An autonomous anomaly detection system, offshore arrangement and associated methods

Abstract

Method (300), offshore arrangement (124) and autonomous anomaly detection system (202) suitable for offshore wind turbines (112), comprising unmanned vehicles, in particular an aerial vehicle (108) and an underwater vehicle (110), the system capable of comparing data derived from captured images and / or measured distances from the unmanned vehicles, with reference data under reference conditions, to detect discrepancies, anomalies or fault conditions in assembly or pre-assembly phases, in particular through a tolerance comparison, which allows for early detection of potential issues, such as misalignments and structural defects, wherein the system may further comprise machine-readable scanning capabilities to detect machine-readable identifier (126) arranged on components of the wind turbine, to facilitate tracking, certificate emission, and may facilitate 3D generation and / or pinpoint location of defects.

Description

[0001] Description

[0002] AN AUTONOMOUS ANOMALY DETECTION SYSTEM, OFFSHORE ARRANGEMENT AND ASSOCIATED METHODS

[0003] FIELD OF THE INVENTION

[0004] The invention relates to an autonomous anomaly detection system for wind turbines , in particular for of fshore wind turbines , that tracks anomalies or fault conditions from an assembly stage . The invention further relates to an of fshore arrangement comprising said system, and associated methods , which employ unmanned vehicles to ensure the structural integrity of wind turbine components throughout various phases of their li fecycle including prior and during assembly .

[0005] BACKGROUND

[0006] The escalating dimensions and consequent weight of wind turbines , especially of fshore wind installations , and particularly with the advent of substantial hydrogen generation platforms , underscore the criticality of exacting assembly tolerances in the wind energy sector . As turbines grow in si ze , the demand for maintaining stringent tolerance thresholds intensi fies , with the slightest deviations potentially leading to signi ficant structural inef ficiencies or failures . The requirement for precision is not merely incremental ; it is , in fact , becoming increasingly rigorous and demanding, as the industry moves toward larger and more complex installations .

[0007] Current methodologies in the wind turbine industry, both within the of fshore and onshore sector, are predominantly focused on visual inspections conducted by drones , which are typically limited to speci fic parts such as blades . While such inspections are valuable for maintenance and post-assembly evaluations , they fall short of detecting the precise tolerances and precise discrepancies which may be crucial for ensuring the overall structural integrity during the critical phases of transportation and assembly .

[0008] The traditional use of drones of fers a fragmented view, concentrating on visible , above-water components and often neglecting the integral subsea structures , which may include the monopile , transition piece , and caissons . Furthermore , this approach based on pictures and surface defects with a drone , may overlook critical misalignments or tolerance breaches that are vital to the wind turbine ' s overall stability and operational ef ficacy . As the industry progresses , the need for a more nuanced and comprehensive inspection capability increases , one that transcends the limitations of current drone technology which is geared more toward capturing general images rather than performing precise tolerance checks of the whole assembly or at least of a plurality of components being mounted adj acently to one another, particularly underneath sea level and above .

[0009] Drone inspection is usually configured to capture speci fic parts , for example pictures or videos of a speci fic portion of a blade or of the tower . The absence of a more comprehensive but at the same time accurate overview of the whole platform, j oints , overall tolerances , including subsea elements on an of fshore wind turbine , impedes a complete structural integrity assessment and further prevents early detection of problems , for example by tracking and documenting from the assembly phase for either early detection or for later usage during a root-cause failure analysis . The need for accurate tolerance detection is further magnified by the demanding conditions of the of fshore setting, where environmental factors such as wind, waves , and saltwater corrosion present additional challenges to structural stability . The vast scale of of fshore wind turbines means that any discrepancy in the assembly tolerances can propagate through the structure , magni fying the impact and potentially leading to failures . This risk is particularly acute when it comes to the installation of heavy components high above the waterline , such as hydrogen platforms , where precision in assembly is not j ust a matter of ef ficiency, but of safety and longevity .

[0010] Known current drone solutions , taking videos and photos , while adept at capturing high-resolution images for visual inspection purposes , does not inherently possess the capability to measure and veri fy the fine tolerances required for ensuring the integrity of j oints , flanges , and other critical connection points . These limitations are not merely a matter of resolution or image clarity but stem from a lack of integration with systems capable of analyzing and interpreting complex spatial data to assess conformity with stringent engineering speci fications .

[0011] Moreover, the inability of existing technology to provide real-time digital tracking and assessment of component conditions throughout the logistical and assembly processes represents a signi ficant gap . This gap not only af fects the quality assurance during the installation but also complicates the maintenance and service operations that are crucial for the long-term operation of of fshore wind turbines . Without a detailed and precise digital record, veri fying compliance with design speci fications and operational tolerances becomes a challenging task, often leading to reactive rather than proactive maintenance strategies .

[0012] In addition, the industry lacks ef ficient digital tracking and documentation of component conditions throughout the stages of transportation, assembly, and post-installation . This deficiency can lead to challenges in accountability and insurance claims , as it becomes di f ficult to determine when damage to components occurred . The traditional reliance on manual measurements for large structures introduces potential for human error and inef ficiencies , further exacerbating the risk of undetected discrepancies that could lead to failures , especially in the demanding of fshore environment .

[0013] The of fshore wind industry currently lacks a standardi zed digital handover process that could provide detailed, autonomous , or semi-autonomous tracking of each component ' s j ourney from manufacture through to installation . There is a lack of quality control but also for streamlining the documentation process , facilitating insurance claims , and ensuring accountability at each stage of the turbine ' s li fecycle .

[0014] Addressing the multi faceted challenges faced by the of fshore wind turbine industry, a signi ficant gap exists in ef fectively detecting and documenting the critical tolerances and discrepancies throughout the transportation upon arriving to the onsite location, during assembly, and further for maintenance stages . The complexity and scale of these turbines , especially with the integration of substantial components like hydrogen platforms , demand an unprecedented level of precision in assembly and maintenance . However, the current industry practices , primarily focused on visual inspections of above-water parts , fall short in proactively addressing these stringent tolerance thresholds .

[0015] Despite the technological advancements in wind turbine design and construction, there remains a notable deficiency in systems or methodologies that can of fer a detailed, all-encompassing assessment spanning both visible and submerged turbine parts . This oversight not only poses risks to the structural integrity and operational ef ficiency of these colossal structures but also highlights a missed opportunity for early detection and resolution of potential issues . The lack of proactive solutions capable of navigating the intricate balance between the above and below water components of wind turbines underscores a critical need in the industry - a need for innovative approaches that can keep pace with the evolving complexities and demands of modern of fshore wind energy generation .

[0016] In addition, known solutions fail to allow a precise tracking, by digital means , of defects or tolerances of components prior, during or after assembly, particularly at j unctions where di f ferent components meet , wherein potential weak points may remain undetected during assembly, resulting in costly and challenging post-installation resolutions . This disadvantages aggregates in of fshore installations where the subsea foundation structures are not taken into consideration even when a drone is used for inspection .

[0017] Furthermore , identi fying damage occurring during transportation or assembly stages is challenging with current technologies , especially during of fshore assembly . Without the capability to detect subtle damages or alterations until after installation, there is an increased risk of safety hazards . This issue is further compounded by the inability to demonstrate whether components arrived damaged or were damaged during assembly and installation, impacting insurance and accountability .

[0018] In light of the complex challenges outlined, there emerges an urgent and unmet need within the of fshore wind turbine industry for a comprehensive solution . Without a detailed 3D visuali zation of the interconnected components , critical tolerance discrepancies that could af fect the entire structure might go undetected . Such oversights are particularly problematic given the increasing complexity and si ze of modern of fshore wind turbines , where even minor deviations can have magni fied consequences . The industry ' s current reliance on segmental inspections , rather than a holistic analysis , heightens the potential for overlooked issues that could compromise the safety, durability, and ef ficiency of these installations . Furthermore , the lack of a broader, more integrated overview of the components mounted or being mounted, and particularly one that includes critical but often underrepresented subsea components like monopiles , transition pieces , and caissons , results in a fragmented understanding of the turbine ' s overall structural integrity .

