Fan regular inspection control method and system based on cooperation of plc and robot

Abstract

The application discloses a fan regular inspection control method and system of PLC and robot cooperation, which sets an initial threshold value by using an edge computing gateway in combination with PLC macro environment data, collects sensor data and attitude information in real time during robot operation, and executes online quality evaluation. For visual data, a weighted evaluation mechanism based on three-dimensional geometric projection fidelity is particularly introduced, distortion Jacobian graph is used to correct the evaluation deviation caused by surface projection, and the quality score is ensured to truly reflect the observation condition of the physical surface. Based on the quality state mark obtained in real time, action intention is automatically decided and generated, and the motion speed of the robot, the sensor parameter or the micro-correction path is dynamically adjusted. This mechanism enables the inspection system to adaptively respond to environmental changes such as light and surface stains, effectively improving the robustness and data effectiveness of fan automatic inspection.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and more specifically, to a method and system for periodic inspection control of wind turbines using a PLC and robot collaboration. Background Technology

[0002] Wind turbines are typically deployed in harsh environments, such as in the field or offshore. Their blades, as the core components for capturing wind energy, are subjected to long-term wind and sand erosion, lightning strikes, temperature fluctuations, and high-intensity mechanical loads, making them highly susceptible to structural defects such as cracks, corrosion, and lightning damage. To ensure the safe and stable operation of wind turbines, regular inspection and maintenance of the blades is crucial. With the development of industrial automation technology, using crawling robots or drones equipped with non-destructive testing equipment to replace manual labor for high-altitude operations has become an industry trend.

[0003] However, existing PLC (Programmable Logic Controller) and robot collaborative control technologies exhibit poor environmental adaptability when facing complex and ever-changing field environments. Current robot motion control logic typically relies on open-loop control based on preset static parameters and fixed paths. This means that the robot cannot perceive and respond to changes in the microenvironment of the blade surface during inspection tasks. For example, when localized icing, oil contamination, or drastic changes in external lighting conditions occur on the wind turbine blade surface, the coupling effect of ultrasonic sensors decreases, and the image captured by the vision camera may be overexposed or lose detail. Due to the lack of a real-time online evaluation and feedback mechanism for the quality of acquired data, the robot cannot dynamically adjust its motion posture, speed, or sensor parameters based on the current detection quality, often continuing to blindly execute the predetermined trajectory. This rigid control method results in low validity of inspection data acquired in harsh environments, easily leading to missed defects or false alarms. Furthermore, in the quality assessment stage of visual inspection of blades, existing technologies often process two-dimensional image data directly without fully considering the geometric distortion and nonlinear mapping relationship generated when the three-dimensional curved surface of the blade is projected onto the two-dimensional image plane. This results in an inaccurate assessment of image clarity and effective field of view, making it difficult to truly reflect the robot's observation quality of the physical surface of the blade, and further restricting the reliability and intelligence level of the automated inspection system.

[0004] Therefore, an optimized control scheme for periodic inspection of wind turbines in collaboration with PLCs and robots is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for periodic inspection control of wind turbines using a PLC and robot collaboration.

[0006] According to one aspect of this application, a method for periodic inspection control of wind turbines in collaboration with a PLC and a robot is provided, comprising: The edge computing gateway loads the inspection task file and requests macroscopic environmental data from the PLC to obtain the operating control parameters and inspection quality thresholds. The robot controller drives the robot to move toward the current waypoint based on motion control parameters and collects robot posture and position data and main inspection sensor data. Online inspection quality indicators are calculated and evaluated based on the main inspection sensor data to obtain real-time quality scores and quality status indicators; Based on the quality status flag, adaptive control decisions are made on the current motion control parameters and the current robot posture and position data to obtain adaptive control commands.

[0007] According to another aspect of this application, a PLC-robot collaborative wind turbine periodic inspection control system is provided, comprising: The data acquisition module is used by the edge computing gateway to load the inspection task file and request macro-environment data from the PLC to obtain the operation control parameters and inspection quality thresholds. The robot control and data acquisition module is used by the robot controller to drive the robot to move to the current waypoint according to the motion control parameters and to collect robot posture and position data and main inspection sensor data. The online inspection quality index calculation and evaluation module is used to calculate and evaluate the quality index of the main inspection sensor data online in order to obtain real-time quality score and quality status indicator; The adaptive control decision module is used to make adaptive control decisions based on the quality status flag, the current motion control parameters, and the current robot posture and position data to obtain adaptive control commands.

[0008] Compared to existing technologies, this solution addresses the technical problems of uncontrollable inspection quality and lack of adaptive adjustment capabilities in complex environments by constructing a closed-loop control system based on real-time data quality feedback. It utilizes an edge computing gateway combined with PLC macroscopic environmental data to set initial thresholds, collects sensor data and posture information in real time during robot operation, and performs online quality assessment. For visual data, a weighted evaluation mechanism based on 3D geometric projection fidelity is specifically introduced. This mechanism corrects evaluation biases caused by curved surface projections using a distortion Jacobian chart, ensuring that the quality score accurately reflects the observed physical surface condition. Based on the real-time quality status indicators, the system automatically makes decisions and generates action intentions, dynamically adjusting the robot's speed, sensor parameters, or triggering micro-correction paths. This mechanism enables the inspection system to adaptively respond to environmental changes such as lighting and surface contamination, effectively improving the robustness and data validity of automated wind turbine inspections. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of a PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow in the PLC and robot collaborative wind turbine periodic inspection control method according to an embodiment of this application; Figure 3 This is a flowchart illustrating the online calculation and evaluation of inspection quality indicators from the main inspection sensor data to obtain real-time quality scores and quality status indicators in the PLC and robot collaborative wind turbine periodic inspection control method according to an embodiment of this application. Figure 4 A flowchart illustrating the PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application, which calculates the gradient energy aggregation value of the image data as the real-time quality score in response to the main inspection sensor data being image data; Figure 5 This is a flowchart illustrating the adaptive control command obtained by making adaptive control decisions based on the current motion control parameters and the current robot posture and position data, according to the PLC and robot collaborative periodic inspection control method of the embodiment of this application. Figure 6 This is a block diagram of a PLC and robot collaborative wind turbine periodic inspection control system according to an embodiment of this application. Detailed Implementation

