An automated diagnostic system for real-time high-resolution defect detection in a solar park and method thereof

Abstract

The invention relates to a novel automated system 10 for high-resolution real-time defect identification in a solar park 12 and a method 300 thereof. A smart vision device 16 includes a receiver module 16.2 which receives power from a solar PV string 20. An Interrogation module 16.3 compares the received power with the power of two neighbouring strings 20. A deployment module 16.4, when received power is lower than the power of two neighbouring strings 20, transmits a deployment signal to trigger the device 16. The device 16 moves to a position of the identified defective string 20. Multiple imaging devices scans imagery of the identified string 20. An image processing module based on machine learning model receives the imagery, analyse and identifies defects associated with the imagery. An alert is transmitted to a maintenance team through an alert module 16.6 to alert the team of the defects.

Description

[0001] AN AUTOMATED DIAGNOSTIC SYSTEM FOR REAL-TIME HIGH- RESOLUTION DEFECT DETECTION IN A SOLAR PARK AND METHOD THEREOF FIELD OF INVENTION The present invention is in the field of solar parks. In particular, the invention relates to an automated diagnostic system for real-time high- resolution defect detection in a solar park and a method thereof. BACKGROUND OF INVENTION There is a need for an appropriate automated real-time monitoring and maintenance of solar parks. This monitoring of solar parks will ensure that there is better detection of any defective photovoltaic modules of solar parks to ensure that such detected defect is addressed effectively and efficiently. Currently, there is a method of analyzing an entire solar park in order to attempt to detect any possible defects of the photovoltaic modules. However, this method is expensive and time-consuming. Often, other methods use multiple sensors which can be placed at various locations within the solar park. The disadvantages with these methods are cost associated with the sensors and the time associated with the frequent maintenance thereof. It is possible to use drones for infra-red (IR) imaging, however, the drones are not sufficiently stable thereby resulting in uncertainty of results that emanate from such methods. Other companies only look for soil-related issues only, leaving out other defects as well as fault- related problems and safety concerns. Further, relying on manual inspection techniques like Electroluminescence (EL) and Infrared (IR) imaging does not provide holistic detection of the defects. For example, US patent no. 10,998,853B2 describes an internet of things enabled solar photovoltaic health IBM monitoring and advising related thereto. The invention utilizes a computer-implemented method to obtain current-voltage samples from specific modules through IoT-triggered switch circuitry between an inverter connected to the module and a current-voltage tracer, detecting anomalies using machine learning, and performing automated root cause analysis by converting data into sequences and using a sequence classifier to identify known anomaly classes. Remedial actions are suggested based on the identified class. However, there is no mention of use on a PV plant or array, especially given how infeasible it would be to connect inverters and tracers for each solar module within a solar park. Another current solution is described by the Korean patent no. 102228512B1 and relates to photovoltaic power generation apparatus with remote control function and function to monitor the failure of solar power plant and fire at connection board using machine learning. The invention relates to a comprehensive photovoltaic power generation setup featuring solar modules, an inverter, a connection box, and monitoring units. The setup is analysed weather data collection, historical data analysis, and machine learning. The monitoring devices track and diagnose faults regression analysis, while also detecting potential fires in the junction box through temperature monitoring. The apparatus employs predicted power rangers based on statistical analysis, flagging failures when actual generated power deviates from the rangers. Failures are counted, and specific fault conditions triggers actions like storing current data. However, this system has a main focus on fires caused in junction boxes and is designed for a specific setup that requires inverters and monitoring units for each module, which would be difficult to implement in the large scale of solar parks. Similarly Korean patent no. 102084784B1 outlines a method for managing the operation of photovoltaic power generation based on machine learning. The invention comprises of solar cell modules, a connection board, and a management server. Real-time data regarding sunlight, current, and the humidity is collected from the solar cell system. Using machine learning, changes in these parameters are predicted by comparing them with pre- analysed reference patterns, accounting for installation location and environmental factors. Predicted trends are communicated to the management server, which detects abnormal predictions and issues control signals to the connection board for necessary actions. A correlation between sunlight amount and current amount is assessed against a present model. Thermal imaging monitors temperature, notifying the server of potential increases and triggering fire prevention measures when needed. Reference patterns for various parameters are generated by the server and shared with the connection board. However, this system also focuses more on fire prevention and is also designed and tested on a single module, rather than a large array like those found in solar parks. Another relevant invention is the Australian patent no.2021102330A4, which presents a solar park monitoring and fault detection system using IoT and machine learning. The system includes the following components: - panel monitoring devices (PMDs) positioned at each solar panel, connected via Wi-Fi Mesh. These PMDs measure electrical parameters and panel temperature. - central monitoring station (CMS) which receives data from PMDs and is equipped with Machine Learning Algorithms (MLAs) for predicting potential faults. However, this invention fails to disclose its complications when being used in large-scale solar parks. It is clear that there is a need for a comprehensive solution which can detect anomalies and classify the defected anomalies effectively and efficiently using both power monitoring mechanisms and mutiple imaging techniques, thereby making solar parks maintenance more accurate and responsive. Thus, a more efficient system involving an automated diagnostic system for real-time high-resolution defect detection with multiple imaging of PV strings is needed to address these challenges. It is accordingly an object of the present invention to provide for a novel automated system of detecting defects in solar parks that at least ameliorate some of the abovementioned drawbacks.