[0019] The aim of the invention is therefore to provide an autonomous 3D modeling and anomaly detection solution for a wind turbine , in particular for of fshore wind turbines , which overcomes at least some of the aforementioned disadvantages .

[0020] BRIEF SUMMARY

[0021] The obj ect of the invention is achieved by the independent claims . The dependent claims describe advantageous developments and modi fications of the invention . In accordance with a first aspect , there is provided an autonomous anomaly detection system for wind turbines , particularly advantageous for of fshore wind turbines .

[0022] At its core , the system may employ one or more unmanned vehicles . Each of these unmanned vehicles may equipped with an image capturing module and / or with a distance measuring module , enabling them to capture detailed images of wind turbine components and / or measure distances to and / or from these components .

[0023] More in particular, the present disclosure is related to an autonomous anomaly detection system that obtains data ( observation data ) , using the one or more unmanned vehicles , during the assembly or preassembly of wind turbine components . Then, the system compares said obtained data ( or observation data ) derived from the image captured and / or distance measured, to / with reference data to detect faulty conditions or anomalies in an assembly phase , in particular in various assembly phases . The way that data is obtained data ( or observation data ) is derived, and the format of reference data can vary across di f ferent embodiments . Below, it is outlined various exemplary implementations regarding how observation data is obtained or derived, and how it is compared against di f ferent types of reference data .

[0024] In one embodiment , observation data comprise raw or pre-processed images captured during assembly . The system compares these images against stored reference images in a database , which may include :

[0025] - Baseline images of defect- free components under controlled conditions .

[0026] - Annotated images of known faulty conditions to facilitate pattern matching . - Multi-angle reference images for more robust comparison .

[0027] - Historical images from prior assemblies to track manufacturing consistency .

[0028] The system may use image-processing algorithms , edge detection, and / or machine learning to highlight visual deviations between captured and reference images .

[0029] In another embodiment , the obtained data involves image-based feature extractions , such as keypoints , textures , or contour maps . Instead of comparing directly to reference images , the system matches these extracted features against a 3D reference model stored in the database . Thus , the reference data may include :

[0030] - CAD models of turbine components , which define expected geometries .

[0031] - Synthetic or simulated images generated from 3D models to improve robustness .

[0032] - Multi-dimensional feature maps based on historical assembly records .

[0033] The system may perform pro ection-based matching between real-world images , which may be trans form to a 3D model which is generated, and synthetic views of the reference 3D model .

[0034] In some embodiment as an alternative or in conj unction with the above , the generated data comprises distance measurements collected using LiDAR, structured light , or stereo imaging . The system compares these distances against stored reference dimensional data, which may be :

[0035] - Predefined tolerance values for each component . Point cloud models of correctly assembled components .

[0036] Reference depth maps to check for misalignment .

[0037] This embodiment may be particularly useful for ensuring mechanical fitment , alignment accuracy, and tolerance validation .

[0038] In some embodiments , the observation data comrpises a dynamically generated 3D model of the component , reconstructed from multiple images or sensor inputs . The system compares this generated model against :

[0039] - A CAD-based reference model defining the expected dimensions .

[0040] - A prior 3D scan of a correctly assembled component .

[0041] - A digital twin representation of the turbine component .

[0042] The system may employ surface matching, volumetric comparison, or geometric deviation analysis to identi fy anomalies .

[0043] In another embodiment , the observation data is processed through a trained machine learning model , which outputs any one of :

[0044] - A classi fication score indicating defect presence .

[0045] - Anomaly probability metrics based on learned patterns .

[0046] - Predicted deviation values compared to a reference standard .

[0047] The machine learning model may use historical labeled datasets as reference data, and the system may refine its detection accuracy over time by incorporating new observation data into its training set . In another embodiment , observation data from multiple assembly phases is logged and analyzed over time . The system compares current generated data with :

[0048] - Previous assembly records of the same turbine component .

[0049] - Statistical tolerances derived from a dataset of prior assemblies .

[0050] - Trend-based anomaly detection models that predict deviations based on historical performance .

[0051] This allows the system to detect progressive faults , degradation, or misalignment issues that develop over multiple assembly phases , improving process monitoring and traceability .

[0052] In each of the embodiments , the system is capable of identifying, detecting and / or determining the presence or absence of an anomaly or a faulty condition an assembly phases , in particular among a plurality of assembly phases . This may allow to determine whether each assembly phase meets quality standards or i f an issue— such as damage , misalignment , or a tolerance deviation has occurred at a particular point in the process .

[0053] This advantage is particularly relevant for accountability and can provide critical data for warranty claims , liability assessments , and / or insurance coverage ,

[0054] The system may further comprise an image processing module that is capable of processing captured data from the captured images , such as by generating a three-dimensional model of one or more components of the wind turbine . This may be derived from the captured images and / or distance measurements (observed data ) , providing a comprehensive and accurate representation of the turbine ' s physical state and assembly tolerances during an assembly phase , in particular capturing anomalies , discrepancy or faults , such as ones raising from the foundation level beneath the sea level .

[0055] The system may also include a data processing module , either integrated with or communicating with the image processing module . This data processing module or the image processing module itsel f may have access to a database containing reference data sets , the data may include dimensional data and / or visual representations of the one or more components at reference conditions .

[0056] In an example , the data may include 3D models of wind turbine components at reference conditions , allowing for a comparison between the generated models and the reference models . Such a comparison allows for identi fying any discrepancies or faulty conditions , for example tolerances , ensuring timely prevention of potential failures before assembly ( during preassembly) and during assembly, for example detecting the precise anomaly assembly phases during or j ust after the phase was performed during the assembly .

[0057] In some embodiments , the comparison may be performed between captured images and / or distance measurements , in particular postprocessed in such a way as to facilitate the comparison, with reference data . The reference data may include visual representations and / or dimensional data of the components at reference conditions , said comparison preferably repeated for multiple assembly phases , and providing as output a faulty assembly phase ( i f any) among the multiple assembly phases . I f a faulty assembly phase is not found, the solution provides allows for quality check, also certi fying that the assembly phases meet quality standards . Otherwise , i f a fault is found, a service action or a reassembly action, may be triggered, or accountability for further action is tracked.

[0058] A faulty assembly phase may be understood as a phase wherein one or more of the components assembled comprise an anomaly or faulty condition, for example a tolerance defect.

[0059] More specifically, the term "assembly phases" in the present invention is intended to broadly encompass different stages, steps, or checkpoints in the wind turbine assembly process and pre-assembly process (after transport but before starting assembly, such that potential damage on the components from transportation or from manufacturing, is detected before formally starting the assembly process) .

[0060] In an example, assembly phases refer to major construction stages of the wind turbine, where different large components are assembled such as tower erection phase, nacelle installation phase, blade attachment phase, etc.

[0061] In an example, assembly phases may refer to a sequence of steps performed within a broader assembly stage, ensuring that every critical action in a process is monitored, for example, during a bolting phase, a flange alignment phase, a torque verification phase, surface finishing phase (e.g. coating or surface quality processes) , etc.

[0062] In some embodiments, assembly phases may be defined as timebased monitoring intervals, where the system periodically checks component integrity at different moments in the assembly timeline. In example, these assembly phases may be:

[0063] - pre-assembly phase - inspecting components before they are used in construction, e.g. after transportation to the assembly location but before formally starting the assembly process .

[0064] - mid-assembly phase - e . g . checking alignment or positioning during assembly . final assembly veri fication - conducting a last check before commissioning .

[0065] For large-scale turbine production, components may be manufactured and partially assembled at di f ferent locations before final installation . Here , assembly phases may refer to di f ferent sites where the assembly process occurs rather than sequential steps . For example :

[0066] - pre-assembly - Detecting possible anomalies caused by handling or shipping, before final assembly on-site , for example directly in a vessel near the assembly site .

[0067] - On-site assembly phase - veri fying that transported components are installed correctly .

[0068] Within the meaning of the invention, reference conditions are referred to allowable conditions or acceptable tolerance levels of the components or set of components during the assembly phase .