[0011] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0012] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0013] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0015] Existing PLC-robot collaborative wind turbine inspection technologies mostly employ open-loop control logic, lacking the ability to perceive and respond to subtle changes in the working environment. This leads to robots mechanically executing preset instructions even in complex on-site conditions such as fluctuating lighting, surface icing, or stains, easily causing a decline in inspection data quality or even misjudgment. This rigid control mode cannot assess the current detection effectiveness in real time, making it difficult for the system to maintain stable detection performance in dynamic environments. Therefore, this application proposes a PLC-robot collaborative wind turbine periodic inspection control method. First, it uses an edge computing gateway combined with macroscopic environmental data acquired by the PLC to set a dynamic quality threshold. During the robot's inspection process, it synchronously collects the robot's posture and position data with the main sensor in real time and performs online quality index calculations. To address the evaluation bias caused by curved surface projection in visual inspection, this solution introduces a weighted evaluation mechanism based on three-dimensional geometric fidelity, using a distortion Jacobian map to correct image gradient features, thereby obtaining a real-time score that truly reflects the observed quality of the physical surface. Based on this real-time quality status, the robot can intelligently identify the quality of the current working conditions and make adaptive decisions accordingly. It can dynamically generate adjustment instructions to change the robot's motion parameters or trigger specific micro-correction actions, thereby ensuring that high-quality inspection data can still be obtained in a variable environment. This achieves a technological leap from fixed trajectory execution to environmental adaptive perception.

[0016] Figure 1 This is a flowchart of a PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in a PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application. Figure 1 and Figure 2As shown, the PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application includes the following steps: S100, the edge computing gateway loads the inspection task file and requests macroscopic environmental data from the PLC to obtain operating control parameters and inspection quality thresholds; S200, the robot controller drives the robot to move towards the current waypoint according to the motion control parameters and collects robot posture and position data and main inspection sensor data; S300, online inspection quality index calculation and evaluation are performed on the main inspection sensor data to obtain real-time quality score and quality status flag; S400, based on the quality status flag, adaptive control decision is made on the current motion control parameters and the current robot posture and position data to obtain adaptive control instructions.

[0017] Specifically, in step S100, the edge computing gateway loads the inspection task file and requests macroscopic environmental data from the PLC to obtain operating control parameters and inspection quality thresholds. It is understandable that, due to the extremely complex and rapidly changing outdoor environment of wind turbines, relying solely on preset static parameters to start the inspection task often fails to adapt to actual on-site lighting, temperature, or wind speed conditions, easily leading to robot instability or improper initial sensor parameter settings, resulting in invalid data or even equipment damage. Therefore, in the technical solution of this application, the edge computing gateway loads the inspection task file and requests macroscopic environmental data from the PLC to obtain operating control parameters and inspection quality thresholds, thereby establishing a control benchmark and quality evaluation standard that matches the current physical environment from the very beginning of the task. This ensures that the robot enters the inspection process in an optimized initial state, effectively avoiding operational risks in the initial stage caused by environmental factors, and laying a solid foundation for subsequent high-quality data acquisition.

[0018] More specifically, in a concrete example of this application, the edge computing gateway first reads a pre-configured inspection task file through a local storage interface. This file defines in detail the target blade number, predetermined trajectory coordinate sequence, and required sensor type configuration for this inspection. Subsequently, the edge computing gateway actively initiates a data request to the wind turbine's main control PLC using the industrial Ethernet communication protocol, acquiring real-time macroscopic environmental data including outside nacelle temperature, relative humidity, and instantaneous wind speed. The sampling frequency is set to 1Hz to ensure data real-time performance and synchronization. After acquiring the above data, the gateway's internal parameter configuration algorithm dynamically corrects the default parameters in the task file based on the environmental data. When the PLC reports an outside nacelle temperature below zero degrees Celsius and high humidity, the algorithm determines there is a risk of icing and automatically reduces the robot's initial moving speed setting while simultaneously increasing the signal-to-noise ratio (SNR) threshold of the ultrasonic sensor echo signal to address potential ice interference. Finally, the calculated set of operating control parameters, including specific speed limits, and the inspection quality threshold, including a lower SNR limit, are loaded into the motion control module and quality assessment module, respectively, completing the initialization preparation before the inspection.

[0019] Specifically, in step S200, the robot controller drives the robot to move towards the current waypoint according to motion control parameters and collects robot attitude and position data and main inspection sensor data. It is understandable that, since the surface of wind turbine blades is an unstructured, complex free-form surface, and the accuracy of defect location directly determines the efficiency of subsequent maintenance work, simply collecting sensor data without accurate spatial attitude correlation will result in the inability to accurately reproduce the physical coordinates of the defect on the blade. Furthermore, the robot's motion behavior must strictly follow the specific control commands generated based on environmental conditions in the preceding steps to ensure operational safety. Therefore, in the technical solution of this application, the robot controller further drives the robot to move towards the current waypoint according to motion control parameters and collects robot attitude and position data and main inspection sensor data, thereby achieving spatiotemporal synchronous acquisition of multi-source heterogeneous data while executing the predetermined inspection trajectory. This ensures that each frame of inspection data has accurate geospatial attributes and provides a raw data stream containing complete spatial context for subsequent online quality assessment.