[0002] SUMMARY OF INVENTION According to a first aspect of the invention, there is provided an automated diagnostic system for real-time high-resolution defect detection in a solar park, made of a plurality of solar photovoltaic (PV) strings which are electrically connected to each other, the system includes smart vision device that comprises: a) a receiver module being operable to receive power from the PV strings; b) an interrogation module being operable to compare the received power with a power of at least two neighbouring PV strings; c) a deployment module, in response to the received power being lower than the power of at least two neighbouring PV strings, being operable to transmit a deployment signal to the receiver module thereby triggering the smart vision device to move from an initial position to a position associated with the identified defective solar PV string, the smart vision device further includes multiple imaging devices being operable to scan imagery of the identified defective solar PV string; d) an image processing module equipped with trained machine- learning model, being operable to receive the scanned imagery of solar PV string, analyse the scanned imagery and identify at least one solar panel defect associated with the scanned imagery of the PV string; and e) an alert module, in response to the identified solar panel defect, being operable to transmit an alert message thereby alerting a maintenance team of the identified defective solar PV panel. The identified solar panel defects may be in a form of broken frame, cracked glass, antireflective coating fault, back sheet cracks, finger discoloration, back sheet discoloration, back sheet chalking, module warpage, encapsulant delamination, encapsulant discoloration, shading, soiling, bird dropping, connector failure, junction box failure, ribbon warpage, ribbon chipping, finger corrosion, finger breakage, potential induced degradation (PID), snail trails, cell cracks, ribbon corrosion or discoloration. The smart vision device may be a rover, drone, or unmanned aerial vehicle (UAV). Preferably, the smart vision device may be a rover. The image processing module is equipped with a trained machine- learning mode based on a deep neural network, trained on images of possible fault occurrences, their signature, and behavior. The deep neural network may preferably be a retrieval augmented generation large language model. The solar panel defect may include a fault status. In another embodiment, the solar panel defect may include a safety status. The alert message may be transmitted to an alert device (for example, a mobile device) associated with the maintenance team. The power of the relevant solar PV string may be compared with the power of at least two neighbouring solar PV strings, continuously. Preferably, the power from the relevant solar PV strings may be compared with the power of at least two neighbouring PV strings, every hour. The receiver module, interrogation module, deployment module, and alert module may be part of central processing unit (CPU) located within the smart vision device. The image processing module may be part of a graphic processing unit located within the smart vision device. The central processing unit may further include a geo-location extraction module being operable to extract the location of the relevant solar PV string so as to identify the location of the defective solar PV string, in use. The smart vision device may use a combination of visible image analysis, infra-red image analysis UV – Florescence (UVF), dark lock-in thermography (DLIT) , and Electroluminescence (EL) to scan the imagery of the defective solar PV string. The image processing module trained with machine-learning model may include a processor and a memory coupled to the processor, to memory containing instructions, which are arranged to cause the processor to analyse patterns in the scanned imagery and identify at least one solar park defect associated with the scanned imagery. According to a second aspect of the invention, there is provided a high- resolution defects identification method, the method comprising: a) receiving power from solar PV string and location thereof; b) comparing the received power with power of at least two neighbouring solar PV strings; c) transmitting, in response to the received power being lower than the power of at least two neighbouring solar PV string, a deployment signal to a smart vision device thereby triggering the smart vision device to move from an initial position to a position associated with the identified defective solar PV string; d) scanning imagery of the identified solar PV string; e) receiving the scanned imagery of the defective solar PV string; f) analysing the scanned imagery, and identifying solar panel defects associated with the scanned imagery; and g) transmitting, in response to the identified solar panel defect, an alert message thereby alerting a maintenance team of the identified defective solar PV string. The method may includes, comparing, the power of the relevant solar PV string with the power of the two neighbouring solar PV string, continuously. In a preferred embodiment, the method may include comparing the power of the relevant solar PV string with the power of the two neighbouring solar PV string, every hour. BRIEF DESCRIPTION OF THE DRAWINGS The objects and features of the present invention will become fully apparent from following the description taken in conjunction with the accompanying drawings. Undertaking that these drawings depict only typical embodiments of the invention and are therefore, not to be considered limiting its scope, the invention will be described and explained with additional specific and detail through the use of the accompanying drawings in which: In the drawings: Figure 1 - shows a defect detection system according to one of the embodiments of the present invention; Figure 2 - shows a high-level block diagram showing a smart vision device of the invention of Figure 1 according to one of the embodiments of the present invention; and Figure 3 - shows a high-level defect detection method according to one of the embodiments of the present invention.