[0069] More in particular, within the meaning of the invention, reference conditions may refer to predefined allowable parameters that determine whether a component or set of components meets acceptable quality, alignment , and structural integrity requirements during the assembly phase . These reference conditions may establish the permissible tolerance levels for dimensions , positioning, material integrity, fastening force , and other critical characteristics necessary to ensure proper assembly . Reference conditions may include dimensional tolerances , such as expected length, width, and / or thickness of components , allowable positioning errors when aligning parts , or maximum deviations in hole placement for bolts or fasteners . They may also define structural and mechanical conditions , including acceptable stress limits during tightening or fastening, proper torque values for bolts and j oints , and material flatness or curvature limits . Additionally, reference conditions may account for geometric and positional constraints , such as required angles , concentricity, or parallelism between connected parts , expected flange alignment before final tightening, or predefined blade pitch or rotor orientation .

[0070] Surface and coating conditions may also form part of the reference conditions , including permitted surface roughness before final finishing, allowed coating thickness for protective layers , or detection of cracks , dents , or irregularities . In some embodiments , reference conditions may further include electrical and sensor calibration parameters .

[0071] The system may access reference conditions from a predefined dataset , including CAD models , manufacturing standards , or historical turbine assembly records . The obtained data derived from the unmanned vehicles ( images and / or distances ) may be compared against these reference conditions to detect deviations , where any measured or observed value exceeding the reference tolerance may be flagged as an anomaly or faulty condition, requiring corrective action . Reference conditions may vary across di f ferent assembly phases , such as pre-assembly, where they may be based on component speci fications before integration, or final assembly, where they may include multi-component alignment criteria . In some embodiments , reference conditions may evolve dynamically, allowing for progressive tolerance adj ustments across di f ferent phases .

[0072] Overall , reference conditions may serve as a benchmark for ensuring that parts are positioned, fastened, and aligned within predefined tolerances , enabling the system to automatically detect and flag anomalies during assembly .

[0073] As aforementioned, the present invention contemplates the use of unmanned vehicles to capture data during the assembly of wind turbine components .

[0074] In some embodiments , unmanned aerial vehicles (UAVs ) , such as drones , are employed to capture images and / or distance measurements from turbine components , particularly at elevated positions or hard-to-reach areas , such as the nacelle or rotor assembly .

[0075] In other embodiments , unmanned underwater vehicles (UUVs ) , such as autonomous or remotely operated underwater drones , may be utili zed for the inspection of of fshore wind turbine foundations and submerged structures , where direct human access is limited .

[0076] Other unmanned vehicles can be contemplated by the skilled person, such as :

[0077] - Unmanned ground vehicles (UGVs ) , which may be wheeled, tracked, or legged robotic systems configured to navigate the assembly site , capturing images and / or measuring distances of turbine components at ground level , such as the tower base or pre-assembled sections before installation .

[0078] - Unmanned climbing robots , which may include magnetically attached crawlers , vacuum suction-based systems , or cable-driven robotic platforms , configured to ascend turbine towers , nacelles , or blade surfaces . Such robots may provide contact-based anomaly detection, such as structural integrity assessments , crack detection, or precise dimensional verification of assembled components .

[0079] - Hybrid multi-platform systems , wherein multiple unmanned vehicles are used in combination, such as a UAV for aerial blade assembly, a climbing robot for tower veri fication during assembly, and / or a UGV for ground-based component analysis . In some embodiments , the system may coordinate di f ferent vehicle types to cross-validate collected observation data, enhancing anomaly detection accuracy across various turbine sections .

[0080] The selection of a speci fic unmanned vehicle type may depend on the location of the turbine component under assembly, environmental factors , and the nature of the observation data required . Regardless of the embodiment , the unmanned vehicles operate in conj unction with image capturing and / or distance measuring modules , ensuring a detection process throughout s multiple assembly phases of the wind turbine , for proper tracking, and quality checks .

[0081] A signi ficance advantageous capabilities of the system may lie on its ability to facilitate precise three-dimensional models , 3D mapping and analysis of wind turbine components extending both bellow and above the surface of the sea, and further monitoring from an assembly phase associated j oints at the interface and beneath of the sea level or water surface and detecting anomalies , discrepancies , and fault conditions , in particular tolerance mismatching within allowable levels . The condition and tolerances of the subsea foundation components may have a big impact on the tower, nacelle , and rotor wind turbine components , capturing the interface and components from bellow and above , and may be important for maintaining their operational integrity and detecting anomalies from the assembly phase .

[0082] In some embodiments , the comparison may comprise a point cloud generation of both 3D models and further a divergence comparison . This involves converting both a generated 3D model and a 3D reference model into point clouds , where each model is represented as a set of vertices in a three-dimensional coordinate system . Techniques like Iterative Closest Point ( ICP ) algorithm may be used to align these point clouds and then compute the di f ferences between them . This may be particularly useful for detecting geometric discrepancies and deviations in shape or si ze .

[0083] In addition, or as an alternative of the above , the comparison may comprise a mesh comparison . For example , both 3D models are represented as meshes composed of vertices , edges , and faces . Techniques such as Hausdorf f distance measurement can be used to compute the maximum distance of a set of points ( from one model ) to the nearest points on the other model . This technique is ef fective for identi fying surface discrepancies , including warps or deformations .

[0084] In addition, or as an alternative of the above , the comparison may comprise a volumetric comparison . This technique involves analyzing the volume occupied by each 3D model . By segmenting the models into corresponding volumetric segments , one can compute and compare the volume of each corresponding segment . This is particularly useful for identi fying discrepancies in the bulk of components , which might be indicative of material wear or erosion .

[0085] In addition, or as an alternative of the above , the comparison may comprise a texture and material analysis : for components where surface texture or material properties are critical , techniques like texture mapping comparison can be applied . This involves examining the surface properties of the models to identi fy di f ferences in texture patterns or material characteristics , which can be indicative of surface wear or corrosion .

[0086] In addition, or as an alternative of the above , the comparison may comprise a feature-based comparison . This featurebased comparison may rely on identi fying key features ( like edges , corners , j oints , flanges or speci fic geometric patterns ) in both models and comparing these features in terms of their position, orientation, and scale . This feature-based technique may be used in combination with other techniques to provide a more comprehensive comparison .

[0087] In addition, or as an alternative of the above , the system may employ one or more machine learning algorithms , for example trained for image or pattern recognition ( like Convolutional Neural Networks ) . In particular, it may be trained to identi fy speci fic types of discrepancies or anomalies between the models , making the comparison more intelligent and adaptive . The system may preferably comprise a trained machine learning algorithm for underwater conditions and a trained machine learning algorithm for above surface conditions , each for the corresponding images captures beneath or above the sea level . Further embodiments employing one or more machine learning models will be explained further bellow . In some embodiments , complementing or as an alternative of the embodiments above , a heat map may be used . For example , after comparing the 3D models using one or more of the above techniques , a visual tool like a heatmap can be generated to visually represent the areas of discrepancy . Areas with no discrepancies can be marked .

[0088] The versatility of the system may be further enhanced by the possibility of including various types of unmanned vehicles , such as underwater vehicles and aerial vehicles (UAVs ) .

[0089] The underwater vehicles , for instance , could be Remotely Operated Underwater Vehicles (ROVs ) or Autonomous Underwater Vehicles (AUVs ) , each selected based on the speci fic environmental conditions and / or operational requirements . This flexibility may be important for thorough inspection and maintenance of wind turbines , particularly of of fshore wind turbine installation, which often have components located both underwater and above water i f the foundation is not a floating platform .

[0090] The captured / generated data and reference data within the meaning of the invention, may include at least two or more adj acently mounted components of a wind turbine , in particular dimensional data or visual representations including at least two or more adj acently mounted components of the wind turbine .

[0091] For example , the database may comprise reference three-dimensional models for these configurations , aiding in the detailed assessment of the connections and interactions between these components . This feature may be particularly beneficial for ensuring the structural integrity and ef ficient functioning of the wind turbine , especially in the areas where different components meet such a as flanges or joints, and for the overview of the structure whereupon tolerances which are linked to the whole assembly (aggregated from assembly tolerance mismatching of other parts) can be determined.