[0020] More specifically, in a concrete example of this application, the robot controller first parses the received motion control parameters and the coordinates of the current path waypoint to be executed. It then uses an inverse kinematics algorithm to calculate the required rotational speed and torque commands for each joint or drive wheel, and drives the servo motor to propel the robot body steadily along the blade surface towards the target waypoint. During this movement, data acquisition commands are synchronously triggered according to the sampling frequency set in the motion control parameters. On one hand, by reading the encoder feedback values ​​of the servo motors and the readings of the onboard inertial measurement unit, the robot's six-degree-of-freedom attitude and position data in the blade coordinate system are calculated in real time, clarifying the precise spatial orientation and angle at the current moment. On the other hand, the onboard main inspection sensor is synchronously triggered to perform operations, such as controlling the vision camera to expose and capture image information of the blade surface when reaching a designated position, or controlling the ultrasonic probe to excite and receive sound wave signals. Finally, the robot attitude and position data collected at the same time are bound to a unified timestamp with the main inspection sensor data, and integrated into a spatiotemporally aligned data packet for output.

[0021] Specifically, in step S300, the main inspection sensor data is used to calculate and evaluate online inspection quality indicators to obtain real-time quality scores and quality status indicators. It is understood that environmental interference factors such as changes in lighting, surface stains, or icing at the wind turbine inspection site directly affect the physical coupling characteristics or imaging quality of the sensors. If unverified raw data is directly used for subsequent defect analysis, it is highly likely to lead to missed detections or false alarms due to signal distortion or image blurring. Furthermore, traditional offline analysis methods cannot promptly detect and remedy such data acquisition failures. Therefore, in the technical solution of this application, the main inspection sensor data is further calculated and evaluated online to obtain real-time quality scores and quality status indicators. This constructs a real-time quality feedback loop between data acquisition and motion control, transforming the previously invisible validity of detection data into quantifiable numerical indicators and clear status levels. This enables the system to instantly perceive the impact of the current working environment on sensor performance, providing accurate quantitative basis for subsequent adaptive control decisions. This avoids the robot continuously operating under invalid observation conditions, ensuring that each acquisition action obtains inspection data with high confidence.

[0022] Figure 3 This document presents a flowchart illustrating the process of determining noise level estimation based on denoised input and raw sensor data in a PLC-robot collaborative wind turbine periodic inspection control method according to embodiments of this application. Figure 3As shown, step S300 includes: S310, in response to the main inspection sensor data being ultrasonic waveform data, calculating the signal-to-noise ratio of the ultrasonic waveform data as the real-time quality score; S320, in response to the main inspection sensor data being image data, calculating the gradient energy aggregation value of the image data as the real-time quality score; S330, determining the quality status indicator based on the comparison between the real-time quality score and the inspection quality threshold.

[0023] In step S310, in response to the main inspection sensor data being ultrasonic waveform data, the signal-to-noise ratio (SNR) of the ultrasonic waveform data is calculated as the real-time quality score. It is understood that, due to the accumulation of dirt, ice, or roughness degradation on the surface of wind turbine blades during long-term operation, these changes in physical surface conditions directly damage the acoustic coupling interface between the ultrasonic probe and the blade, leading to drastic fluctuations in the effective echo signal amplitude or its being submerged by background noise, causing amplitude-based defect judgment to fail. Therefore, in the technical solution of this application, in response to the main inspection sensor data being ultrasonic waveform data, the SNR of the ultrasonic waveform data is calculated as the real-time quality score, thereby quantifying the current acoustic coupling efficiency and signal clarity in real time, transforming the abstract waveform quality into a comparable numerical indicator. This allows for the immediate identification of low-quality data frames caused by poor contact or environmental noise before the signal enters the defect analysis stage, preventing the misjudgment of an echo-free state caused by coupling failure as a defect-free state, and ensuring the authenticity and reliability of the internal structural flaw detection data.

[0024] More specifically, in a specific example of this application, in response to the main inspection sensor data being ultrasonic waveform data, the signal-to-noise ratio of the ultrasonic waveform data is calculated as the real-time quality score, including: extracting the effective echo signal amplitude and background noise signal amplitude from the ultrasonic waveform data; and calculating the signal-to-noise ratio of the ultrasonic waveform data based on the effective echo signal amplitude and background noise signal amplitude.

[0025] Accordingly, the effective echo signal amplitude and background noise signal amplitude are extracted from the ultrasonic waveform data. It is understandable that since wind turbine blades are typically composed of multi-layered composite materials, their complex internal layered structure generates inherent material scattering noise. Combined with electromagnetic interference introduced by the operation of on-site electrical equipment, the original ultrasonic data inevitably contains background clutter. Without precise signal component separation, it is difficult to discern the true strength and effectiveness of the echo signal. Therefore, in the technical solution of this application, the effective echo signal amplitude and background noise signal amplitude are further extracted from the ultrasonic waveform data. This decouples and quantifies the deterministic signal representing effective detection information from the random noise representing environmental interference in the time domain. This provides physically meaningful parameter inputs for subsequent signal-to-noise ratio (SNR) calculations, ensuring that the quality assessment results objectively reflect the propagation loss and coupling status of the sound beam in the medium.

[0026] More specifically, in a concrete example of this application, the signal processing unit inside the edge computing gateway first performs time-domain gating processing on the received discretized A-Scan ultrasonic waveform data. Based on a blade material sound velocity and thickness model, the unit defines a signal monitoring time window on the waveform axis covering the range where the expected defect echo or bottom wave appears. A peak-finding algorithm is used to search for the maximum voltage amplitude within the window and lock it as the effective echo signal amplitude. Simultaneously, in the silent section after the waveform's initial pulse and before the signal monitoring time window, or in the tail wave section after the bottom wave, an independent noise sampling time window is defined. The peak value or root mean square value of the signal amplitude within this window is statistically analyzed to determine the background noise signal amplitude. Based on this, the extracted effective echo signal amplitude and background noise signal amplitude are calculated according to a preset logarithmic ratio formula to obtain a precise value characterizing the acoustic signal-to-noise ratio of the current detection point.

