control interface of the high-voltage excitation module 101 is also connected to the system host to achieve unified scheduling.

[0057] Safety isolation: Ensure that the 200 strings of photovoltaic modules under test are in an open circuit state with no current flowing, and are physically isolated from the inverter / combiner box.

[0058] Camera parameters: The system host automatically sets the operating parameters of the SWIR camera. For strong daylight backgrounds, the exposure time is set to a short exposure mode (e.g., <5ms) to avoid pixel saturation, and a narrow-band filter with a center wavelength of 1150nm is used to physically suppress 99% of visible light interference.

[0059] Figure 2 A schematic diagram of a photovoltaic module 200 defect inspection system provided in this application embodiment is shown below. Figure 2 In some embodiments, the photovoltaic module 200 defect inspection system includes the photovoltaic module defect detection system 100 described above, and a mobile platform 300; the image acquisition module 102 of the photovoltaic module defect detection system 100 is mounted on the mobile platform 300.

[0060] In this embodiment, considering the need for large-scale on-site inspection, the system is designed to be mobile and modular. All components are mounted in a robust frame or enclosure, equipped with necessary protection to withstand outdoor environments and vibrations during movement. The system supports installation on a mobile platform 300, such as a drone, tracked inspection robot, or automated guided vehicle. The robot carries the system between the photovoltaic module arrays 200, enabling automated inspection of multiple modules. The mobile platform 300 moves to the front of each photovoltaic module 200 to stop and take pictures, or simultaneously takes pictures while moving at low speed. A GPS hard synchronization mechanism coordinates current excitation and camera exposure imaging, sequentially completing the inspection of the entire row of modules. This gives the system flexible synchronous control and mobile deployment capabilities, making it suitable for on-site inspection of large-scale photovoltaic power plants.

[0061] Furthermore, this system also provides interfaces for integration with drones or rail inspection vehicles. The image acquisition module 102 and wireless image transmission equipment can be shrunk and mounted on a drone gimbal to capture electroluminescent images of the components from the air. Simultaneously, a ground vehicle tows a high-voltage power supply to inject current into the 200 strings of photovoltaic modules under test, and the two work together via wireless synchronization signals. Alternatively, the system can be installed on a trolley with a pre-set track between rows of photovoltaic panels to achieve automatic reciprocating inspection along a fixed route. These expansion methods enable the defect monitoring system to adapt to the defect detection needs of large-area photovoltaic arrays in different scenarios, thereby achieving automated and integrated defect inspection operations and greatly improving the efficiency of on-site inspection.

[0062] Specifically, the mobile platform 300 can employ a ground-based tracked / wheeled robot, integrating the system onto a small tracked robot. The robot's lifting mechanism aligns the camera with the component under test. The robot automatically travels along the row of components, stopping at predetermined locations where the system excites and photographs the current photovoltaic module 200 before moving to the next location. Throughout the process, the robot plans its path and positions itself using its navigation system, while the system receives control commands and uploads image data via a wireless network. Due to the robot's limited power supply, high-voltage power can be provided by a battery-powered inverter booster carried by the robot, or continuous power can be obtained from ground power via a tow cable. Field tests show that the defect detection system 100 mounted on this robot can inspect hundreds of photovoltaic modules 200 per day, far exceeding the efficiency of manual operation.

[0063] Specifically, the mobile platform 300 can utilize unmanned aerial vehicles (UAVs) for rapid inspection of large-area photovoltaic power plants. The image acquisition module 102 of the defect detection system 100 can be mounted on a multi-rotor or fixed-wing UAV. The UAV hovers above the modules to capture infrared images, while ground personnel use a portable high-voltage power supply to excite the solar photovoltaic modules 200 string by string. The advantages of the UAV-based solution are its extremely wide coverage and fast response speed, making it particularly suitable for photovoltaic power plants with complex terrain or those difficult to access manually.

[0064] Specifically, the mobile platform 300 can be a fixed-track robot. In some large photovoltaic power plants, tracks or guide vehicles for operation and maintenance may be pre-installed in front of the photovoltaic arrays. The defect detection system 100 can be installed on the track vehicle, moving along the fixed track to perform defect detection on each of the photovoltaic modules 200 along the line. This method provides stable movement and precise positioning, and charging and data aggregation stations can be set at the end of the track to achieve long-term automatic operation.

[0065] In summary, regardless of the mobile platform 300, the defect detection system 100 reserves standard mechanical and communication interfaces for quick installation and integration with upper-level control systems. For example, via REST API or industrial Ethernet protocol, the operation and maintenance center can remotely issue inspection tasks to robots or drones, which then call the system's interfaces to perform defect detection and send the results back to the database. This highly automated and intelligent inspection method can further improve the efficiency and precision of photovoltaic power plant operation and maintenance.

[0066] In this way, various deployment scenarios for the photovoltaic module defect detection system 100 can be obtained: Scenario 1: Air-Ground Collaborative Inspection (for mountainous, water-based, or complex terrain) In this scenario, the mobile platform 300 is an industrial-grade unmanned aerial vehicle (UAV). Specifically, the image acquisition module 102 (including a SWIR camera and edge computing unit) is mounted on the UAV, utilizing its high maneuverability and field of view to perform aerial imaging of the components. The heavier high-voltage excitation module 101 is deployed at a fixed ground station or moves with a ground vehicle, connected to the photovoltaic string via cables. The two achieve logical coordination and timing alignment through a low-latency wireless communication link (such as LoRa or a dedicated data radio). The UAV hovers or cruises at low speed above each photovoltaic module 200 to take pictures, while the ground power supply outputs coded current, achieving efficient "shooting from the air, powering from the ground" operation.