[0092] An optional embodiment of the system may involve the use of a machine-readable identifiers on each component, and more particularly on different portions of the wind turbine components. The unmanned vehicles may be configured to scan these identifiers, streamlining the process of tracking and analyzing various parts of the turbine. This approach not only enhances the organization of maintenance and inspection data but also contributes to a more targeted and efficient troubleshooting process.

[0093] In addition to the core functionalities, the system may further refine its analytical capabilities. For instance, the data processing module may utilize the scanned machine-readable identifiers for various purposes. These include accessing corresponding reference data of the wind turbine component scanned, for example, , pinpointing discrepancies, anomalies, or faulty conditions directly to the corresponding portion scanned, and / or issuing or recording compliance information of the assembly phase where the scanning took place. This multi-faceted use of data enhances the system's ability to provide detailed and actionable insights, which is advantageous for effective tracking and repair strategies. Furthermore, this may facilitate to leverage the computer resources, preempting for which portion or one component of the whole structure corresponds to the images and distances taken.

[0094] I.e., one or more components, or portions of a component may be labeled with a unique machine-readable identifier, and the unmanned aerial vehicle and underwater vehicle may be configured to scan these unique machine-readable identi fiers to identi fy the components thereof during an assembly phase .

[0095] In addition, the scanned unique machine-readable identi fier may be used by the system to document the location of any identi fied damage or anomaly in a digital record .

[0096] For example , a digital certi fication signal may emit and save an approval signal based on the comparison when the comparison results in no structural defects or anomalies , preferably said approval signal comprising a proof of compliance certi ficate . This may be integrated in a speci fic module configured to emit and receive information from the machine-readable identi fiers or may be integrated in a data processing module which may further post-process the images and execute the comparisons .

[0097] The machine-readable identi fier may be a QR code , a barcode , or any other equivalent suitable machine-readable identi fier, such further suitable machine-readable identi fiers will become apparent in view of the present description and drawings .

[0098] Each one or more unmanned vehicles may also comprise an image capturing module with diverse image capturing elements such as high-resolution digital cameras , infrared cameras , multi- spectral imaging sensors , and side-scan sonar . Each element provides a unique advantage , such as capturing images in varying light conditions or underwater environments , thereby broadening the scope of the system' s capabilities . One or more types of images capturing elements may be integrated in the image capturing module . Similarly, the distance measuring module may include a variety of distance measuring elements like Lidar devices, sonar devices, radar-based sensors, Ultra-Short Baseline (USBL) , and Doppler Velocity Logs (DVL) . The inclusion of these diverse technologies ensures that distance measurements are accurate and reliable regardless of environmental conditions. One of more of these distance measuring elements may be integrated in the distance measuring module.

[0099] In some embodiments, the system may comprise a data processing module which may incorporate or communicate with one or more machine learning algorithm. These algorithms may be trained using a dataset of previously captured images that include discrepancies and faulty conditions from various wind turbine components. The advantage of this feature is twofold: it allows the system to learn from past data, thereby improving its diagnostic capabilities over time, and it provides a more nuanced understanding of the range of potential issues that might affect wind turbine components. Furthermore, reference data (e.g. visual representations) , for example of the three-dimensional models, may be only used for its training, leveraging computer resources and efficacy of the comparison.

[0100] In an embodiment, the machine learning algorithm may be further trained using the reference data (e.g. 3D reference models) as benchmarks, enhancing its ability to identify deviations or faults. This continuous learning and adaptation may make the system increasingly efficient and accurate in anomaly detection.

[0101] In some embodiments, a redundant method may be used, using both the trained machine learning algorithm and a further comparison with the reference data. The comparison may comprise a tolerance comparison within, for example, a generated three-dimensional models and a reference three-dimensional models .

[0102] The machine learning algorithm may also be trained to recogni ze and categori ze di f ferent types of discrepancies , such as structural misalignments , wear and tear, or material defects . By distinguishing between these various types of issues , the system can provide more speci fic and useful information for future action, such as repair, leading to more targeted and ef fective interventions .

[0103] In a second aspect of the invention, it is provided an autonomous anomaly detection system, incorporating the one or more trained machine learning algorithms . Once it is trained, the system may not use the reference data anymore ( e . g . reference three-dimensional models ) , since the machine learning algorithm might be faster, consume less computational resources , and be more ef fective .

[0104] This embodiment of the system is similarly equipped with unmanned vehicles featuring image capturing and, optionally, distance measuring modules , along with the trained machine learning algorithms . The integration of the machine learning in this manner ensures that the system is adept at analyzing the data ( dimensional data and / or visual representations of the captured images ) to detect discrepancies or faulty conditions .

[0105] The machine learning algorithm in this embodiment may use reference three-dimensional models as part of its training process , serving as a benchmark . This ensures a high standard of accuracy in the algorithm ' s diagnostic capabilities . In some embodiments , the one or more machine learning algorithms comprises at least two machine learning modules , a first machine learning algorithm ( 226 ) trained at underwater conditions , configured to detect an absence or existence of a one or more anomalies or faulty conditions for an underwater component , and a second machine learning algorithm ( 224 ) trained at surface conditions , configured to detect an absence or existence of a one or more anomalies or faulty conditions for an above-water component .

[0106] The first machine learning algorithm may be trained to recogni ze discrepancies , anomalies or fault conditions in dimensional data or visual representations ( e . g . 3D model ) that are speci fically indicative of conditions encountered in underwater components , such as corrosion, biofouling, or structural damage unique to submerged environments .

[0107] In addition, or alternatively to the above , the first machine learning algorithm may be calibrated or trained to account for unique imaging characteristics associated with underwater environments , such as light refraction, varying visibility conditions , and potential distortions .

[0108] In some embodiments , a single machine learning algorithm is trained speci fically to in the fusion and analysis of 3D models representing components located both above and beneath the water surface . In particular, further trained on datasets with known anomalies or faulty conditions that include integrated 3D models featuring the interface between above water and beneath water components , to enhance its capability in identi fying discrepancies that may occur at or near the transition zone between these two environments . Similarly, the image processing module or the data processing module may incorporate speciali zed image preprocessing steps to enhance the quality of captured images derived from underwater images , including noise reduction, contrast adj ustment , and color correction techniques .

[0109] The data processing module in this context may be then further configured to perform a validation step, which includes comparing the captured / observed data ( optionally post processed) with the reference data and determining the presence of any discrepancies or faults based on this comparison .

[0110] This step may be particularly relevant for accountability in multi-step manufacturing processes and can provide critical data for warranty claims , liability assessments , and / or insurance coverage , allowing manufacturers to determine whether each assembly phase meets quality standards or i f an issue- such as damage , misalignment , or a tolerance deviation has occurred at a particular point in the process .

[0111] In a yet further aspect , it is provided a method of using the aforementioned system, in any of its various embodiments and aspect of the inventions herein explained . One embodiment includes a method step for assisting in the assembly of wind turbines , wherein the system may be employed during one or more assembly phases . This approach may allow for the identification of faulty assembly phases , involving discrepancies or conditions that deviate from the allowable tolerances or conditions . The advantageous features of this method, including the capability to identi fy and recti fy issues during the assembly phase , can be combined with other aspects or embodiments of the invention to enhance overall ef ficiency and reliability . The identi fication of satis factory or faulty assembly step, in other words , the identi fication and tracking during assembly, may be particularly relevant for accountability in multi-step manufacturing processes and can provide critical data for warranty claims , liability assessments , and / or insurance coverage , allowing manufacturers to determine whether each assembly phases meets quality standards or i f an issue— such as damage , misalignment , or a tolerance deviation has occurred at a particular point in the process .

[0112] Additionally, the disclosure also aids in assisting in the maintenance or service of of fshore wind turbines . Service or maintenance actions may be triggered, guided, and / or diagnosed based on the detailed information provided by the system such as a tolerance discrepancy, a fault condition or any other discrepancy detected during the determination .