[0027] Accordingly, the signal-to-noise ratio (SNR) of the ultrasonic waveform data is calculated based on the effective echo signal amplitude and the background noise signal amplitude. It is understandable that due to the multi-layered structure of the wind turbine blade composite material and the complexity of the on-site electromagnetic environment, relying solely on the absolute voltage amplitude of the signal cannot objectively evaluate the effectiveness of the detection data. This is because a strong signal under high gain may be accompanied by extremely high structural scattering noise, thus masking the true small defect echo and resulting in an excessively low SNR that fails to meet imaging requirements. Therefore, in the technical solution of this application, the SNR of the ultrasonic waveform data is further calculated using a specific logarithmic operation model based on the effective echo signal amplitude and the background noise signal amplitude. This transforms the absolute voltage value, which is originally affected by hardware gain, into a relative quality index reflecting signal contrast. This allows for a normalized quality metric for detection data with different gain settings and different thicknesses, ensuring sufficient discriminability between obstacle echoes and background noise, thereby providing high-confidence input data for subsequent automatic defect identification algorithms.

[0028] More specifically, in a particular example of this application, the computing unit within the edge computing gateway substitutes the two key voltage feature values ​​extracted in the preceding steps into a preset signal-to-noise ratio (SNR) calculation model. This model uses a logarithmic scale to process the signal-to-noise ratio, thereby expanding the dynamic range and conforming to industry standards for acoustic measurements. Specifically, the SNR of the ultrasonic waveform data is determined using the following formula:

[0029] in, The ultrasonic signal-to-noise ratio is expressed as a real-time quality score, in decibels. This represents the effective echo signal amplitude extracted from the signal gate in the previous step. This represents the amplitude of the background noise signal extracted from the noise gate. For example, when an inspection robot moves to the blade sparsity area for detection, if the probe is well coupled to the surface, the extracted internal interlayer echo... It is 0.8 volts, while the matrix material scattering noise is... It is 0.04 volts, calculated according to the above formula. A score of approximately 26 decibels is high enough to directly trigger an excellent status indicator, allowing the robot to maintain its current speed and continue operating. Conversely, if uneven distribution of the coupling fluid due to localized surface roughness reduces the effective echo... If the noise level is attenuated to 0.08 volts while remaining constant, the calculation result will drop to 6 dB. This low score will be immediately identified as unusable data, triggering an adaptive control command that requires the robot to reduce its speed and increase the probe pressure for retesting.

[0030] In step S320, in response to the main inspection sensor data being image data, the gradient energy aggregation value of the image data is calculated as the real-time quality score. It is understandable that, due to the complex free-form surface shape of wind turbine blades, simply performing gradient aggregation processing on the original image data during visual inspection of wind turbine blades fails to fully consider the projection geometry from the three-dimensional blade surface to the two-dimensional image. This nonlinear mapping between physical space and the imaging plane leads to significant differences in the physical surface area represented by pixels in different regions of the image due to lens distortion and surface shape differences. Furthermore, using a preset region of interest fixes the focus on the image region, failing to adjust the imaging range and distortion degree of the focused physical region in the image in real time according to the dynamic changes in the robot's posture.

[0031] Therefore, in the technical solution of this application, in response to the main inspection sensor data being image data, the gradient energy aggregation value of the image data is calculated as the real-time quality score. In this process, a QoS adaptive weighted evaluation mechanism based on the geometric projection fidelity of the three-dimensional surface is introduced to quantify the geometric fidelity of pixel information during the three-dimensional to two-dimensional projection process, and the gradient features are weighted and aggregated based on this. This avoids the drawback of directly aggregating the gradient energy of all pixels, which is equivalent to mixing data with mismatched physical meanings, generating a QoS score that is closer to physical reality. This makes the quality evaluation results of the system's inspection data less susceptible to interference from geometric projection effects. For example, when the robot moves to a high-curvature area at the leading edge of a blade to take a picture, even if the image edges are stretched and blurred due to the tilted viewpoint, this mechanism can automatically reduce the weight of these low-fidelity areas, focusing on the central effective area with less geometric distortion. This ensures that the output quality score truly and accurately reflects the fine quality of the blade's physical surface, thus reliably guiding subsequent adaptive control.

[0032] Figure 4 This is a flowchart illustrating the PLC-robot collaborative wind turbine periodic inspection control method according to an embodiment of this application, which calculates the gradient energy aggregation value of the image data as the real-time quality score in response to the main inspection sensor data being image data. (See flowchart for example.) Figure 4 As shown, step S320 further includes: S321, determining the distortion Jacobian map based on robot posture and position data, camera intrinsic parameters, and blade 3D model; S322, performing parallel calculations of pixel-level geometric fidelity weights and gradient features on the image data based on the distortion Jacobian map to obtain a geometric fidelity weight map and a pixel gradient map; S323, determining the gradient energy aggregation value of the image data based on the geometric fidelity weight map and the pixel gradient map.

[0033] In step S321, a distortion Jacobian map is determined based on robot posture and position data, camera intrinsic parameters, and the 3D model of the blade. It is understood that due to the nonlinear and complex curved surface geometry of wind turbine blades, when the robot's vision camera captures images of the blades at close range, the pixel resolution of the central and edge regions of the image is inconsistent in physical space due to limitations in the imaging angle and the curvature of the blade surface. If this projection distortion is not quantified and corrected, it will lead to subsequent misjudgments of defect size and texture density. Therefore, in the technical solution of this application, a distortion Jacobian map is further determined based on robot posture and position data, camera intrinsic parameters, and the 3D model of the blade to quantify the degree of geometric distortion in the mapping relationship between 3D blade surface points and 2D image pixels. This allows the system to accurately understand the actual area proportion represented by each pixel in the image space on the 3D physical surface, providing a basis for subsequent weight calculations. This establishes a geometric bridge between the 3D physical world and 2D imaging, providing a basis for geometric correction for subsequent quality assessment.