[0067] Scenario 2: Ground / Water Surface Mobile Inspection (for flat land power stations or photovoltaic rows) In this scenario, the mobile platform 300 is a ground mobile robot (such as a tracked vehicle, AGV, or photovoltaic cleaning robot) or a surface unmanned vessel. Specifically, the image acquisition module 102 is mounted on top of the robot via a lifting rod or gimbal to obtain the best shooting angle. In this case, the deployment of the high-voltage excitation module 101 is more flexible: Trailer type: For tracked vehicles with strong load capacity, the high-voltage excitation module 101 can be directly carried or towed by the robot and move with the vehicle to achieve "independent operation of a single vehicle".

[0068] Split-type: For lightweight robots, the power supply remains fixed at the end of the string, and the robot is only responsible for carrying the camera to move between rows and take pictures. Compared with drones, ground robots have the advantages of smoother operation, longer battery life, and the ability to pause for multi-angle detailed shooting, making them particularly suitable for high-precision acceptance testing in flat terrain.

[0069] Whether it's air-to-ground coordination or ground inspection, the system employs the aforementioned "hard synchronization + delayed triggering" mechanism. The coordination control unit ensures that the coded current output by the ground / vehicle power supply precisely matches the camera's exposure window on a microsecond-level time axis at the instant the mobile platform moves or hovers, thereby acquiring high-quality, motion-free EL defect images while in motion.

[0070] Optionally, when deploying on the mobile platform 300, in order to reduce the transmission latency of massive amounts of raw image data, the system can also adopt an "edge-cloud collaborative" computing architecture: Edge computing devices (device side): Deployed on the mobile platform 300, they utilize embedded AI chips (such as NPUs) to perform image preprocessing and preliminary analysis tasks. They perform inter-frame registration and correlation operations on the acquired SWIR image sequences, remove the effects of ambient light and device noise, generate a single high-quality EL image, and output image quality evaluation metrics (including signal-to-noise ratio (SNR), sharpness, and alignment). The quality of the generated EL image (sharpness, exposure, alignment) is evaluated in real time. If excessive drone shaking and image blurring occur due to gusts of wind, the edge device immediately triggers a "retake" command to ensure the validity of the uploaded data. The edge computing device can also apply deep learning models to perform preliminary image analysis, detect the presence of obvious defects, and generate preliminary defect candidate regions and classification results. Finally, only the processed EL result image and preliminary defect coordinates are transmitted back via wireless networks (such as 5G Wi-Fi, LTE, 5G).

[0071] Cloud servers (cloud-side): Deployed in remote data centers, they receive data transmitted from the edge, utilize powerful computing capabilities to run complex deep learning models (such as U-Net and ResNet), perform refined defect segmentation, classification, and lifespan prediction, and generate a final inspection report. This final diagnostic report is then transmitted back to the user terminal for further processing and decision-making by maintenance personnel. In this scenario, cloud servers can integrate more computing resources to perform deeper defect identification and classification. Furthermore, cloud servers can not only perform more complex defect detection but also further optimize defect detection algorithms and improve detection accuracy through historical data and big data analysis.

[0072] The cloud computing layer (deployed on a remote server) transmits the "structured data" (EL result image + defect coordinates + electrical waveform clips) after compression and cleaning at the edge to the cloud via 5G / 4G or a private network. The cloud server then leverages its powerful computing capabilities to perform in-depth value mining. 1. Refined Defect Diagnosis: Run large-scale deep learning models (such as large visual models based on Transformer) to perform sub-pixel-level segmentation and classification of minute hidden cracks and broken gates.

[0073] 2. Lifespan Prediction and Asset Assessment: Based on historical inspection data (EL / IV), a full lifecycle health record for photovoltaic modules is established. Through time-series prediction algorithms, defect evolution trends are extrapolated, and the remaining lifespan and future power degradation curves of the modules are assessed, providing a quantitative basis for asset transactions and financial loss assessment of photovoltaic power plants.

[0074] 3. Closed-loop feedback of results: Diagnostic conclusions from the cloud can be pushed to the handheld terminals of on-site maintenance personnel in real time, guiding them to conduct targeted review or replacement. At the same time, edge computing devices can be upgraded via OTA based on the latest model parameters issued from the cloud, continuously evolving their detection capabilities.

[0075] The system constructs a fully automated closed loop encompassing "hardware spread spectrum sensing + real-time edge despreading + cloud-based AI diagnostics." The acquired raw shortwave infrared (SWIR) sequences are registered and despread at the edge, and the generated intermediate EL images are transmitted to the cloud or ground station for deep learning enhancement and classification. For the inspection of large photovoltaic power plants, this architecture enables the system host to output a comprehensive diagnostic report, including defect type, location coordinates, and electrical safety status, within milliseconds after each photovoltaic module is photographed (200 images are captured), completely replacing inefficient manual screening.

[0076] The introduction of edge computing devices makes the entire inspection process more intelligent and real-time, enabling preliminary data processing and defect detection on a mobile platform 300. This reduces reliance on cloud servers, lowers latency and bandwidth consumption, and improves system response speed and stability. Furthermore, edge computing devices can be customized to meet specific on-site needs, satisfying the defect detection requirements of different types of photovoltaic power plants. This hardware-software synergy, combining edge computing with cloud computing, ensures the system maintains high efficiency, reliability, and flexibility in defect detection of large-scale photovoltaic arrays.