[0113] More in particular, the method may comprise the steps of : capturing images and / measuring distances , using one or more unmanned vehicles ( 108 , 408 , 208 , 110 , 410 , 210 ) , obtaining data based derived from the captured images and / or measured distances , the generated data comprising dimensional data and / or visual representations , providing or accessing reference data, said reference data comprising dimensional data and / or visual representations of one or more components under a reference conditions , and comparing, the obtained data with the reference data, identi fying, based at least in part on the comparison, the absence or presence of one or more anomalies or faulty conditions in a plurality of assembly phases and / or pre-as- sembly phases .

[0114] In an example , the method may thus identi fy a faulty assembly phase ( i f any) among the multiple assembly phases analyzed . Otherwise , the method confirms that all assembly phase analyzed meet quality controls ( e . g . absence of anomalies in all phases analyzed) .

[0115] Therefore , the method according to the present disclosure , improving traceability and quality control in the assembly phases of wind turbine components , ensuring that faults are pinpointed precisely in time . In other words , this tracking during assembly, may allow manufacturers to determine whether each assembly phase meets quality standards or i f an issue- such as damage , misalignment , or a tolerance deviation— has occurred at a particular point in the process .

[0116] In a further embodiment , the method may comprise the steps of : operating one or more unmanned vehicles to capture images one or more components of the wind turbine , optionally, further measuring distances from and / or to the at least one component of the wind turbine , generating a three-dimensional model of the at least one component based at least in part on the captured images and / or distance measurements , providing or accessing one or more reference three-dimensional models ( 3DR) of the one or more components of the wind turbine , and determining, based at least in part of the generated three-dimensional model ( 3DG) and the one or more reference three-dimensional models , the absence or presence of one or more anomalies or faulty conditions in an assembly phases , in particular further configured for repeating the comparison across a plurality of assembly phases .

[0117] It ' s important to note that , within the meaning of the present invention, advantages and features of the method, can obviously be combined with other aspects of the invention such as the system or further aspects detailed next, and vice versa, unless explicitly stated otherwise.

[0118] In a yet further aspect, it is provided an offshore arrangement comprising one or more components of the wind turbine, such as a subsea foundation arrangement, a tower, and a vessel or any suitable boat equipped with communication and power delivery modules for the unmanned vehicles. This arrangement might further comprise a further vessel equipped with known tools for transporting and installing offshore wind turbines, or all corresponding equipment might be comprised in the same offshore arrangement transporting vessel.

[0119] In some embodiments, the arrangement may further comprise the autonomous system of the first aspect, as per any of the embodiments, which may be configured to identify discrepancies or faults in the subsea foundation arrangement, either before, during, or after an assembly step. Further, it may be further configured to detect anomalies, discrepancies, or faults on a component and / or one or more components extending across the water surface, namely, above and beneath thereof.

[0120] The integration of such a system into an offshore arrangement, with its capability to preemptively identify and address potential issues, is a significant advantage of the solution provided.

[0121] In addition, as stated for previous aspects, any component such as the transition piece, the monopile, the tower, inter alia, may comprise a unique machine-readable identifier, and more particularly located on different portions of said components. The unmanned vehicles may be configured to scan these identifiers, streamlining the process of tracking, and analyzing various parts of the turbine both bellow and above sea level . This approach not only enhances the quality tracking across assembly phases but also contributes to a more targeted and ef ficient troubleshooting process and future failure root analysis and maintenance operations .

[0122] Further, any previous stated features of the system, the method, and the arrangement , can obviously be combined therebetween, unless stated otherwise .

[0123] In yet a further aspect , it is provided a computing apparatus comprising a processor and a memory storing instructions that , when executed by the processor, configure for : receiving or accessing captured data, said captured data based at least in part on a captured images and / or measured distances received or accessed from an image capturing module and / or a distance measuring module ( 214 ) of one or more unmanned vehicles , determining the absence or presence of one or more discrepancies or faulty conditions in a plurality of assembly and / or pre-assembly phases , said determining based at least in part on : accessing a database ( 216 ) that comprises reference data of one or more wind turbine components under reference conditions , and comparing the received or accessed data with the reference data, and / or using a one or more trained machine learning algorithm ( 226 , 224 ) trained on datasets of a previously captured images comprising discrepancies , anomalies and / or faulty conditions of the one or more components and / or two or more components mounted thereof . In some embodiments , the computing apparatus comprises a processor and a memory storing instructions that , when executed by the processor, is configured for : generate or access data, the data comprising a generated 3D modelbeing based at least in part on images received from an image capturing module of one or more unmanned vehicles during assembly, determining the absence or presence of one or more discrepancies or faulty conditions in an plurality of assembly or preassembly phases , said determining based at least in part on any one of the following steps : accessing a database that includes one or more reference three-dimensional models of the one or more components of the wind turbine , and performing a comparison of the generated 3D model with the one or more reference three-dimensional models , and / or using a trained machine learning algorithm trained with three-dimensional models with known anomalies or faulty conditions with anomalies and / or faulty conditions , in particular further using the one more reference three-dimensional model as part of a training step of the trained machine learning algorithm .

[0124] In a yet further aspect , it is provided a non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to : retrieve or access observed data, the data observed derived from in either or both of : images captured from an image capturing module of one or more unmanned vehicles during assembly, and distance representations from a distance measuring module of one or more unmanned vehicles during assembly, determine the presence or absence of discrepancies or faulty conditions in a plurality of assembly or preassembly phases , based at least in part on : accessing a database containing a reference data, the reference data including visual representations and / or dimensional data, and comparing the captured data with the reference data ; and / or using a trained machine learning algorithm, trained on datasets of visual representations and / or dimensional data of the wind turbine components with known anomalies or faulty conditions in an a plurality of assembly or preassembly phases .

[0125] In an example , the computer-readable storage medium including instructions that when executed by a computer, cause the computer to : generate , a three-dimensional model of one or more components of a wind turbine based at least in part on images received or accessed from an image capturing module comprised in a one or more unmanned vehicles , determine , the absence or presence of one or more discrepancies or faulty condition in an assembly step, said determining based at least in part on any one of the following steps :

[0126] - accessing, a database that includes one or more reference three-dimensional models of the one or more components of the wind turbine , and performing a comparison of the generated 3D model with the one or more reference three-dimensional models , and / or

[0127] - using a trained machine learning algorithm trained with three-dimensional models with known anomalies or faulty conditions with anomalies and / or faulty conditions , in particular further using the one more reference three-dimensional model as part of a training step of the trained machine learning algorithm .

[0128] The invention ' s focus on a detailed, accurate data gathering, tracking, and anomaly detection analysis , reflects an advancement in the field of wind turbine failure analysis , assembly tolerances and deviation documentation, tracking of components conditions received after transport and during assembly, and for performing maintenance or service operations as a result of the data gathered by the system . Other purposes and advantages will become apparent from the teachings of the present description and figures .

[0129] BRIEF DESCRIPTION OF THE DRAWINGS

[0130] Embodiments of the invention are now described, by way of example only, with reference to the accompanying drawings , of which :

[0131] Figure 1 Shows an of fshore arrangement comprising a wind turbine , a subsea sub arrangement comprising in turn a monopile on a seafloor foundation, a vessel , and the system of Figure 2 comprising unmanned vehicles , according to an embodiment .

[0132] Figure 2 Shows the autonomous anomaly detection system for wind turbines installation, the operational workflow and modules interaction according to an embodiment thereof .

[0133] Figure 3 Shows a flowchart of an autonomous anomaly detection method, according to an embodiment . Figure 4 Shows the use of the autonomous anomaly detection system throughout various assembly phases of the offshore arrangement of Figure 1, according to an embodiment .

[0134] DESCRIPTION OF THE DRAWINGS

[0135] The illustration in the drawings is in schematic form. It is noted that in different figures, similar or identical elements may be provided with the same reference signs. Figures represent a way of carrying out exemplary embodiments and it is to be understood that the solution is not limited by the disclosed examples, and that numerous additional modifications and variations could be made thereto by a person skilled in the art without departing from the scope of the invention .

[0136] FIG. 1 shows an offshore arrangement (124) that includes an array of unmanned vehicles (108, 110) which form part of the autonomous 3D modeling and anomaly detection system (202) which is depicted in FIG. 2. The system (202) which will be explained in more detail further ahead may be part of the offshore arrangement.