[0003] DETAILED DESCRIPTION OF AN EXAMPLE EMBODIMENT While various inventive aspects, concepts, and features of the invention may be described and illustrated herein as embodied in combination with the exemplary embodiments, these various aspects, concepts, and features may be used in many alternative embodiments, either individually or in various combinations and sub-combinations thereof. Unless expressly excluded herein all such combinations and sub-combinations are intended to be within the scope of the present invention. Still further, while various alternative embodiments as to the various aspects, concepts, and features of the invention - such alternative structures, configurations, methods, devices and components, alternatives as to form, fit and function, and so on - may be described herein, such descriptions are not intended to be a complete or exhaustive list of available alternative embodiments, whether presently known or later developed. Those skilled in the art may readily adopt one or more of the inventive aspects, concepts, or features into additional embodiments and uses within the scope of the present invention even if such embodiments are not expressly disclosed herein. Still further, exemplary or representative values and ranges may be included to assist in understanding the present disclosure; however, such values and ranges are not to be construed in a limiting sense and are intended to be critical values or ranges only if so expressly, stated. Moreover, while various aspects, features, and concepts may be expressly identified herein as being inventive or forming part of an invention, such identification is not intended to be exclusive but rather there may be inventive aspects, concepts, and features that are fully described herein without being expressly identified as such or as part of a specific invention. Embodiments will now be described in detail with reference to the accompanying drawings. To avoid unnecessarily obscuring the present disclosure, well-known features may not be described or substantially the same elements may not be redundantly described, for example. This is for ease of understanding. The following description is provided to enable those skilled in the art to fully understand the present disclosure and is in no way intended to limit the scope of the present disclosure as set forth. The embodiments described herein provide a novel automated diagnostic system for real-time high-resolution defect detection in a solar park. According to an embodiment of the present invention, with reference to the drawings and in particular to Figure 1, in which like features are indicated by like numerals, a novel automated system for real-time high-resolution defect detection in a solar park is generally indicated by reference number 10. A solar park 12 is made of a plurality of solar photovoltaic (PV) strings 20, which are electronically connectable to each other. It will be appreciated to those skiled in the art that the solar PV strings 20 can be electrically connectable to each other via any suitable mechanism. The system 12 includes a power monitoring device 14 configured within the solar PV string 20, to monitor power of each of the solar PV strings 20. The system 10 includes a smart vision device 16 which is deployed, if there is a change in power of the solar PV string in comparison with the power of at least two neighbouring solar PV string 20 so as to detect any anomalies on the solar park 14, in particular, on a particular solar PV string 20. In use, the information about the detected anomalies is transmitted to a server 22 associated with a maintenance team so as to enable the maintenance team to react accordingly. The power monitoring device 14, the smart vision device 16 and the server 22 are in communication with one another, either directly or indirectly via a communication arrangement 18. Referring also to Figure 2 which shows a high-level block diagram depicting modules that form part of the smart vision device 16 . The smart vision device 16 includes a server, which, in turn, includes a central processing unit 16.1 defining plurality of conceptual modules 16.2, 16.3, 16.4, 16.5, and 16.6 which correspond to functional tasks performed by the central processing unit (CPU) 16.1. The CPU 16.1 includes a receiver module 16.2, an interrogation module 16.3, a deployment module 16.4, an analysis module 16.5 and an alert module 16.6. Image processing module is part of the Graphic processing unit. Referring also to Figure 3 which illustrates a method 300 of detecting a defect on a solar park 12. In particular, it is a method 300 of detecting a defective solar PV string 20. The power monitoring device 14 continously monitors power of solar PV string 20 to which it is connected. The receiver module 16.2 receives, at block 302, power from the solar PV string 20. The interrogation module 16.3 compares, at block 304, the received power with a power of at least two solar PV strings 20. The interrogation module 16.3 compares the power of the relevant solar PV string 20 with the power of two neighbouring solar PV strings continuously. In a preferred embodiment, the received power and the power of the two neighbouring solar PV strings 20 are compared hourly. If the received power is lower than the power of two neighbouring PV strings 20, the deployment module 16.4 transmits, at block 306, a deployment signal to the smart vision device 16. In simple terms, the deployment signal is a trigger signal transmitted to the smart vision device 16 to trigger its movement. Upon the receipt of the deployment signal, the smart vision device 16 moves from its initial position to a position associated with the location of the defective