Method and system for inspecting photovoltaic modules using image analysis

Abstract

The invention relates to a method for inspecting a module in a photovoltaic power plant with a plurality of modules, each consisting of a plurality of cells. This method comprises: processing (SIGNcar) an image (IM) of the inspected module to determine a characteristic signature of the inspected module; comparing (COMP) the characteristic signature with reference signatures stored in a database (DB) that associates each of a plurality of reference modules with a reference signature and a serial number; when the characteristic signature coincides with one of the reference signatures, assigning to the inspected module the serial number (IDref) of the reference module whose reference signature matches the specific signature. Figure 1 (for the abstract)

Description

Title of the invention: Method and system for inspecting photovoltaic modules by image analysis technical field

[0001] The field of the invention is that of the inspection of photovoltaic modules by image analysis. Previous technique

[0002] A technique for characterizing a photovoltaic module consists of observing a Light is emitted by the electroluminescence phenomenon from the various cells that make up the module. To achieve this, a current is injected into the module, and a near-infrared image of the module is acquired using a camera. The recovered image shows the quality of the cells and information about any defects within the module.

[0003] Generally, the operator of a photovoltaic solar power plant seeks to source photovoltaic modules from suppliers offering the best value for money. To this end, they conduct quality audits of the suppliers. In particular, they collect electroluminescence images, obtained during the quality control performed by the suppliers on the manufactured modules, for the purpose of evaluating samples.

[0004] Then, when the operator purchases modules, all the images acquired during quality control are transmitted to him. He thus has an electroluminescence image of each of the modules purchased, associated with the serial number or identifier of the module.

[0005] During the construction of the power plant, the modules are deployed without recording the position at which a particular module is installed, i.e., a module corresponding to a given serial number. Therefore, during the operational phase of the power plant, it is not possible to know the serial number of a module deployed at a given position, except by going on-site to read this number on the back of the module, which proves impractical and very time-consuming since a power plant can contain more than 100,000 modules.

[0006] To locate the modules in a power plant, one solution would be to use the electronic and communicating devices attached to the back of the modules, whose main function is to optimize the electrical power produced. This solution would require programming each of these devices manually to store the serial number of the corresponding module. However, on the one hand, these devices are expensive, particularly in terms of labor. And, on the other hand, this manual data collection is far too time-consuming a strategy in a power plant that can have more than 100,000 modules.

[0007] It follows that there is currently no solution to ensure the traceability of a module from its production to its operation and thus to allow, for example, monitoring of its health status and, where appropriate, identification of whether defects affecting it are related to its production, transport, installation, or use (aging). Description of the invention

[0008] One objective of the invention is to improve the analysis and monitoring of the performance of a photovoltaic solar power plant. To this end, the invention proposes a method for inspecting a module of a photovoltaic power plant having a plurality of modules, each consisting of a plurality of cells, comprising: - the processing of an image of the inspected module to determine a characteristic signature of the inspected module; - the comparison of the characteristic signature with reference signatures stored in a database which associates each of a plurality of reference modules with a reference signature and an identifier; - when the characteristic signature coincides with one of the reference signatures, the assignment to the inspected module of the identifier of the reference module whose reference signature coincides with the specific signature.

[0009] Some preferred but not limiting aspects of this process are as follows: - the characteristic signature of the inspected module is a map of the characteristics of the image of the inspected module; - the image processing of the inspected module includes a segmentation in the image of each of the cells of the inspected module and, for each of the segmented cells, a feature map, the characteristic signature of the inspected module consisting of the union of the feature maps of the segmented cells; - the mapping of features of the image of the inspected module is obtained by means of an object detection model which has previously been machine-learned using annotated images of modules; - the mapping of features of the image of the inspected module includes a mapping of defects of the module; - the mapping of image characteristics of the inspected module also includes a mapping of brightness variations in the image of the inspected module; - image processing includes pre-processing comprising the successive operations of isolating the inspected module, resizing the inspected module and transforming the inspected module into a rectangle; - it also includes the issuing of an alert when the characteristic signature does not coincide with any of the reference signatures; - it includes repeating the processing step from an image of the inspected module acquired at a later date and comparing the characteristic signatures of the inspected module determined at each iteration of the processing step; - it includes a preliminary step consisting, for each of the reference modules, of processing an image of the reference module to determine the reference signature of the reference module; - the image of the inspected module is an electroluminescence image; - the image of the inspected module is an image acquired by a drone. Brief description of the drawings

[0010] Other aspects, objects, advantages and features of the invention will become more apparent upon reading the following detailed description of preferred embodiments thereof, given by way of non-limiting example, and made with reference to the accompanying drawings in which:

[0011] - Fig. 1 is a diagram representing different steps implemented in a possible implementation of the process according to the invention;

[0012] - Figure 2 represents an electroluminescence image acquired by a drone flying over a power plant at different zoom levels;

[0013] - Figure 3 represents various defects that may be observed by means of a electroluminescence image of a cell;

[0014] - [Fig. 4] illustrates variations in brightness in an image electroluminescence of a module;

[0015] - [Fig. 5] is a diagram representing a map of imaged features by an image of a module.