[0077] Figure 3 This application provides a flowchart illustrating a method for detecting defects in photovoltaic modules under daytime conditions, as illustrated in the embodiments of this application. Figure 3 In some embodiments, the photovoltaic module defect detection method includes the following steps: S301, input a coded current excitation signal based on code domain modulation to at least one photovoltaic module to excite the photovoltaic module to generate a time-varying electroluminescent signal; wherein, the coded current excitation signal is an alternating current or a pseudo-random sequence current.

[0078] In this process, a non-constant coded current based on code domain modulation is injected into the photovoltaic module through a high-voltage excitation module. The coded current excitation signal is preferably a pseudo-random sequence (PRBS) or an orthogonal sequence code with specific autocorrelation characteristics. Through this spread spectrum modulation, the electroluminescent signal generated by the photovoltaic module has a unique identifier in the code domain, thus possessing extremely high recognizability under strong sunlight and multi-machine operation interference.

[0079] Specifically, a coded current excitation signal based on code domain modulation is input to at least one photovoltaic module through a high-voltage excitation model.

[0080] S302 continuously acquires infrared images of photovoltaic modules to obtain infrared image sequences.

[0081] Optionally, after receiving the synchronization trigger signal, the image acquisition module executes a preset delay to avoid the current ramp-up period of the high-voltage excitation module, and then starts exposure during the stable plateau period of the current waveform.

[0082] Specifically, the image acquisition module continuously acquires infrared images of the photovoltaic modules to form an infrared image sequence.

[0083] S303 acquires electroluminescent images of photovoltaic modules based on infrared image sequences.

[0084] Specifically, the acquired infrared image sequence is first subjected to mathematical demodulation processing (such as the aforementioned difference operation or cross-correlation operation) to extract the preliminary electroluminescent image of the photovoltaic module from the strong background.

[0085] Although mathematical demodulation can eliminate most linear noise, under strong sunlight, the original image still retains a large amount of nonlinear residual noise and image artifacts caused by factors such as cloud changes and camera thermal drift. These nonlinear artifacts cannot be effectively eliminated by traditional averaging or difference operations, which constitutes a bottleneck in the current technology from mathematical demodulation to high-definition image reconstruction. Therefore, this embodiment introduces deep learning enhancement techniques to overcome this technical obstacle.

[0086] Meanwhile, in order to further improve the signal-to-noise ratio and visual quality of the image, this embodiment also introduces deep learning enhancement technology after obtaining the preliminary electroluminescent image; the specific operation is as follows: the preliminary electroluminescent image after mathematical despreading is input into the pre-constructed image denoising and enhancement model to obtain a high-definition electroluminescent image after denoising processing.

[0087] The image denoising and enhancement model is obtained by training a deep learning network, preferably using a convolutional neural network (CNN), U-Net network, or generative adversarial network (GAN) architecture.

[0088] To enable the model to operate under strong sunlight, this embodiment employs a specific "data synthesis training strategy": Constructing the Ground Truth: Collect a large number of high signal-to-noise ratio EL images taken in a standard darkroom environment as the ideal target; Simulated noise injection: Simulated strong sunlight background, cloud shadow changes and sensor thermal noise are superimposed on the above ground value image to generate the corresponding "noisy input set"; Supervised training: The deep learning network is trained in a supervised manner using the paired data (noisy input - darkroom truth) to learn to separate and recover the texture of weak EL signals from strong interference.

[0089] This step effectively removes nonlinear residual noise that is difficult to eliminate through mathematical operations, significantly improving detection accuracy and enabling the final output image to clearly show micro-cracks and broken grid features for use by subsequent defect detection modules.

[0090] In this process, the infrared image sequence is processed to obtain the electroluminescent image of the photovoltaic module.

[0091] Optionally, after acquiring the electroluminescence image of the photovoltaic module, the electroluminescence image is input into a pre-built image denoising and enhancement model to obtain the electroluminescence image after denoising processing by the image denoising and enhancement model; wherein, the image denoising and enhancement model is trained based on a second deep learning model.

[0092] In the process of image processing and signal extraction, the system can also introduce a convolutional neural network or GAN model to train a model for denoising and extracting defect features from the original electroluminescent image containing strong noise, so as to improve the reliability and intelligence of defect detection.

[0093] Optionally, to further improve the visual quality of the image, after obtaining the electroluminescent image of the photovoltaic module through relevant calculations, the system introduces a "deep learning enhancement" step: the mathematically despread electroluminescent image is input into a pre-constructed image denoising and enhancement model. This model is trained based on a deep learning network (preferably a convolutional neural network (CNN), a residual network (ResNet), or a generative adversarial network (GAN).

[0094] The specific training strategy involves constructing a "denoising autoencoder" or a "generative adversarial network." A large number of high signal-to-noise ratio (SNR) EL images captured in a standard darkroom environment are collected as "ground truth," and corresponding "noisy input sets" are generated by superimposing simulated strong sunlight noise, cloud shadows, and sensor thermal noise. This paired data is used to supervise the training of the model, enabling it to recover the texture of fine cracks from residual noise. This step effectively removes nonlinear noise that is difficult to eliminate through mathematical operations, significantly improving detection accuracy.