solar PV string 20. The smart vision device 16 can be in a form of a drone. Alternatively, the smart vision device 16 can be a rover where the deployment signal is used to trigger the rover. It will be appreciated by those skilled in the art that determination of whether or not the received power is lower than the power of at least two neighbouring PV strings will be dependent on a predefined critical power threshold. The deployment signal includes an solar PV string identifier of the defective solar PV 20 and associated location. The smart vision device receiver module, receives the deployment signal and extracts (from the deployment signal) geo-location of the solar PV string. The central processing unit can further include a geo-location extraction module that extract the location of the solar PV string 20 associated with the received solar PV string identifier, so as to identify the location of the defective solar PV string 20. In a preferred embodiment, the smart vision device 16 uses a geo-tagging technique in order to identify the location of the defective solar PV string 20. The smart vision device 16 then moves, at block 308, from an initial position to a position associated with the relevant solar PV string 20. In other words, the deployment signal is a form of a trigger signal which is transmitted to the smart vision device 16 to trigger its movement. Upon receipt of the deployment signal, the smart vision device 16 moves from its initial position (which could be a stationary position) to the position of the defective solar PV string 20. The smart vision device 16 includes a multiple imaging device that scans, at block 310, images of the defective solar PV string 20. In a preferred embodiment, the smart vision device 16 can use any one of (or a combination of any of) the following imaging techniques, visible imaging, IR imaging, UV – Florescence (UVF) imaging, Dark Lock-In Thermography (DLIT) and Electroluminescence (EL), so as to clearly scan any anomalies depicted on the solar PV string 20. In an example embodiment, the smart vision device 16 can scan front and rear end of the solar PV string 20 using the abovementioned techniques An image processing module based on a trained machine-learning model receives the scanned imagery of the defective solar PV string 20 and the analysis module 16.5 analyse the scanned imagery of the solar PV tring 20. The analysis module 16.5 could form part of the image processing module. Therefore, the image processing module analyses the scanned imagery and identify a solar park defect associated with the scanned imagery. In particular, the image processing module based on a trained machine- learning model, includes a processor and a memory coupled to the processor, to memory containing instructions, which are arranged to cause the processor to analyse patterns in the scanned imagery and identify solar panel defects associated with the scanned imagery of the defective solar PV string 20. According to one example embodiment, the machine learning model is based on deep neural network, trained on images of possible fault occurrences, their signature, and behavior. According to another example embodiment the deep neural network is retrieval augmented generation large language model. The combination of multiple imaging system and image processing module based on a trained machine-learning model, retrieval augmented generation (RAG) large language model (LLM) will be used for solar PV string defect exploration and classification in the solar park. The image processing module can be pre-trained such that it eventually has object detection and segmentation capabilities for defect classification of solar PV string specific defects. The image processing module may be part of a graphic processing unit. Deep neural network that is used to train the image processing module is typically based on deep learning techniques, such as Retrieval Augmented Generation, RAG; Retrieval enhancement generation: is a method combining search and generation techniques aimed at improving the performance of search and generation tasks in Natural Language Processing (NLP), where two key components of search and generation cooperate to more intelligently generate text or answer a user's query. By combining the advantages of retrieval and generation, the RAG enables the system to be more comprehensive and accurate when answering user queries, and can adapt to different contexts and tasks. The technique is widely applied to search engines, dialogue systems, and other natural language processing tasks. For example, the defect classification can include the steps of sorting the IR, UVF, EL, and DLIT images. Initially, visible imagery is analysed, followed by IR imagery, UVF imagery, EL imagery and finally analysing the DLIT images. Therefore, the solar panel defect can include a fault status of the solar PV string 20. In another embodiment, the solar panel defect can include the safety status of the solar PV string 20. In other embodiments, the solar park defect can include (but not limited to) any one of the following: broken frame, cracked glass, antireflective coating fault, back sheet cracks, finger discoloration, back sheet discoloration, back sheet chalking, module warpage, encapsulant delamination, encapsulant discoloration, shading, soiling, bird dropping, connector failure, junction box failure, ribbon warpage, ribbon chipping, finger corrosion, finger breakage, potential induced degradation (PID), snail trails, cell cracks, ribbon corrosion / discoloration etc. The comparison of defects identified by individual imaging systems and defects identified through multiple imaging devices of present invention is summarized in Table 1 below:

[0004] Defects IR* UVF* EL* DLIT* Present Invention** Broken / Detached frame X X X X✓Broken / Cracked Glass X X X X✓Antireflective ReflectiveX X X X✓Coating fault Back sheet cracks X X X X✓Finger discoloration X X X X✓Back sheet discoloration X✓X X✓Back sheet chalking X✓X X✓Module Warpage✓X X X✓Encapsulant delamination✓X X X✓Encapsulant discoloration✓X X X✓Shading / Soiling / Bird dropping✓X X X✓Connector failure✓X X X✓Junction box failure✓X X✓ ✓Ribbon warpage X X✓ ✓ ✓Ribbon chipping X X✓ ✓ ✓Finger corrosion X X✓ ✓ ✓Finger breakage X X✓ ✓ ✓Potential Induced✓X✓ ✓ ✓Degradation (PID)Snail Trails✓ ✓ ✓ ✓Cell cracks✓ ✓ ✓ ✓ ✓Ribbon ✓ ✓ ✓ ✓ ✓ corrosion / discoloration *Detectable with individual imaging systems. ** Detectable using a smart vision device of the present invention. Table 1 From the above data, it is clear that smart vision device 16 of the present invention, which use a combination of multiple imaging and image analysis module provides high-resolution defect identification and classification in comparison to individual image analysis. An alert module 16.6, transmits, at block 312, an alert message subsequent to a solar PV string defect being identified. The alert message alerts a maintenance team of the identified solar park defect on the defective solar PV string 20. The alert message can be transmitted to the server 22 which can form part of a monitoring device. The monitoring device associated with a maintenance team, whereby, in use, once the maintenance team is alerted of the solar park defect, the maintenance team is able to take appropriate remedial action.