[0034] More specifically, in a concrete example of this application, the geometry processing unit within the edge computing gateway performs the following operation: First, by inputting robot pose data (including spatial coordinates and three-axis Euler angles), camera intrinsic parameters (including focal length, principal point coordinates, and distortion coefficients), and a blade 3D model, a mapping function from the blade's 3D surface coordinates to the image's 2D pixel coordinates is established based on the pinhole camera model and rigid body transformation theory. Specifically, the blade surface points... Using the robot pose matrix Transformed into camera coordinate system points Combined with the camera intrinsic parameter matrix Projected to pixel coordinates This process is equivalent to simulating the real camera imaging optical path in digital space. Then, the Jacobian matrix mapped to any point on the 3D surface is calculated. In a mathematical and physical sense, the Jacobian matrix describes the scaling transformation that occurs when a tiny area element on a 3D curved surface is projected onto a 2D image plane. Next, the determinant of this matrix is ​​calculated. This determinant directly quantifies the area scaling factor of the region; a value deviating from 1 represents the degree of projection distortion—the larger the value, the larger the area corresponding to the pixel on the physical surface, and vice versa. For example, when the camera is directly facing the blade sparsity plane, the Jacobian determinant is close to 1, indicating minimal imaging distortion; however, when the line of sight passes over the highly curved leading edge of the blade, the determinant value increases significantly, meaning that the image pixels at that location are stretched, with each pixel covering an excessively large physical area, leading to a decrease in effective resolution. Finally, the purpose of this step is to output a detailed distortion Jacobian map, providing geometric transformation information for each pixel during the projection process. This map will be passed as a pixel-by-pixel confidence index table to the subsequent feature extraction module.

[0035] In step S322, based on the distortion Jacobian map, pixel-level geometric fidelity weights and gradient features are calculated in parallel on the image data to obtain a geometric fidelity weight map and a pixel gradient map. It is understandable that, during the visual inspection of wind turbine blades, the two-dimensional image is merely a projection of the three-dimensional physical surface. If each pixel in the image is treated indiscriminately, the imaging distortion caused by viewing angle tilt and surface curvature will be ignored, leading to severely stretched or compressed pixel areas misleading the final image quality assessment. Therefore, in the technical solution of this application, pixel-level geometric fidelity weights and gradient features are further calculated in parallel on the image data based on the distortion Jacobian map to obtain a geometric fidelity weight map and a pixel gradient map. This obtains the geometric transformation information of each pixel, which is then converted into correction weights, while simultaneously extracting the basic visual features of the image. In this way, a weighted map that indicates the physical correlation of information of each pixel in the image can be generated, and the original gradient features can be obtained for weighted aggregation in subsequent steps. This ensures that only those pixels that truly reflect the physical surface texture details contribute significantly to the quality score, and eliminates interference data caused by geometric projection distortion.

[0036] More specifically, in a concrete example of this application, the parallel computing unit within the edge computing gateway simultaneously processes the input raw image stream and the distorted Jacobian map generated in the preceding steps. Specifically, the geometric fidelity weights are first calculated. Read the value at the corresponding pixel position in the Jacobian map, and then calculate the value by using the determinant of the Jacobian matrix. The value is compared with 1, and a decay function is applied to obtain the pixel. The geometric fidelity weight. In a physical sense, this weight reflects the degree of compression or stretching of the physical region represented by the pixel in the image, assigning higher weights to regions with less distortion. For example, the flat area in the middle of a blade has minimal projection distortion and is assigned a high weight; while the tip or leading edge of the blade with large curvature is assigned a low weight due to severe projection distortion.

[0037] In step S323, the gradient energy aggregation value of the image data is determined based on the geometric fidelity weight map and the pixel gradient map. It is understood that in the actual scenario of visual inspection of wind turbine blades, image edges or areas with high curvature are often accompanied by severe projection distortion. Simply averaging the gradient of the entire image would incorrectly include noise or blurry information from these distorted areas in the quality assessment system, resulting in a score that cannot represent the true level of physical observation. Therefore, in the technical solution of this application, the gradient energy aggregation value of the image data is further determined based on the geometric fidelity weight map and the pixel gradient map. This combines the geometric fidelity weight generated in the previous step with the pixel gradient features to calculate a QoS score that truly reflects the inspection quality of the physical surface. In this way, through geometric correction, the QoS score no longer simply favors the average gradient of the image center or edge, but better represents the observation sharpness on the physical surface. This weighted aggregation mechanism ensures that gradient information from areas with high geometric fidelity on the physical surface (i.e., areas with small projection distortion) plays a more important role in the final QoS score, while low-fidelity information from severely distorted areas is suppressed. This avoids directly aggregating distorted pixel information, making the calculated QoS score more stable and accurate in reflecting the physical detail observation quality of the blade surface, and effectively reducing the false alarm and missed detection rates caused by environmental factors such as insufficient lighting, reflection, dirt, or icing.