[0095] Specifically, the image denoising and enhancement model learns the noise distribution pattern in the image and performs denoising, texture enhancement, and contrast stretching operations to generate a high-resolution electroluminescent image for final diagnosis, which can then be used for subsequent defect detection.

[0096] S304, based on electroluminescent images, obtains defect detection results for photovoltaic modules.

[0097] Optionally, the electroluminescent image is input into a pre-built defect classification and detection model to obtain the defect detection results output by the defect classification and detection model; wherein, the defect classification and detection model is trained based on a first deep learning model, and the defect detection results include defect location, defect type, and damage degree.

[0098] In terms of defect identification, a defect detection and classification model can be trained. A convolutional neural network is constructed to analyze the input electroluminescent image and output the identified defect type, its location in the image, and the extent of damage. This reduces reliance on human experience and improves the digitalization level of operation and maintenance.

[0099] Specifically, for defect detection and classification models, deep learning models can be trained using a dataset of electroluminescent images labeled with defects, enabling the deep learning models to learn to identify defect patterns from the distribution of luminescence intensity.

[0100] Specifically, the first deep learning model is preferably an image processing network such as a multiplicative neural network (CNN), YOLO, or U-Net; this defect classification and detection model can automatically extract spatial morphological features from the image and classify and accurately locate the defects based on their topological structure (hidden cracks, broken grids, black cores, process contamination, etc.).

[0101] In this embodiment, a code-domain modulated current electroluminescence detection, defect detection, defect classification detection model, and image denoising and enhancement model are connected in series. The acquired infrared image data is transmitted to the image denoising and enhancement model. First, the original electroluminescence image is denoised and enhanced, and then the denoised electroluminescence image is sent to the defect classification detection network, thereby realizing full automation from original image acquisition to final defect detection results. For the inspection of large photovoltaic power plants, the system host can immediately provide defect detection results and defect location after each photovoltaic module is photographed, which greatly saves the workload of defect detection compared to traditional manual screening and analysis methods.

[0102] This method employs code-domain-based spread spectrum modulation combined with a correlation algorithm, resulting in a high signal-to-noise ratio and strong anti-interference capability. It can significantly suppress interference from strong background light and environmental noise on the electroluminescence signal. Compared to traditional single-frequency phase-locked loop (PLL) methods, thanks to the broadband spectrum and autocorrelation characteristics of the pseudo-random sequence, this defect detection method is less sensitive to changes in illumination and can stably extract defect signals even under conditions of large light intensity fluctuations.

[0103] exist Figure 3 The photovoltaic module defect detection method shown below requires acquiring electroluminescence images. Figure 4 The technical solution for the above-mentioned photovoltaic module defect detection method further describes the content related to acquiring electroluminescent images.

[0104] Figure 4 This is a flowchart illustrating a method for acquiring electroluminescent images according to an embodiment of this application. (See attached diagram.) Figure 4 In some embodiments, the alternating current includes a periodically spaced forward current and a zero current, and the infrared image sequence includes a set of luminescent images and a set of background images. The set of luminescent images includes multiple luminescent image frames acquired during the forward current cycle, and the set of background images includes multiple background image frames acquired during the zero current cycle. The method for acquiring electroluminescent images includes the following steps: S401, perform frame-by-frame image registration and averaging on the luminous image set and the background image set respectively to obtain the average luminous image and the average background image.

[0105] During the detection process, infrared images are acquired by alternately applying a positive current (+I) and a zero current. The image with the positive current contains both electroluminescence signals and background signals, while the image with the zero current contains only background signals. The difference between the two can eliminate most of the background interference, thereby extracting the weak electroluminescence defect signals.

[0106] Specifically, the high-voltage excitation module, driven by the code domain-based spread spectrum modulation control unit, periodically switches its output state. The output state includes an excitation state and a background state.

[0107] Excitation state: A constant forward current (e.g., +10A) is output, causing the component to emit a strong light. The image captured by the infrared detector under forward current conditions. .in, Indicates an electroluminescent signal. Indicates sunlight noise. This indicates thermal noise.

[0108] Background state: Output zero current (0A) or a weak non-conducting voltage. Zero current avoids applying excessive reverse current, preventing accidental conduction or overheating damage to the photovoltaic module's bypass diodes, ensuring absolute safety during the detection process. Image acquired by the infrared detection device at zero current. This 'zero-current background acquisition' mode avoids the risk of bypass diode conduction that may be caused by applying reverse voltage, ensuring absolute safety in the detection process.

[0109] Modulation parameters: The symbol rate can be flexibly set according to the camera frame rate and power supply response speed, preferably within the range of 20–50Hz, to balance the signal-to-noise ratio and ambient light changes caused by cloud movement. The modulation frequency or code sequence rate can be flexibly set according to the frame rate of the infrared detector, and an appropriate frequency can be selected within the range of 20–50Hz to balance signal amplitude and environmental changes.

[0110] Based on the above, the specific process during the infrared image acquisition stage includes: the power supply periodically switching between two current amplitudes, +I and 0, to output alternating current. +I represents the forward current applied to the photovoltaic module string; for example, the current is set to 10A, while 0 represents zero current. The current waveform is a square wave, alternating between forward and zero current at a frequency of 25Hz, with each positive or zero pulse lasting 20ms. It should be noted that when zero current is applied, the photovoltaic module is in an unexcited state; due to the unidirectional conduction characteristic of photovoltaic diodes, the photovoltaic module does not produce electroluminescence.