[0016] DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS

[0017] The invention relates to a method for inspecting a photovoltaic module in a photovoltaic solar power plant equipped with a plurality of photovoltaic modules, each consisting of a plurality of photovoltaic cells. By way of example, the power plant may comprise on the order of 100,000 to 200,000 modules, and each module may consist of 144 cells in a matrix arrangement of 6*24 cells.

[0018] In what follows, we will take as a privileged example an inspection based on electroluminescence images of modules. However, the invention is not limited to this inspection method and in fact extends to any module imaging method from which it is possible to extract differentiating features of an imaged module and thus determine a characteristic signature of the module. picture.

[0019] The invention can therefore also be implemented by exploiting infrared thermal images, photoluminescence images, synchronous detection thermography images (“Lock-In Thermography” in English), active thermography images, UV fluorescence images or even optical images.

[0020] With reference to [Fig. 1], the inspection process is preceded by a phase T0 of creating a reference database (RD) from electroluminescence images of reference modules previously acquired during an ACQini step. Each of these images is associated with an IDref identifier of the corresponding reference module (for example, a serial number). Creating the reference database (RD) involves processing the electroluminescence images during a SIGNref step to determine a reference signature for each of the reference modules. The RD then associates, for each reference module, the IRref identifier of the reference module with the reference signature of the reference module.

[0021] The electroluminescence images of the reference modules are typically images provided to the power plant operator by a third party. For example, as previously mentioned, when a power plant operator purchases modules from a supplier, all the electroluminescence images produced during the supplier's quality control are transmitted to them. They thus possess an electroluminescence image of each module delivered by the supplier, associated with a unique module identifier. For example, the file name containing the image includes the module's serial number. The ACQini step is therefore implemented by the module supplier.

[0022] Alternatively, the ACQini step can be implemented by the operator, for example in the laboratory after receipt of the modules and before their deployment on site.

[0023] In the context of the invention, the inspected module is a module that has been delivered to the power plant operator. With reference to [Fig. 1], an electroluminescence image of the inspected module is acquired by the operator during an ACQinsp step, for example in the laboratory (particularly when the ACQini step is implemented by a third party) or on-site in a darkened room or outdoors (at night or during the day). By way of example, the image may be an image representing one or more modules deployed on-site, acquired using a tripod mounted on the modules' support structures or using a flying drone. [Fig. 2] illustrates in this regard an electroluminescence image acquired by a drone flying over a power plant at different zoom levels. In the IM image zoomed to the scale of a module, cracks can be observed, probably caused by improper handling of the module during its delivery. Damage or improper handling of the module during transport or on-site installation.

[0024] The module inspection process is implemented during a subsequent phase T1 following the database creation phase T0, after the modules have been delivered to the operator and before or after their deployment on site. This process includes the implementation by a processor of a data processing device of steps for obtaining the electroluminescence image representing the inspected module acquired during the ACQinsp step and for processing this image during a SIGNcar step to determine a characteristic signature of the inspected module.

[0025] The process continues with a COMP step comparing the characteristic signature with the reference signatures stored in the BdD database which associates each of the reference modules with its reference signature and its identifier.

[0026] When the COMP comparison step concludes that the characteristic signature coincides with one of the reference signatures (i.e. these signatures have a coincidence rate greater than a threshold, for example 80%), the method includes assigning to the inspected module the identifier of the reference module whose reference signature coincides with the specific signature.

[0027] Traceability of the inspected module is thus ensured, making it possible, for example, during the analysis and monitoring of the power plant's performance, to trace back to the manufacturing batch of a module that is the cause of underperformance in the power plant. Or, as described later, to monitor the health status of a module.

[0028] The process can also be continued by using the identifier thus assigned to the inspected module to identify the inspected module on a map of the modules deployed in the power plant. By repeating the process described above for each of the modules deployed in the power plant, it then becomes possible to locate any given module using its identifier. In particular, drone inspection of the modules makes it possible to obtain such a map of the modules' positions with their identifiers.