[0038] More specifically, in a concrete example of this application, the image processing unit within the edge computing gateway performs weighted gradient aggregation and geometric correction QoS score calculation. This unit iterates through each pixel in the image matrix, multiplying each gradient value in the pixel gradient map by its corresponding weight in the geometric fidelity weight map. This process essentially performs a physical validity screening of the visual information. For example, when a robot is photographing the main beam area of ​​a blade, the tiny crack in the center of the image is significantly magnified due to its high fidelity weight, while the stretching blur caused by curvature at the image edges is ignored due to its extremely low weight. Then, the weighted gradient values ​​of all pixels are summed and divided by the sum of all weights to obtain the final geometric correction QoS score. The gradient energy aggregation value of the image data is determined by the following formula:

[0039] In this formula, The visual QoS score (i.e., gradient energy aggregation value) refers to the inspection quality score calculated based on visual sensors. This indicates an accumulation operation on all coordinate points in the two-dimensional matrix of the image. Represents pixels The geometric fidelity weight, which is derived from the calculation results of the distorted Jacobian plot, Represents pixels The gradient magnitude reflects the texture sharpness at that point. This calculation outputs a geometrically corrected QoS score that is unaffected by significant geometric distortion and accurately reflects the physical inspection quality, providing a reliable quality metric for subsequent adaptive control.

[0040] In step S330, the quality status flag is determined based on the comparison between the real-time quality score and the inspection quality threshold. It is understood that since the real-time quality score calculated in the preceding steps is essentially a continuously fluctuating numerical signal, directly inputting it into the logic controller makes it difficult to define clear action switching boundaries and cannot directly correspond to specific control strategies. The control system requires discrete logical states with clear semantics. Therefore, in the technical solution of this application, the quality status flag is further determined based on the comparison between the real-time quality score and the inspection quality threshold, thereby discretizing the continuously changing physical quality metric into logical state levels with clear action guidance significance. This simplifies the complex numerical analysis results into state inputs that can be directly called by the decision tree, enabling the system to quickly identify whether the current state is transitioning from excellent to suboptimal or falling to failure, thus providing a precise logical trigger source for subsequent hierarchical adaptive control.

[0041] More specifically, in a concrete example of this application, the status determination unit within the edge computing gateway first reads the inspection quality threshold set set based on macroscopic environmental data during the task initialization phase. This set specifically includes an excellent threshold for distinguishing high-quality data and a minimum tolerance threshold for distinguishing usable data. Subsequently, the determination unit executes multi-level numerical comparison logic, comparing the currently calculated ultrasonic signal-to-noise ratio or image gradient energy aggregation value with the aforementioned thresholds in real time. When the real-time quality score is higher than the excellent threshold, the determination unit determines that the current acquisition conditions are ideal, assigns the quality status flag to excellent, and instructs the system to maintain the current efficient operating mode. When the real-time quality score is between the excellent threshold and the minimum tolerance threshold, the determination unit identifies that environmental interference has caused a decline in data quality but is still within an acceptable range, and assigns the quality status flag to suboptimal, warning of potential risks. When the real-time quality score is lower than the minimum tolerance threshold, the determination unit confirms that the current data cannot be used for subsequent defect analysis due to a severely low signal-to-noise ratio or image blur, and immediately assigns the quality status flag to failure. Finally, this explicit enumerated status flag is output to the control decision module as the core judgment basis for the next control cycle.

[0042] Specifically, in step S400, based on the quality status flag, adaptive control decisions are made on the current motion control parameters and the current robot posture and position data to obtain adaptive control commands. It is understandable that traditional wind turbine inspection robots typically employ a pre-programmed open-loop control mode, lacking a real-time response mechanism to sudden environmental changes. When faced with localized strong light interference, surface icing, or dirt causing a sharp decline in data quality, if the robot mechanically maintains its original motion trajectory and speed, it will inevitably lead to missed defects in critical areas or the collection of a large amount of invalid data unusable for analysis. Therefore, in the technical solution of this application, adaptive control decisions are further made on the current motion control parameters and the current robot posture and position data based on the quality status flag to obtain adaptive control commands. This constructs an intelligent decision-making center connecting the perception layer and the execution layer, transforming abstract data quality evaluation results into specific equipment action adjustment strategies in real time. This ensures that the system can immediately and autonomously intervene when it detects fluctuations or failures in data quality, dynamically generating deceleration optimization, parameter fine-tuning, or local re-inspection search instructions based on the current position. This enables the robot to proactively adapt to the working environment and self-correct, maximizing the integrity of inspection tasks and the validity of data under complex dynamic working conditions.

[0043] Figure 5 This is a flowchart illustrating the adaptive control command obtained by making adaptive control decisions based on current motion control parameters and current robot posture and position data using a PLC-robot collaborative periodic inspection control method according to an embodiment of this application. Figure 5 As shown, step S400 further includes: S410, determining the action intention based on the quality status flag; S420, generating action instructions based on the action intention, current motion control parameters, and current robot posture and position data; S430, performing adaptive instruction synthesis on the action intention and action instructions to obtain adaptive control instructions.

[0044] In step S410, the action intent is determined based on the quality status flag. It is understandable that since the quality status flag is merely a diagnostic label for the validity of the current data, it does not directly contain tactical action instructions for the control system. Directly passing the diagnostic results to the underlying drive unit would lead to a high degree of coupling between decision-making and execution logic, hindering the expansion and maintenance of complex control strategies. Therefore, in the technical solution of this application, the action intent is further determined based on the quality status flag, thereby establishing a semantic mapping layer between status diagnosis and specific instruction generation, transforming different quality levels into clear high-level tactical objectives. This ensures that the control system first determines, from a macro-strategic perspective, whether to maintain, optimize parameters, or initiate an emergency recovery procedure, providing a clear logical guide for the subsequent generation of specific numerical instructions.