[0111] While the high-voltage excitation module outputs the aforementioned alternating current, the image acquisition module continuously acquires short-exposure images at a frame rate of 50 frames per second. Through software synchronization design, the system host ensures that the infrared camera frame exposure is in phase with the current waveform. The system host records the timestamp of each frame from the camera and, based on the known power waveform period, marks the corresponding current phase for each image. The system host program calculates the time of the next current reversal and sends a trigger command to the infrared detection device with a slight delay. At the moment of +I and 0 switching, "luminescent" image frames and "background" image frames are acquired respectively. Each frame has been timestamped to ensure correspondence. After acquiring this alternating sequence of infrared images, the image processing module divides them into two groups according to the frame markings: electroluminescent frames under forward current (denoted as the luminescent image set: frames containing EL signals) and background frames under zero current (denoted as the background image set: frames containing only background noise).

[0112] For processing infrared image sequences, slight shaking during hovering / movement of drones or robots can cause edge ghosting when directly subtracting. Therefore, to prevent artifacts introduced during subtraction, the image processing module can use an image registration algorithm to align and superimpose all frames, and then average the aligned frames to improve the image signal-to-noise ratio. Specifically, the image processing module first uses an image registration algorithm to align all frames to a unified coordinate system, and then performs a weighted average of the aligned frames.

[0113] For example, suppose 25 frames of emission images and 25 frames of background images are acquired within a 1-second code period. Averaging these separately yields a high signal-to-noise ratio average emission image. and average background image Shot noise is significantly suppressed.

[0114] Image registration is the process of aligning multiple images through geometric transformations (such as translation, rotation, scaling, or perspective transformation). Its core lies in finding the spatial correspondences between images. Image registration provides a precise basis for spatial alignment, thereby eliminating image misalignment problems caused by differences in viewpoint, time, or device. This embodiment employs a hybrid registration strategy, and the specific image registration algorithms include: Coarse registration: Extract image feature points (such as battery cell edges and grid line intersections), and use the SIFT / ORB algorithm to calculate the homography matrix.

[0115] Fine registration: Based on coarse registration, optical flow is used to perform sub-pixel level fine-tuning to eliminate the effects of high-frequency jitter, and robust estimation (such as the RANSAC algorithm) is used to eliminate mismatched points.

[0116] Image averaging involves averaging (or weighted averaging) the corresponding pixel values ​​of multiple images. In spread spectrum imaging, this is equivalent to integration, which effectively improves the signal-to-noise ratio (SNR), gradually revealing hidden cracks and textures obscured by noise, preserving details, and reducing graininess. Furthermore, averaging can be used to eliminate random motion interference (such as jitter) or to generate background templates to separate foreground objects.

[0117] S402 performs differential processing on the average luminescence image and the average background image to obtain the electroluminescence image.

[0118] Among them, the averaged background image (Corresponding to current off / zero current state) From the average emission image Subtracting from (the corresponding current conduction state) yields the differential image. The calculation formula is as follows:

[0119] Where k represents a coefficient.

[0120] In ideal circumstances (coefficient) ), Includes solar background scattered light and sensor thermal noise and The corresponding components are basically the same. Through the above differential operation, the background noise is mathematically canceled out, retaining only the weak electroluminescent signal emitted by the component under forward bias. This eliminates the dominant solar background noise and thermal noise, thereby extracting the weak electroluminescent defect signal. The processed signal is then obtained. The images clearly show internal defects in the components: cracks appear as dark lines, and failed cells appear as dark spots. Finally, the image processing module... Adaptive gamma correction or pseudo-color enhancement is performed to facilitate further identification of defect types by human or AI models.

[0121] In this embodiment, a code-domain-based spread spectrum modulation technique is employed, which has a fundamental advantage in anti-interference compared to traditional techniques: 1. Resistant to ambient light drift: Even in scenarios where rapidly moving clouds cause drastic fluctuations in background light intensity, the signal can still be stably extracted through correlation calculations using the white noise statistical characteristics of pseudo-random sequences, without producing artifacts.

[0122] 2. Anti-crosstalk characteristics: Utilizing orthogonal code sequences, it supports multi-machine parallel operation without mutual interference.

[0123] 3. Safety optimization: Compared with traditional square wave modulation (which may apply reverse voltage), this application preferably adopts the "pulse on / off" mode, which avoids the component being in a reverse bias state, thereby completely eliminating the risk of bypass diode conduction and heat generation, and realizing non-destructive testing.

[0124] 4. Field test verification: The measured data shows that even under extreme conditions of strong midday sunlight (irradiance of about 800–1100 W / m²), the system described in this embodiment can still acquire SWIR electroluminescence images with excellent signal-to-noise ratio and clarity close to the darkroom standard, which fully verifies the effectiveness and robustness of the method in engineering applications.

[0125] Figure 5 A flowchart illustrating another method for acquiring electroluminescent images provided in this application embodiment is shown below. Figure 5 In some embodiments, the pseudo-random sequence current includes a pseudo-randomly distributed forward current and a zero current, and the infrared image sequence includes multiple infrared image frames acquired within the pseudo-random sequence current period. The method for acquiring electroluminescent images includes the following steps: S501, extract the grayscale value sequence of each pixel in an infrared image frame across multiple consecutive infrared image frames.