[0137] Referring now to FIG. 1, at the top, the wind turbine includes a tower (102) supporting a nacelle (116) from which extend rotor blades (118) rotatable coupled thereof, collectively representing some of the above-water components of the wind turbine (112) . The nacelle (116) is located at the uppermost section of the tower (102) .

[0138] In the vicinity of the tower (102) , there is depicted an unmanned aerial vehicle (108) depicted in flight, equipped with both an image capturing module (206) and a distance measuring module (214) as illustrated in FIG. 2. This aerial vehicle

[0139] (108) is configured to capture images and measuring distances to and / or from components of the wind turbine (112) , particularly those extending above the sea surface, in the context of this exemplary embodiment, for the purpose of generating three-dimensional models and detecting any anomalies.

[0140] Submerged beneath the sea surface is a monopile (104) , which anchors the structure to the seafloor. The monopile (104) is rigidly connected to a transition piece (106) , which in turn supports the tower (102) . A caisson (414) is designed to be rigidly attached to both the monopile (104) and the transition piece (106) to provide additional stability to the structure. The transition piece 106 may be partly submerged and partly on the surface, being at least in part on the interface between the surface portion and the subsea portion.

[0141] An unmanned underwater vehicle (110) is illustrated next to the monopile (104) , similarly equipped with an image capturing module (206) and distance measuring modules (214) as the aerial vehicle (108) , tasked with capturing images and measurements of the subsea components of the wind turbine (112) .

[0142] A vessel (114) is shown on the water surface, which is equipped with a communication module. This vessel (114) serves as the operational base for transmitting data and / or electrical power to the underwater vehicle (110) , facilitating the autonomous operation of the underwater vehicle (110) .

[0143] FIG. 1. further illustrates that the offshore arrangement 124 may further comprise a platform (120) for hydrogen generation and storage (120) , which is attached to the tower (112) of the offshore wind turbine arrangement (124) . This platform (120) is positioned on the tower (102) above the water surface, enabling it to harness the mechanical energy produced by the wind turbine's rotor blades (118) for the purpose of hydrogen generation.

[0144] The platform (120) serves a dual function. Firstly, it contains equipment necessary for the electrolysis process, which splits water into hydrogen and oxygen using the electricity generated by the wind turbine (124) . This process of hydrogen generation is particularly advantageous for offshore wind turbines, as it provides a way for energy storage and transportation that is more efficient than electrical cabling, especially over long distances.

[0145] Secondly, the platform (120) may be equipped with storage facilities for the produced hydrogen. The hydrogen can be compressed or liquified, depending on the technology used, and stored in tanks supported by said platform (120) until it is ready to be transported. The vessel (114) may also serve as the means for this transport, collecting the stored hydrogen from the platform (120) and bringing it to shore or to another location for use.

[0146] The presence of the platform (120) necessitates rigorous oversight during assembly to ensure that all components are mounted within strict tolerance thresholds. This vigilance is not only pivotal during the assembly phase but also in the subsequent operational phases to maintain the structural integrity and functionality of the entire assembly as it adds a significant hundreds of tones to the structure overhanging just above sea level.

[0147] Ensuring the structural integrity and operational functionality of the hydrogen generation and storage platform (120) is of greatest importance, as any anomalies or faults could impact the overall efficiency and safety of the energy conversion and storage process and / or add vibrations or tolerances defects to the rest of components of the wind turbine 112.

[0148] The autonomous 3D modeling and anomaly detection system (202) if FIG. 2 plays a crucial role in this context, as it enables continuous monitoring and verification that the platform (120) and associated components are within acceptable tolerance levels, thus safeguarding against potential assembly errors or post-installation deformations that could affect the overall stability and efficiency of the wind turbine system.

[0149] More specifically, Figure 2 illustrates the operational workflow and component interaction within the autonomous 3D modeling and anomaly detection system (202) for wind turbines installations, particularly suited for offshore environments.

[0150] Unmanned vehicles (208, 210) , in an embodiment both an aerial vehicle (208) and an underwater vehicle (210) are encompassed, each equipped with an image capturing module (206) and a distance measuring module (214) . The image capturing module (206) is responsible for acquiring visual data of the wind turbine's components, while the distance measuring module (214) provides precise measurements from or to the one or more components (102,104,106, 116, 118, 120) for accurate spatial analysis and 3D modeling.

[0151] The acquired data from these modules are then processed by an image processing module (204) , which synthesizes the captured images and distance measurements to generate a three-dimensional model (3DG) of the wind turbine (112) . This model serves as a detailed digital representation of the physical state and assembly tolerances of one or more components (102,104,106, 116, 118, 120) .

[0152] The generated three-dimensional model (3DG) is then fed or accessed by a data processing module (222) , which communicates with a database (216) that contains reference three-dimensional models (3DR) . These reference three-dimensional models (3DR) represent the wind turbine (112) components (102,104,106, 116, 118, 120) under reference or ideal conditions, serving as a reference against which the generated three-dimensional model (3DG) may be compared.

[0153] Building upon the system outlined in Figure 2, the comparison between the generated 3D model (3DG) and the reference 3D models (3DR) may be a multi-faceted process that employs various analytical techniques to ensure comprehensive anomaly detection. This comparison can include point cloud generation, mesh comparison, volumetric analysis, texture and material analysis, and feature-based comparison. Such techniques allow for the detection of geometric discrepancies, surface irregularities, and volumetric deviations which might indicate conditions like wear, erosion, or assembly faults. The comparison may be devoid of any machine learning algorithms or used the machine learning algorithms maybe only used once trained for further checking. Furthermore, the comparison step may be only done throughout the training of the machine learning algorithms.

[0154] In some embodiments, for the purpose of anomaly detection and condition assessment, the system (202) , as illustrated in FIG.2, the system may employ two machine learning algorithms: the first machine learning algorithm (226) and the second machine learning algorithm (224) . Each algorithm is trained to recognize specific types of discrepancies, anomalies, or faulty conditions associated with the wind turbine components. The first machine learning algorithm (226) may be trained for analyzing underwater conditions and components, while the second machine learning algorithm (224) focuses on above-water conditions and components.

[0155] The machine learning algorithms may be trained using three- dimensional models with known anomalies or faulty conditions (3DK) . Through these steps, the system can detect deviations from the expected conditions, such as structural misalignments, wear, tear, or material defects.

[0156] Using at least two machine learning algorithms (224,226) , including the first machine learning algorithm (226) trained for underwater conditions and the second machine learning algorithm (224) trained for above-water conditions, play a pivotal role in enhancing the determination of anomalies or faulty conditions.

[0157] Upon identifying any discrepancies, the system can document the anomaly's location using machine-readable identifiers (126) as depicted in FIG. 1, which may be located on different portions of the wind turbine components. This documentation facilitates targeted maintenance operations and contributes to a more efficient root analysis for future maintenance and repairs.

[0158] The use of machine-readable identifiers (126) , such as QR codes or barcodes, is integrated into the system to streamline the tracking and analysis of the wind turbine components. Unmanned vehicles equipped with scanning capabilities can identify these identifiers during the survey or 3D modeling process, during preprocessing or post-processing, linking the captured images and measurements directly to the corresponding component. This facilitates the documentation of any identified damage or anomaly in a digital record, enhancing the efficiency of maintenance operations and the organization of inspection data.

[0159] More in particular, in some embodiments, the one or more unmanned vehicles are configured to scan machine-readable identifier (126) to perform any one or more of the following steps : accessing and / or selecting a reference three-dimensional model (3DR) from the database (216) , integrating two or more provisionally generated three- dimensional models into a merged generated three-dimensional model ( 3DG) , pinpointing one or more locations of discrepancies or faulty conditions identified, and / or issuing or recording a compliance or a non-compliance information to the corresponding machine-readable identifier (126) based on a discrepancy or a faulty conditions detection outcome for the scanned one or more components (102,104, 106,116, 118, 120) or part thereof .