[0045] More specifically, in a concrete example of this application, the control decision module maintains a preset state-intention mapping table, which defines the types of response strategies the system should adopt under different quality states. The module first reads the input quality state flag and immediately performs logical branch matching. When the flag is excellent, the decision module recognizes that the current operating environment is stable and the data quality meets the requirements, and thus locks the action intention to continue normal execution, meaning the system will maintain its current trajectory and acquisition rhythm. When the flag changes to suboptimal, it indicates that environmental interference has begun to affect data quality but has not yet caused complete data failure. The module then determines the action intention to adjust motion parameters, aiming to counteract environmental noise by fine-tuning speed or gain. When the flag is identified as failure, it means that the current acquisition parameters can no longer cope with the adverse operating conditions. The module immediately escalates the action intention to execute a recovery action, which instructs the system to interrupt the current main path task and enter a special troubleshooting or recoupling search mode.

[0046] In step S420, action commands are generated based on the action intent, current motion control parameters, and current robot posture and position data. It is understood that since the action intent is merely an abstract strategy direction, and the robot's underlying servo drive system cannot directly execute abstract concepts, it requires speed commands containing specific values ​​or path sequences containing precise coordinates. Furthermore, any adjustment must be calculated relative to the robot's current motion state and spatial position; blind adjustments detached from the current state can easily lead to sudden motion changes or collisions. Therefore, in the technical solution of this application, action commands are further generated based on the action intent, current motion control parameters, and current robot posture and position data. This instantiates the macroscopic tactical intent into specific instruction code that the robot controller can directly parse and execute. This ensures that adaptive adjustments not only conform to the strategy's expectations but also maintain kinematic continuity and safety, achieving precise implementation from decision logic to physical execution.

[0047] More specifically, in a concrete example of this application, the instruction generation module first receives a determined action intention and retrieves the real-time updated current motion control parameters and robot posture position data from memory based on the intention type. When the action intention is identified as adjusting motion parameters, the module reads the current set velocity value, multiplies it using a preset attenuation factor (e.g., an adaptive coefficient with a value between 0.6 and 0.8), and calculates a new velocity value after reduction to suppress sensor noise. Simultaneously, if the current main inspection sensor is a vision camera, the module can also synchronously increase the exposure time or adjust the gain parameters to compensate for the amount of light received, and encapsulate this as a parameter update instruction. Conversely, when the action intention is identified as performing a recovery action, the module uses the currently read robot posture position coordinates as the geometric center and plans a small-range grid-shaped or spiral search path on the blade surface, aiming to find the optimal ultrasonic coupling point or visual focus plane again through subtle pose changes. This micro-correction task, containing a fine waypoint sequence, is then encapsulated as a subroutine call instruction. Finally, the generated instructions are pushed to the send queue, ready to be sent to the actuator.

[0048] In step S430, the action intent and action command are adaptively synthesized to obtain adaptive control commands. It is understood that since the action commands generated in previous steps typically only contain specific parameter values ​​or path coordinate data, lacking execution logic attributes (such as command priority, whether to interrupt the current action, or only smooth updates in the background) that instruct the robot controller how to schedule these data, directly issuing the original commands may cause the robot to be unable to correctly distinguish between emergency avoidance actions and routine parameter fine-tuning, leading to motion conflicts or response delays. Therefore, in the technical solution of this application, the action intent and action command are further adaptively synthesized to obtain adaptive control commands. This encapsulates the abstract strategic intent and specific control data into a complete execution package conforming to the industrial fieldbus protocol, and assigns corresponding execution permissions and timing logic to different types of commands. This ensures that the robot controller accurately understands the tactical level of immediate intervention commands, correctly suspends the current task within a millisecond-level time window to prioritize micro-corrections, or seamlessly switches sensor parameters without interrupting the main motion, ensuring the stability and safety of adaptive control.

[0049] More specifically, in a concrete example of this application, the instruction synthesis module first receives a determined action intention and its corresponding action instruction content. When the action intention is to perform a recovery action and the action instruction contains a micro-correction path sequence, the module identifies this as a high-priority task and synthesizes the instruction into a specific interrupt-insertion control message. The highest-level execution mask is placed in the message header, forcing the robot controller to immediately suspend the current main path interpolation operation and prioritize processing the micro-correction sequence. Conversely, when the action intention is to adjust motion parameters and the action instruction only involves updating speed or gain values, the module synthesizes the instruction into an immediate update message, marks it as background non-blocking processing, and instructs the robot controller to smoothly transition to the new parameter settings while maintaining the current trajectory tracking accuracy. For intentions to continue normal execution, the module synthesizes a heartbeat hold or no-operation instruction to ensure the communication link remains active without interfering with normal operation. Finally, this synthesized adaptive control instruction is sent to the robot controller for execution via a real-time Ethernet communication port.

[0050] In summary, the PLC-robot collaborative wind turbine periodic inspection control method according to the embodiments of this application is explained. It solves the problems of uncontrollable inspection quality and lack of adaptive adjustment capability in complex environments by constructing a closed-loop control system based on real-time data quality feedback. It utilizes an edge computing gateway combined with PLC macroscopic environmental data to set initial thresholds, collects sensor data and posture information in real time during robot operation, and performs online quality assessment. For visual data, a weighted evaluation mechanism based on 3D geometric projection fidelity is specifically introduced. Distortion Jacobian diagrams are used to correct evaluation biases caused by curved surface projection, ensuring that the quality score truly reflects the observed physical surface condition. Based on the real-time quality status indicators, automatic decision-making and action intentions are generated, dynamically adjusting the robot's movement speed, sensor parameters, or triggering micro-correction paths. This mechanism enables the inspection system to adaptively respond to environmental changes such as lighting and surface contamination, effectively improving the robustness and data validity of automated wind turbine inspection.

[0051] Furthermore, a PLC-robot collaborative wind turbine periodic inspection control system is also provided.