[0126] The high-voltage excitation module outputs a series of rapidly switching pseudo-random current pulses to the photovoltaic module, rather than a simple square wave. The pseudo-random modulated signal is a signal form with specific patterns but approximate random characteristics. It is generated by a deterministic algorithm, seemingly disordered yet precisely reproducible, combining the wide-spectrum characteristics of random signals with the controllability of periodic signals. By rapidly switching carrier parameters (such as frequency, phase, or amplitude), its spectrum approximates white noise distribution, effectively improving anti-interference capabilities. Specifically, the pseudo-random sequence current configuration length is As a modulation sequence, the output frequency is 100Hz, each symbol lasts for 10ms, and the symbol value "1" corresponds to a 5A forward current and "0" corresponds to no current. Within a complete sequence period, "1" and "0" each appear about half the time. From a macroscopic perspective, the total input energy is greatly reduced compared to continuous DC. At the same time, the sequence is randomly distributed in a short time window, making its spectral characteristics close to white noise.

[0127] During the detection process, after the pseudo-random sequence modulated coded current excitation signal is sent to the photovoltaic module, the image acquisition module continuously acquires infrared images at a fixed frame rate of 100fps, synchronized with the code elements. The image acquisition module is started via software to acquire and record the timestamp of the start of each frame's exposure. Simultaneously, the system host records the time reference for the start of the power modulation sequence output and assigns a corresponding sequence symbol index to each frame of the image accordingly. This yielded a series of image frames. ,in Indicates the frame number. It represents the pixel coordinates in an image frame and knows the modulation state corresponding to each frame. .

[0128] Due to variations in ambient lighting and camera noise, the electroluminescence signal in a single frame image is extremely weak or even invisible. A cross-correlation method is used to accumulate and extract the signal for each pixel. Take the pixel in a continuous sequence grayscale value sequence in a frame .

[0129] S502 performs cross-correlation calculations on the gray value sequence of each pixel on the infrared image frame and the modulation sequence of the pseudo-random sequence current to obtain an electroluminescent image.

[0130] Among them, the extracted grayscale value sequence With known modulation sequences Perform cross-correlation calculations. Through this calculation, the weak EL signal submerged in strong noise is "fished out" to generate a high signal-to-noise ratio EL result image.

[0131] Specifically, when using code-domain based spread spectrum modulation, a cross-correlation algorithm is preferred, which calculates the intensity sequence of each captured pixel over time against a known excitation sequence. This method, combined with active electrical fault diagnosis and air-ground collaborative deployment, not only significantly improves anti-interference capabilities and enables online detection in strong outdoor light environments, but also provides comprehensive technical support for the safe operation and maintenance of photovoltaic assets. The cross-correlation calculation satisfies the following formula: First, define the modulation reference sequence as follows: Its value is defined according to different excitation methods: Binary on / off modulation: Take ,definition The range of values ​​is .in, Indicates the state of the modulation sequence. Alternating pulse current (+I / O): directly taken (Where -1 represents the physical zero current background state).

[0132] Three-level (+I / 0 / -I) modulation: [Take...] .

[0133] For the short-wave infrared (SWIR) electroluminescence image sequence after inter-frame registration / alignment, by pixel Perform cross-correlation / despreading operations to obtain the pixel output of the electroluminescent image. The specific calculation formula is as follows:

[0134] in, Represents pixels The final correlation value (EL signal strength); Represents pixels The One grayscale value; This indicates the optional removal of trend / background items; Represented as the code sequence energy normalization factor, this value is [value] in the binary / alternating pulse case. In the three-level case, only the energy of non-zero symbols is counted to avoid diluting the signal amplitude by the zero state; This indicates the total number of frames in the image sequence.

[0135] For example, in the case of binary on / off modulation... And remove trend items When the above operation is performed, it simplifies to:

[0136] in, This indicates the on / off state of the k-th modulation (1 for on excitation, 0 for off excitation).

[0137] The above calculations utilize the orthogonality and autocorrelation properties of pseudo-random sequences to efficiently reveal the degree of synchronization between pixel grayscale changes and excitation current changes: For pixels containing EL signals (faulty or normal cells): whenever current is turned on When the pixel brightens, the current is turned off. The pixel dims. At this time... with sequence They show a strong positive correlation, and the cumulative results are as follows: The value will be significantly higher than zero.

[0138] For pixels containing only background noise (such as borders and ground): their grayscale changes are dominated by ambient light and have no logical relationship (irrelevant) with pseudo-random power switching. With sufficiently long sequence accumulation, positive and negative terms cancel each other out, ultimately... Approaching 0.

[0139] For example, in the three-level (+I / 0 / -I) case, the detrending term The preferred configuration is the pixel average of the "0-state frame" (i.e., the image frame acquired at the moment of zero current output). Take the energy term that is counted by non-zero symbols.

[0140] For example, detrending items The value selection method also includes taking the mean of the entire sequence of images, and can be combined with moving average or low-pass filtering algorithms to suppress slow drift or changes in ambient light.

[0141] By performing the above calculations on each pixel of the entire image, the system generates a correlation image, whose grayscale intensity directly reflects the electroluminescence intensity at the corresponding physical location. Finally, the image processing module performs linear stretching and adaptive filtering on the correlation image to remove residual speckle noise, thus obtaining the final high-definition image of electroluminescence defects. In this image, normally luminescent cell areas exhibit a certain brightness, while areas with hidden cracks, broken grids, or black cores, due to weak EL emission, appear as clear dark spots or streaks, thereby achieving precise visualization of defects.

[0142] Considering the physical dynamic characteristics of high-power power supplies (e.g., current settling time of approximately 5ms), to ensure the accuracy of related calculations, this embodiment preferably sets the symbol rate to 5Hz-50Hz (i.e., symbol width of 20ms-200ms). This ensures that the power supply can output a complete flat-top waveform while also being sufficient to resist low-frequency environmental interference such as cloud movement.