[0029] When the COMP comparison step concludes that the characteristic signature of the inspected module does not coincide with any of the reference signatures, the process includes issuing an alert indicating a risk of fraud. This alert can be issued, in particular, during a quality control check of the modules implemented by the operator after receipt of the modules. The operator can thus ensure that all the inspected modules correspond to the expected modules. Otherwise, the operator may be faced with fraud (for example, the manufacturer provided an image of a module without defects while the inspected module actually has one or more defects) or not (the manufacturer made a mistake by not transmitting the image of the correct module with the correct serial number). To resolve the doubt following Upon issuance of the alert, an operator can travel to the site of the power plant to record the identifier of the inspected module whose characteristic signature does not coincide with any of the reference signatures and compare the image of the inspected module with the image of the reference module in question (retrieved from the reference database with the identifier of the module recorded).

[0030] The method according to the invention may also include a new inspection of the module during a T2 phase subsequent to the T1 phase, for example, one or two years later. This new inspection includes obtaining an image of the inspected module acquired during an ACQinsp* step at the later date, SIGNcar* processing of this new image to determine a new characteristic signature of the inspected module, and then COMP* comparing the characteristic signatures determined in the T1 and T2 phases of the inspected module at each iteration of the processing step. If the characteristic signatures determined in the T1 and T2 phases do not coincide, the method may include issuing an alert indicating possible deterioration of the module. It should be noted that assigning the module identifier to the T1 phase makes such monitoring of the module's health status over time possible.This T2 phase can also be repeated over time, for example every year.

[0031] A possible embodiment of determining the reference signatures of the reference modules and the characteristic signatures of the inspected modules is detailed below. According to this embodiment, a module signature is a feature map of the module image, specific to the module.

[0032] This feature mapping may include (where appropriate) a module defect mapping, these defects being apparent in the module image. As illustrated in [Fig. 3], these defects are, for example, electroluminescence-revealed defects such as cracks or microcracks or cross-shaped defects C (on the left in [Fig. 3]), black spots T (in the center of [Fig. 3]), ring-type defects R (on the left in [Fig. 3]).

[0033] Feature mapping may also include mapping of brightness variations in the image of the inspected module. As shown in [Fig. 4], this mapping of brightness variations may indicate high-brightness cells (CH), low-brightness cells (CL), or include black border regions (BN) at the periphery of cells.

[0034] In an embodiment exploiting optical images of the modules, the modules may bear a unique mark, deliberately inscribed on the module by the manufacturer (such as a QR code engraved on the glass edge of a module) and visible during the acquisition of the module image. In this case, the signature of a module corresponds to a map of image patterns that correspond to this unique mark.

[0035] To optimize image processing performance for determining module signatures, these images are segmented into cells, and feature mapping is performed cell by cell. Preferably, image processing of a module (reference module or inspected module) involves segmenting the image of each of the module's cells and, for each segmented cell, creating a feature map as described above. The characteristic signature of the inspected module or the reference signature of a reference module then consists of the union of the feature maps of the segmented cells.

[0036] To perform this segmentation, the image of a module (reference or inspected) can be subjected to preprocessing comprising the successive operations of isolating the inspected module, resizing the image of the inspected module, and transforming this image of the inspected module into a rectangle. For example, this preprocessing includes extracting the parts of the image that relate to one or more modules. Each entire module in this extraction is then isolated to provide an image of a single module. The non-entire modules in this extraction are removed. The image of a module, distorted by the capture into a parallelogram, is then resized and transformed into a rectangle of predetermined size. The image is then segmented into cells, for example, according to an automatic segmentation method that considers an expected size for the cells or by identifying the cell contours in the image.

[0037] In one possible embodiment, the mapping of imaged features is obtained using an object detection model (or a semantic segmentation model) that has previously undergone machine learning with annotated images of modules and / or cells. The object detection model is, for example, the Yolov8 model trained on thousands of electroluminescence images to detect defects on each cell.

[0038] The resulting map can thus consist of a set of rectangles encompassing each of the features identified in the image. Each rectangle records the position and area of ​​a feature while being associated with a feature label (or category). Figure 5 illustrates in this regard an example of a map of imaged features displayed on a 6 x 24 cell matrix. In this map, the label associated with a feature category corresponds to the background pattern of the corresponding encompassing rectangle, which in this example allows the identification of a defect type (C, T, R) or a brightness type (BN, CL, CH). This map can be saved in a file that constitutes a unique fingerprint of the analyzed module. This map can also be superimposed on the module image (or, where applicable, the pre-processed image).

[0039] A possible realization of the comparison of signatures (characteristic signature of an inspected module compared to the reference signatures stored by the database or comparison of characteristic signatures of the same module determined at different times in time) consists of intersection over union measurement ("Intersection over Union" or loU in English) which describes the level of overlap (or rate of cover) between the maps, typically between the different encompassing rectangles mentioned above.