Claims

CLAIMS 1. An automated system for defect detection in a solar photovoltaic (PV) installation comprising a plurality of PV strings which are electronically connectable to each other, the system comprising: a monitoring unit configured to obtain operational data indicative of the performance of each PV string; a comparison unit configured to compare the operational data of each PV string with that of at least one other neighbouring PV string to identify a PV string exhibiting anomalous performance; a mobile inspection device configured to be selectively deployed to the location of an identified anomalous PV string in response to the comparison data received from comparison unit, wherein the mobile inspection device comprising at least one imaging module configured to acquire image data of the identified PV string using at least one imaging modality; and a processing module configured to analyse the acquired image data using a machine learning model trained on PV string image data to detect defects.

2. The system of claim 1, wherein the mobile inspection device comprises the processing module and further comprises a communication module configured to transmit defect information of the anomalous PV string to a maintenance system or operator.

3. The system of claim 1, wherein the at least one imaging modality is selected from a group comprising of: visible, ultraviolet, electroluminescence, and thermographic imaging.

4. The system of claim 3, wherein thermographic imaging is Dark Lock-In Thermography or infrared Thermography.

5. The system of claim 1, wherein the mobile inspection device is selected from a drone, rover, or unmanned aerial vehicle.

6. The system of claim 1, wherein the machine learning model is based on a deep neural network trained on images selected from a group comprisingPV string defects, PV fault occurrences, PV signatures, and PV behaviour.

7. The system of claim 1, wherein the at least one imaging module is configured to acquire image data in at least two different imaging modalities, and the processing module is configured to combine or fuse the image data from the different modalities to improve defect detection and classification.

8. A method for detecting defects in a solar PV installation comprising a plurality of PV strings which are electronically connectable to each other, the method comprising: obtaining operational data for each PV string; comparing the operational data of each PV string with that of at least one other neighbouring PV string to identify a PV string exhibiting anomalous performance; deploying a mobile inspection device to the location of the identified anomalous PV string; acquiring, via the mobile inspection device, image data of the identified anomalous PV string using at least one imaging modality; and analysing the acquired image data using a machine learning model trained on PV string image data to detect defects.

9. The method of claim 8, wherein the at least one imaging modality is selected from a group comprising of: visible, ultraviolet, electroluminescence, and thermographic imaging.

10. The method of claim 9, wherein thermographic imaging is Dark Lock-In thermography or Infrared Thermography.

11. The method of claim 8, wherein the mobile inspection device is a drone, rover, or unmanned aerial vehicle.

12. The method of claim 8, wherein the machine learning model is based on a deep neural network trained on images selected from a group comprising PV string defects, PV fault occurrences, PV signatures, and PV behaviour.

13. The method of claim 8, wherein image data is acquired in at least two different imaging modalities, and the analysis comprises combining or fusing the image data from the different modalities.

Images