[0160] The preferred way of carrying out the above-mentioned steps, preferably, if available, by the system (200) , will be explained in greater detail next, with the aid of Figures 3 flowchart .

[0161] More specifically, FIG. 3 outlines a method (300) for autonomous 3D modeling and anomaly detection of one or more components (102,104, 106, 116, 118, 120) of the wind turbine (112) .

[0162] As will be appreciated, some embodiments of the present invention may be embodied as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment , an entirely software embodiment ( including firmware , resident software , microcode , etc . ) or an embodiment combining software and hardware aspects all generally referred to herein as a "circuit" , "module , " or " system" . Furthermore , the present invention may take the form of a computer program product on a computer-usable storage medium having computer-usable program code embodied in the medium . As used herein, the terms " software" or " firmware" are interchangeable and include any computer program stored in memory for execution by a processor, including RAM memory, ROM memory, EPROM memory, EEPROM memory, and nonvolatile RAM (NVRAM) memory . The above memory types are exemplary only and are thus not limiting as to the types of memory usable for storage of a computer program .

[0163] Any suitable computer readable medium may be utili zed . The computer-usable or computer-readable medium may be , for example but not limited to , an electronic, magnetic, optical , electromagnetic, infrared, or semiconductor system, apparatus , de-vice , or propagation medium . More speci fic examples ( a non-exhaustive list ) of the computer-readable medium would include the following : an electrical connection having one or more wires , a portable computer diskette , a hard disk, a random access memory (RAM) , a read-only memory (ROM) , an erasable programmable read-only memory (EPROM or Flash memory) , an optical fiber, a portable compact disc read-only memory ( CD- ROM) , an optical storage device , a transmission media such as those supporting the Internet or an intranet , or a magnetic storage device . Note that the computer-usable or computer- readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for instance , optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, i f necessary, and then stored in a computer memory. In the context of this document, a computer-usable or computer-readable medium may be any medium that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0164] Computer program code for carrying out operations of the present invention may be written in an object-oriented programming language such as Java?, Smalltalk or C++, or the like. However, the computer program code for carrying out operations of the present invention may also be written in conventional procedural programming languages, such as the "C" programming language, or a similar language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a re-mote computer or entirely on the remote computer. In the latter scenario, the re-mote computer may be connected to the user's computer through a local area network (LAN) or a wide area network

[0165] (WAN) , or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider) .

[0166] More specially, referring back to Figure 3, step (302) initiates the method (300) , where one or more unmanned vehicles are operated to capture images of at least a component of the wind turbine (112) . These components could include, but are not limited to, the tower (102) , nacelle (116) , blades (118) , monopile (104) , transition piece (106) , or any other relevant component of a wind turbine.

[0167] Step (304) involves the measurement of distances from and / or to the at least one component of the wind turbine. This step (302) is important for acquiring accurate spatial data necessary for constructing the 3D model.

[0168] At step (306) , the method (300) generates a three-dimensional model (3DG) of the at least one component based on the captured images and / or distance measurements. This model serves as a detailed representation of the current physical state of the wind turbine components.

[0169] At step (308) , the method comprises accessing or providing one or more reference three-dimensional models (3DR) .

[0170] Next, step (310) is a decision-making point where the method determines the absence or presence of one or more discrepancies or faulty conditions in the one or more components. This is where the detailed analysis for anomaly detection occurs.

[0171] For doing such determination at step (310) , the depicted Method (300) involves two complementary alternatives, depicted as branches into two pathways at step (310) , which can be conducted in tandem or as alternatives:

[0172] Pathway (312) involves performing a comparison of the generated three-dimensional model (3DG) with one or more reference three-dimensional models (3DR) . These reference models represent the expected state of the components under normal, fault-free conditions. The comparison may be done without a trained machine learning model using one or more of the techniques aforementioned and other suitable techniques.

[0173] Pathway (314) utilizes a trained first machine learning algorithm (226) that has been previously trained with known three-dimensional models with anomalies and faulty conditions (3DK) . This algorithm can intelligently analyze the generated model (3DG) to identify patterns or features characteristic of known issues, thus assisting in the detection of anomalies or faulty conditions.

[0174] Both pathways provide a robust solution, and both can be combined to increase the quality and reliability of the output for the anomaly detection.

[0175] Figure 4 depicts various phases of installation within an offshore arrangement (124=, showcasing how the autonomous 3D modeling and anomaly detection system (200) can be implemented at different phases of assembly to ensure the integrity of components and the accuracy of the assembly process.

[0176] The leftmost portion of Figure 4 provides a system's (200) pre-installation overview of the supporting platforms for hydrogen generation and storage (420) , for assessing its condition prior to the commencement of the installation process. This pre-installation inspection is aimed at ensuring there has been no damage during transportation and establishing accountability for the component's state before it is assembled in the wind turbine.

[0177] The damage or anomalies detection may be performed by step (308) of the method as depicted in Figure 3. Similarly, the system (202) may reproduce said method according to any of the embodiments, at every step of the assembly processes explained next.

[0178] The isolated representation of the platform (420) in the topleft image (408) signifies a detailed pre-assembly model which might be used as a standard for inspection. This may be performed not only for the platform which exemplifies it, but for any one or more components of the wind turbine, The system can utilize the generated model (3DG) , the reference three-dimensional models (3DR) and / or the machine learning algorithms (226, 224) to meticulously examine any component of the 412 for detecting any signs of damage, such as cracks, deformations, or any other anomalies that could have occurred during transportation. For example, by capturing images and measuring distances from critical points of the platform, the system can compare these real-time data points against the undamaged reference model to ascertain the platform's integrity.

[0179] The system might focus on high-stress points, connection interfaces, or other areas susceptible to damage during handling and transit. This detailed inspection helps in establishing a clear accountability trail, documenting the component's condition upon arrival at the installation site.

[0180] The inspection process at this pre-assembly phase serves several advantages. It serves as a safeguard against the incorporation of compromised components into the offshore wind turbine structure, which could lead to weakened structural performance or failure. By confirming that the platform is free from transport-related damages, the system ensures that the assembly begins on a solid foundation, both literally and figuratively. This proactive approach in the early phase of assembly (i.e. pre-assembly phase) supports the overall quality assurance of the construction process and contributes to the long-term reliability of the offshore wind turbine installation. Furthermore, the problem of accountability to detect where the component was damaged, this is, if during transportation or during installation, is solved. FIG. 4 further depicts a monopile (404) of a subsea sub-arrangement. The system (202) making use of the underwater vehicle (410) is capable of executing steps of the method 300 to detect the present or absence of anomalies or faulty conditions of the monopile 404, further checking flanges and segments of the monopile (404) before a further assembly phase thereat.

[0181] Directly below this, it is depicted a detailed view of a subassembly, focusing on a joining portion or flange wherein a caisson (414) is mounted between the monopile 404 and the transition piece (406) . Again, the system 202 making use of the underwater vehicle (410) is capable of executing steps of the method (300) as aforementioned for FIG. 3, to detect the present or absence of anomalies or faulty conditions any of these components, namely, so far, the monopile (404) , the transition piece (406) and the caisson (414) .

[0182] Each section may have a machine-readable identifier (126) , which enable the unmanned vehicles (408,410) to recognize and catalog each part accurately. These machine-readable identifiers (126) allow the system (202) to track the components through different assembly phases and ensure each is assembled in the correct sequence and orientation.

[0183] To the right, a detailed view shows the interface between the monopile (104) and the transition piece (106) , focusing the joint areas which may be monitored and checked for potential deficiencies. Here, the system may check the alignment and fitting of the flanges or connection points, confirming that they match the reference model and that there are no anomalies or misalignments that could compromise structural integrity. Above this, an assembly phase is depicted, where an aerial vehicle 408 equipped with an image capturing module (206) and a distance measuring module 214 is configured to generate a 3D model, and for further anomalies or faulty condition determination, this is, for example, the alignment and the condition of components as they are joined. This real-time inspection may be important to detect any discrepancies during the actual assembly process.

[0184] The top-right image shows a complete assembly phase, where the fully erected tower (402) or monopile (404) is in place. Here, the system (202) performs a comprehensive scan of the entire structure, comparing it against the reference three- dimensional models to ensure that the final assembly conforms to design specifications and that no new defects have arisen during the assembly process.