[0040] This measure may, in particular, consist of determining whether a given percentage X of features extracted from a first image are found in the features extracted from a second image. Here, the first image is, for example, an image of a reference module, and the aim is to verify whether the features extracted from an image of an inspected module conform to more than X%, for example, more than 80%, the features extracted from the image of the reference module. If so, the signatures associated with the first and second images coincide.

[0041] This measurement may include calculating the recovery rate of each feature of the first image, and if the average of all the recovery rates is greater than X%, then the first and second images are images of the same module. Alternatively, each of the recovery rates must be greater than X% to conclude that the first and second images are images of the same module. In one possible embodiment, recovery rates may be calculated for a given type of feature considered as reference features, for example, defects that can only appear during the manufacturing phase.

[0042] Moreover, in the case where the number of features differs between the first image (for example that of a reference module) and the second image (that of an inspected module), for example features present in the first image are not identified in the second image, it is still possible to consider that the two images coincide and represent the same module if more than Z% of features extracted from the first image are present in the set of features extracted from the second image.

[0043] Alternatively, this measure may consist of determining whether a given percentage Y of features extracted from a first image are not found in the features extracted from a second image. Here, the two images may be those of the same inspected module, identified as such by the identifier assigned to it according to the invention. The first image may be that of the module inspected at a later date than the earlier date of acquisition of the second image. In this case, the aim is to verify whether the features extracted from an image of the module inspected at the later date differ by more than Y%. For example, more than 20% of the features extracted from the image of the module inspected at the earlier date. If so, the inspected module can be identified as having more defects at the later date than at the earlier date. This degradation of the inspected module is further quantified by the result of the loU measurement.

[0044] Furthermore, in order to improve identification, it is possible to refine this quantification by type of characteristic, for example by type of defect, by including, for instance, defects that can only appear during the manufacture of the module (e.g., ring-type defects) or during its transport or installation (e.g., crack-type defects of a certain shape), or defects likely to appear during the operation of the power plant (e.g., defects caused by wind or a heavy load such as snow). Alternatively, it is possible to exclude certain characteristics (e.g., manufacturing defects) and retain only certain others (e.g., defects likely to appear during the operation of the power plant).

[0045] The invention is not limited to the method described above and extends to an inspection system for a module of a photovoltaic power plant, comprising a processor configured to implement the method described above, as well as to a computer program product comprising instructions which, when executed by a computer, cause the computer to implement the method described above.

Claims

Demands

1. A method for inspecting a module of a photovoltaic power plant having a plurality of modules, each consisting of a plurality of cells, comprising: - processing (SIGNcar) an image (IM) of the inspected module to determine a characteristic signature of the inspected module; - comparing (COMP) the characteristic signature with reference signatures stored in a database (DB) which associates with each of a plurality of reference modules a reference signature and an identifier; - when the characteristic signature coincides with one of the reference signatures, assigning to the inspected module the identifier (IDref) of the reference module whose reference signature coincides with the specific signature.

2. A method according to claim 1, wherein the characteristic signature of the inspected module is a mapping of features revealed by the image of the inspected module.

3. A method according to claim 2, wherein the processing (SIGNcar) of the image of the inspected module comprises a segmentation in the image of each of the cells of the inspected module and, for each of the segmented cells, a feature map, the characteristic signature of the inspected module consisting of the union of the feature maps of the segmented cells.

4. A method according to any one of claims 2 and 3, wherein the mapping of features revealed by the image of the inspected module is obtained by means of an object detection model which has previously been machine-learned using annotated images of modules.

5. A method according to any one of claims 2 to 4, wherein the mapping of features revealed by the image of the inspected module includes a mapping of defects of the module.

6. A method according to claim 5, wherein the mapping of features revealed by the image of the inspected module further comprises a mapping of brightness variations in the module image inspected.

7. A method according to any one of claims 1 to 6, wherein the image processing includes a pre-processing comprising the successive operations of isolating the inspected module, resizing the inspected module and transforming the inspected module into a rectangle.

8. A method according to any one of claims 1 to 7, further comprising issuing an alert when the characteristic signature does not coincide with any of the reference signatures.

9. A method according to any one of claims 1 to 8, comprising repeating the processing step (SIGNcar*) from an image of the inspected module acquired at a later date and comparing (COMP*) the characteristic signatures of the inspected module determined at each of the iterations of the processing step.

10. A method according to any one of claims 1 to 9, comprising a preliminary step consisting, for each of the reference modules, of processing an image of the reference module to determine the reference signature of the reference module.

11. A method according to any one of claims 1 to 10, wherein the image of the inspected module is an electroluminescence image.

12. A method according to any one of claims 1 to 11, wherein the image of the inspected module is an image acquired by a drone.

13. Inspection system for a module of a photovoltaic power plant, comprising a processor configured to implement the method according to any one of claims 1 to 12.

14. Product computer program comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 12.

Images