[0143] In this embodiment, the advantage of using pseudo-random sequence (spread spectrum) modulation is that: 1. Spectrum Spreading: The sequence covers a wide range of frequency components, making it difficult for fixed-frequency interference (such as power frequency flicker and inverter noise) to have a sustained impact on it. The noise energy is significantly diluted during the despreading process.

[0144] 2. Anti-drift capability: Even under conditions of slow changes in illumination (such as cloud cover of 0.1Hz-1Hz), since such changes are low-frequency signals relative to rapidly switching pseudo-random sequences, cross-correlation operations naturally have high-pass filtering characteristics, which can effectively filter out such drift.

[0145] 3. By accumulating the correlation over a complete sequence period (e.g., 5-10 seconds), the system can extract the EL signal, which was originally submerged in strong sunlight, with a high signal-to-noise ratio. Experiments show that, in conjunction with an InGaAs camera and 1ms hard synchronization technology, this method can stably detect minute microcracks down to the millimeter level inside the component.

[0146] Specifically, the aforementioned cross-correlation (despreading) calculation process can be accelerated on the parallel hardware of the edge computing unit, such as FPGA, DSP, or GPU, to achieve millisecond-level real-time processing of high-resolution SWIR image sequences.

[0147] In summary, when using code-domain-based spread spectrum modulation (SDM) combined with relevant algorithms, the system exhibits stronger anti-interference capabilities (processing gain). Utilizing the autocorrelation (similar to an impulse function) and cross-correlation (close to zero) properties of pseudo-random sequences, the system can not only extract weak signals from strong background light but also effectively suppress co-frequency crosstalk during multi-machine parallel operation by allocating orthogonal code sequences, ensuring the stability and reliability of large-scale cluster inspections.

[0148] For example, by using a correlation accumulation period of approximately 10 seconds, the electroluminescence signal that was originally submerged in noise can be extracted. Experiments show that an industrial camera with a 10-bit ADC, or even an 8-bit camera, can reconstruct electroluminescence images under daytime outdoor conditions. It is worth noting that the code length and rate of the pseudo-random sequence can be adjusted as needed: longer sequences offer higher processing gain and better anti-interference capabilities, but also longer acquisition times; excessively high code rates are limited by the camera's frame rate limit and photon accumulation effects, requiring a balance. Therefore, the modulation scheme can be dynamically adjusted based on the ambient light intensity and the severity of defects to achieve the best imaging results.

[0149] exist Figure 3 The photovoltaic module defect detection method shown also includes an electrical fault detection method, which will be discussed below. Figure 6 The technical solution for the above-mentioned photovoltaic module defect detection method further introduces the content of the electrical fault detection method.

[0150] Figure 6 This is a flowchart illustrating a method for acquiring electroluminescent images according to an embodiment of this application. (See attached diagram.) Figure 6 In some embodiments, the electrical fault detection method includes the following steps: S601, monitor the loop response waveform of the encoded current excitation signal; wherein the loop response waveform is a voltage response waveform and / or a current response waveform.

[0151] In this circuit, the high-voltage excitation module and the photovoltaic module form a loop. After the high-voltage excitation module inputs the coded current excitation signal to the photovoltaic module, the high-voltage excitation module receives the loop signal of the coded current excitation signal, such as current and / or voltage, and obtains the loop response waveform of the loop signal.

[0152] S602, Based on the dynamic impedance spectrum or high-frequency noise characteristics of the circuit response waveform, determine whether the photovoltaic module is abnormal.

[0153] When abnormal and severe fluctuations are detected in the dynamic impedance spectrum at high frequencies, or when the characteristic high-frequency noise (hash noise) in the voltage waveform exceeds a preset threshold, the system determines that there is poor contact in the string circuit (such as loose MC4 connector) or potential DC arcing.

[0154] S603, if the photovoltaic module malfunctions, stop inputting the coded current excitation signal to the photovoltaic module.

[0155] In this embodiment, electrical safety diagnostics are integrated to achieve "accompanying health check." While acquiring images, the system utilizes the high-precision readback capability of the high-voltage power supply to monitor the dynamic impedance and frequency response characteristics of the photovoltaic string circuit in real time. This allows for the simultaneous detection of electrical hazards such as poor contact and DC arcing without adding extra workload, upgrading the simple "defect localization" to "comprehensive safety diagnosis," providing more comprehensive protection for power plant assets.

[0156] Specifically, to enhance the safety and functionality of the testing process, the system integrates active electrical fault diagnosis functionality in its hardware. The high-voltage excitation module incorporates a high-precision voltage / current readback unit (preferably with a sampling rate greater than 1kHz) to simultaneously acquire the voltage and current response data of the circuit while outputting the coded current excitation signal. This response data is used to analyze the dynamic impedance characteristics of the photovoltaic string circuit in real time. When severe impedance fluctuations are detected at high frequencies, or characteristic high-frequency noise (hash noise) appears in the voltage waveform, the system determines that there is poor contact (such as a loose MC4 connector) or a potential DC arcing hazard in the string circuit. At this point, the system immediately controls the high-voltage excitation module to stop outputting, or utilizes the bidirectional characteristics of the power supply to perform a reverse suction operation to instantly extinguish the arc, thereby achieving simultaneous diagnosis of the electrical connection status of the modules during EL imaging.