[0185] Each detailed view provides a close-up inspection of critical junctions, such as where the transition piece 406 meets the tower 402. The system's ability to analyze these points ensures that all seals, bolts, and other fastening mechanisms are properly installed and that there are no gaps, misalignments, or other issues that could lead to failures in the future .

[0186] In summary, Figure 4 illustrates the use of the autonomous 3D modeling and anomaly detection system (202) throughout the entire assembly process of an offshore arrangement (124) .

[0187] From the initial condition check of individual components to the final assembly verification, the system ensures that each phase of the assembly process is executed to the highest standards possible, preventing potential issues that could affect the wind turbine's performance and lifespan. The offshore arrangement (124) comprising the system (202) , the system (202) , and the associated method (300) , for 3D modeling and anomaly detection, in particular aided by machine-readable identifiers (126) and machine learning algorithms (224,226) specifically trained as aforementioned, represents a significant advantage in the field of wind turbine maintenance and assembly. It ensures a correct autonomous tracking from an early phase or pre-assembly phase, before installation, during installation and during operation, with any potential issues being identified and addressed promptly, autonomously, and digitally. This solution becomes increasingly important particularly for offshore platforms, whose dimension, weights and integration with hydrogen platforms requires abiding by critical tolerance thresholds during assembly and digital tracking of the components from an early stage for correct accountably and future failure root assessments .

[0188] It should be noted that the use of "a" or "an" throughout this application does not exclude a plurality, and "comprising" does not exclude other steps or elements. Also, elements described in association with different embodiments may be combined. It should also be noted that reference signs in the claims should not be construed as limiting the scope of the claims .

Claims

CLAIMS1. An autonomous anomaly detection system (202) to detect anomalies or fault conditions during assembly and / or pre-as- sembly phases of a wind turbine (412) , the system comprising: one or more unmanned vehicles (408, 108, 208, 110, 410, 210) each comprising:- an image capturing module (206) configured to capture images of a one or more wind turbine components, and / or- a distance measuring module (214) configured to measure distances from and / or to the one or more wind turbine components, the system further comprising a processing module (204) configured to obtain data derived from the captured images and / or the measured distances, the obtained data comprising dimensional data and / or visual representations of the wind turbine components, a database (216) storing reference data, said reference data including dimensional data and / or visual representations of the wind turbine components under reference conditions, wherein the processing module (204) or a further a data processing module (222) is configured to: compare, the generated data with the reference data, and identify, based at least in part on the comparison, the presence or absence of anomalies or faulty conditions in a plurality of assembly phases and / or preassembly phases.

2. The system (202) of claim 1, wherein the pre-assembly phase comprises a phase after transportation and before assembly, during which the system is configured to determine the presence or absence of anomalies or faulty conditions inthe one or more wind turbine components prior to assembly, in particular for identifying potential faults or anomalies that may have arisen during transportation or manufacturing.

3. The system (202) of any one of claims 1 to 2, wherein the processing module (204) is configured to obtain data derived from the images captured, by generating a three-dimensional model (3DG) of the one or more wind turbine components, .

4. The system (202) of claim 3, wherein the database (216) comprises one or more reference three-dimensional models (3DR) of the one or more wind turbine components, such that the reference data comprises said one or more reference three-dimensional models (3DR) .

5. The system (202) of claim 4, wherein the processing module (204) is configured to perform a comparison of the generated three-dimensional model (3DG) and the one or more reference three-dimensional models (3DR) , in particular the comparison involving at least a tolerance comparison within the models (3DG, 3DR) .

6. The system (202) of claim any one of claims 1 to 5, wherein the one or more unmanned vehicles (408, 108, 208, 110, 410, 210) include an underwater vehicle (110, 410, 210) and / or an aerial vehicle (108, 408, 208) .

7. The system (202) of any one of claims 1 to 6, wherein the generated data comprises dimensional data and / or visual representations of at least two or more adjacently mounted wind turbine components, and the database (216) comprises one or more reference data comprising dimensional data and / or visual representations which include at least said two or more adjacently mounted wind turbine under reference conditions.

8. The system (202) of claim 7, wherein the two or more adjacently mounted components comprise an underwater component captured by the underwater vehicle (110, 410, 210) , and an above-water component captured by the aerial vehicle (108, 408, 208) .

9. The system (202) of any one of claims 1 to 8, wherein the data processing module (222) or the processing module (204) , incorporates or is communicated with a machine learning algorithm (226, 224) trained using a dataset of a previously captured images comprising discrepancies, anomalies and / or faulty conditions of the one or more components (102,402,104, 404, 106, 406, 414, 116, 416, 118, 418, 120, 420) and / or two or more components mounted thereof.

10. An autonomous anomaly detection system (202) , incorporating one or more trained machine learning algorithms (226, 224) , the system (202) comprising: one or more unmanned vehicles (408, 108, 208, 110, 410, 210) , each comprising: an image capturing module (206) configured to capture images of one or more components (102,402,104, 404, 106, 406, 414, 116, 416, 118, 418, 120, 420) of a wind turbine (412) , wherein the one or more machine learning algorithms (226, 224) is trained on a dataset of images comprising discrepancies and / or faulty conditions from the wind turbine components , wherein the one or more trained machine learning algorithms (226, 224) is configured to: determine, based at least in part on the captured images, the presence or absence of anomalies or faultyconditions in a plurality of assembly phases and / or pre-as- sembly phases.

11. The system (202) of claim 10, wherein the one or more machine learning algorithms (224,226) comprises: a first machine learning algorithm (226) trained at underwater conditions, configured to detect an absence or existence of a one or more anomalies or faulty conditions for an underwater component, and / or a second machine learning algorithm (224) trained at surface conditions, configured to detect an absence or existence of a one or more anomalies or faulty conditions for an abovewater component.

12. An offshore arrangement (124) , comprising: a subsea sub-arrangement (122) comprising a monopile (104,404) and a transition piece (106,406) intended to be rigidly mounted to the monopile (404, 104) , in particular the subsea sub-arrangement (122) further comprising a caisson (414) intended to be rigidly attached to the monopile (104, 404) and to the transition piece (106, 406) , a tower (102, 402) intended to be securely mounted on the transition piece (106, 406) , a vessel (114) or boat, equipped with a communication module configured to transmit data and / or electrical power to a one or more unmanned underwater vehicle (110, 410, 210) , wherein the offshore arrangement (122) further comprises the system (202) of any one of claims 1 to 11.

13. A method (300) for detecting anomalies during assembly or pre-assembly of a wind turbine, the method comprising: capturing images and / measuring distances, using one or more unmanned vehicles (108, 408, 208, 110, 410, 210) ,obtaining data based derived from the captured images and / or measured distances , the generated data comprising dimensional data and / or visual representations , providing or accessing reference data, said reference data comprising dimensional data and / or visual representations of one or more components under a reference conditions , and comparing, the obtained data with the reference data, identi fying, based at least in part on the comparison, the absence or presence of one or more anomalies or faulty conditions in a plurality of assembly phases and / or pre-as- sembly phases .14 . The method of claim 13 , further comprising performing a maintenance , service or re-assembly action, which is triggered, guided and / or diagnosed by the data obtained .15 . A computing apparatus comprising a processor and a memory storing instructions that , when executed by the processor, configure the apparatus to the steps of : receiving or accessing captured data, said captured data based at least in part on captured images and / or measured distances received or accessed from an image capturing module and / or a distance measuring module ( 214 ) of one or more unmanned vehicles , determining the absence or presence of one or more discrepancies or faulty conditions in a plurality of assembly and / or pre-assembly phases , said determining based at least in part on : accessing a database ( 216 ) that comprises reference data of one or more wind turbine components under reference conditions , and comparing the received or accessed data with the reference data, and / orusing a one or more trained machine learning algorithm ( 226 , 224 ) trained on datasets of a previously captured images comprising discrepancies , anomalies and / or faulty conditions of the one or more components and / or two or more components mounted thereof .

Images