[0157] Optionally, and further, the system combines synchronously acquired electrical impedance data to perform a multimodal assessment of the defect severity. For example, when the image shows a hidden crack and the impedance data shows an abnormally high loop resistance, the system will automatically increase the risk level of the defect and recommend immediate maintenance.

[0158] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0159] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0160] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0161] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0162] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0163] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the appended claims.

[0164] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A photovoltaic module defect detection system for daytime environments, characterized in that, include: A high-voltage excitation module is used to input a coded current excitation signal based on code domain modulation to at least one photovoltaic module to excite the photovoltaic module to generate a time-varying electroluminescent signal; wherein the coded current excitation signal is an alternating current or a pseudo-random sequence current; An image acquisition module is used to continuously acquire infrared images of the photovoltaic module and obtain an infrared image sequence; An image processing module is used to acquire an electroluminescent image of the photovoltaic module based on the infrared image sequence; A defect detection module is used to obtain defect detection results of the photovoltaic module based on the electroluminescent image.

2. The photovoltaic module defect detection system under daytime conditions according to claim 1, characterized in that, The high-voltage excitation module includes a programmable high-voltage DC power supply and a spread spectrum modulation control unit; The spread spectrum modulation control unit is used to control the programmable high voltage DC power supply so that the programmable high voltage DC power supply outputs the coded current excitation signal.

3. The photovoltaic module defect detection system under daytime conditions according to claim 1, characterized in that, The image acquisition module includes an infrared detector, which is used to acquire infrared images of the photovoltaic module.

4. The photovoltaic module defect detection system according to claim 3, characterized in that, The infrared detector's lens is equipped with a narrowband optical filter or a long-pass optical filter.

5. A photovoltaic module defect inspection system for daytime environments, characterized in that, Includes a photovoltaic module defect detection system under daytime conditions as described in any one of claims 1-4, and a mobile platform; The image acquisition module of the photovoltaic module defect detection system under daytime conditions is mounted on the mobile platform.

6. A method for detecting defects in photovoltaic modules under daytime conditions, characterized in that, The photovoltaic module defect detection system based on any one of claims 1-4 under daytime conditions includes: A coded current excitation signal based on code domain modulation is input to at least one photovoltaic module to excite the photovoltaic module to generate a time-varying electroluminescent signal; wherein the coded current excitation signal is an alternating current or a pseudo-random sequence current; Infrared images of the photovoltaic module are continuously acquired to obtain an infrared image sequence; Based on the infrared image sequence, the electroluminescence image of the photovoltaic module is obtained; Based on the electroluminescent image, the defect detection results of the photovoltaic module are obtained.

7. The method for detecting defects in photovoltaic modules under daytime conditions according to claim 6, characterized in that, The alternating current includes periodically spaced forward current and zero current. The infrared image sequence includes a set of luminescent images and a set of background images. The set of luminescent images includes multiple luminescent image frames acquired during the forward current cycle, and the set of background images includes multiple background image frames acquired during the zero current cycle. Based on the infrared image sequence, the electroluminescence image of the photovoltaic module is obtained, including: The set of luminescent images and the set of background images are respectively subjected to frame-by-frame image registration and averaging to obtain average luminescent images and average background images; The average luminescence image and the average background image are differentially processed to obtain an electroluminescent image.

8. The method for detecting defects in photovoltaic modules under daytime conditions according to claim 6, characterized in that, The pseudo-random sequence current includes pseudo-randomly distributed forward current and zero current. The infrared image sequence includes multiple infrared image frames acquired within the period of the pseudo-random sequence current. Based on the infrared image sequence, the electroluminescence image of the photovoltaic module is obtained, including: Extract the grayscale value sequence of each pixel in the infrared image frame across multiple consecutive infrared image frames; An electroluminescent image is obtained by performing cross-correlation operations on the grayscale value sequence of each pixel on the infrared image frame and the modulation sequence of the pseudo-random sequence current.

9. The method for detecting defects in photovoltaic modules under daytime conditions according to any one of claims 6-8, characterized in that, Based on the electroluminescent image, the defect detection results of the photovoltaic module are obtained, including: The electroluminescent image is input into a pre-constructed defect classification and detection model to obtain the defect detection result output by the defect classification and detection model; wherein, the defect classification and detection model is trained based on a first deep learning model, and the defect detection result includes defect location, defect type, and damage degree; the first deep learning model is a convolutional neural network, a U-Net network, or a generative adversarial network.

10. The method for detecting defects in photovoltaic modules under daytime conditions according to any one of claims 6-8, characterized in that, After acquiring the electroluminescence image of the photovoltaic module, the method further includes: The electroluminescent image is input into a pre-constructed image denoising and enhancement model to obtain an electroluminescent image after denoising processing by the image denoising and enhancement model; wherein, the image denoising and enhancement model is trained based on a second deep learning model; the second deep learning model is a convolutional neural network, U-Net, or generative adversarial network.

11. The method for detecting defects in photovoltaic modules under daytime conditions according to any one of claims 6-8, characterized in that, The method also includes Monitor the loop response waveform of the encoded current excitation signal; wherein the loop response waveform is a voltage response waveform and / or a current response waveform; Based on the dynamic impedance spectrum or high-frequency noise characteristics of the loop response waveform, determine whether the photovoltaic module is abnormal. If the photovoltaic module malfunctions, the input of the coded current excitation signal to the photovoltaic module shall be stopped.

Images