[0052] Figure 6 This is a block diagram of a PLC-robot collaborative wind turbine periodic inspection control system according to an embodiment of this application. Figure 6As shown, the PLC-robot collaborative wind turbine periodic inspection control system 100 according to an embodiment of this application includes: a data acquisition module 110, used to load the inspection task file by the edge computing gateway and request macroscopic environmental data from the PLC to obtain operating control parameters and inspection quality thresholds; a robot control and data acquisition module 120, used to drive the robot to move to the current waypoint according to the motion control parameters and collect robot posture and position data and main inspection sensor data; an online inspection quality index calculation and evaluation module 130, used to perform online inspection quality index calculation and evaluation on the main inspection sensor data to obtain real-time quality score and quality status flag; and an adaptive control decision module 140, used to perform adaptive control decision on the current motion control parameters and current robot posture and position data based on the quality status flag to obtain adaptive control instructions.

[0053] As described above, the PLC-robot collaborative wind turbine periodic inspection control system 100 according to the embodiments of this application can be implemented in various types of computing devices or control units. For example, it can be deployed in an edge computing gateway within the wind turbine nacelle or tower base, a programmable logic controller responsible for wind turbine logic control, or a high-performance industrial computer integrated into a wind farm remote operation and maintenance center. In one possible implementation, the PLC-robot collaborative wind turbine periodic inspection control system 100 according to the embodiments of this application can be integrated into the computing device as a software module and / or hardware module. For example, the PLC-robot collaborative wind turbine periodic inspection control system 100 can be an intelligent collaborative control application within the operating system of the computing device. This software module is configured to perform spatiotemporal synchronous acquisition of robot posture and main inspection sensor data, online inspection quality index calculation and evaluation based on the weighted gradient of ultrasonic signal-to-noise ratio and image geometric fidelity, action intent determination based on real-time quality status indicators, and adaptive motion control parameter adjustment and micro-correction path generation based on multi-source heterogeneous data feedback. Alternatively, it can be a dedicated wind turbine automated inspection control algorithm program developed for the computing device. Of course, the PLC and robot-coordinated wind turbine periodic inspection control system 100 can also be one of the many hardware modules of the computing device or control unit, or it can be embedded in a field-programmable gate array circuit to accelerate the construction of the image distortion Jacobian matrix and pixel-level gradient feature extraction in parallel, or it can be a multimodal sensor signal processing integrated circuit for a specific application.

[0054] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for periodic inspection control of wind turbines using PLC and robot collaboration, characterized in that, include: The edge computing gateway loads the inspection task file and requests macroscopic environmental data from the PLC to obtain the operating control parameters and inspection quality thresholds. The robot controller drives the robot to move toward the current waypoint based on motion control parameters and collects robot posture and position data and main inspection sensor data. Online inspection quality indicators are calculated and evaluated based on the main inspection sensor data to obtain real-time quality scores and quality status indicators; Based on the quality status flag, adaptive control decisions are made on the current motion control parameters and the current robot posture and position data to obtain adaptive control commands.

2. The PLC-robot collaborative wind turbine periodic inspection control method according to claim 1, characterized in that, Online inspection quality indicators are calculated and evaluated based on the main inspection sensor data to obtain real-time quality scores and quality status indicators, including: Since the main inspection sensor data is ultrasonic waveform data, the signal-to-noise ratio of the ultrasonic waveform data is calculated as the real-time quality score. Since the main inspection sensor data is image data, the gradient energy aggregation value of the image data is calculated as the real-time quality score; The quality status indicator is determined by comparing the real-time quality score with the inspection quality threshold.

3. The PLC-robot collaborative wind turbine periodic inspection control method according to claim 2, characterized in that, In response to the main inspection sensor data being ultrasonic waveform data, the signal-to-noise ratio of the ultrasonic waveform data is calculated as the real-time quality score, including: Extract the effective echo signal amplitude and background noise signal amplitude from the ultrasonic waveform data; Based on the effective echo signal amplitude and the background noise signal amplitude, the signal-to-noise ratio of the ultrasonic waveform data is calculated using the following formula: in, The signal-to-noise ratio (SNR) is the value used as the real-time quality score for ultrasound. For the effective echo signal amplitude, This represents the amplitude of the background noise signal.

4. The PLC-robot collaborative wind turbine periodic inspection control method according to claim 1, characterized in that, In response to the main inspection sensor data being image data, the gradient energy aggregation value of the image data is calculated as the real-time quality score, including: Based on robot posture and position data, camera intrinsic parameters, and blade 3D model, the distortion Jacobian map is determined. Based on the distortion Jacobian map, pixel-level geometric fidelity weights and gradient features are computed in parallel to obtain geometric fidelity weight maps and pixel gradient maps. The gradient energy aggregation value of the image data is determined based on the geometric fidelity weight map and the pixel gradient map.

5. The PLC-robot collaborative wind turbine periodic inspection control method according to claim 4, characterized in that, Determining the gradient energy aggregation value of the image data based on the geometric fidelity weight map and pixel gradient map includes: determining the gradient energy aggregation value of the image data using the following formula, wherein the formula is: in, This represents the aggregated value of gradient energy. This indicates an accumulation operation on all coordinate points in the two-dimensional matrix of the image. Represents pixels Geometric fidelity weights Represents pixels The gradient magnitude.

6. The PLC-robot collaborative wind turbine periodic inspection control method according to claim 1, characterized in that, Based on the quality status flag, adaptive control decisions are made using the current motion control parameters and the current robot posture and position data to obtain adaptive control commands, including: Determine action intentions based on quality status indicators; Generate action commands based on the action intent, current motion control parameters, and current robot posture and position data; Adaptive control commands are obtained by adaptively synthesizing action intentions and action instructions.

7. A PLC-robot collaborative wind turbine periodic inspection control system, characterized in that, include: The data acquisition module is used by the edge computing gateway to load the inspection task file and request macro-environment data from the PLC to obtain the operation control parameters and inspection quality thresholds. The robot control and data acquisition module is used by the robot controller to drive the robot to move to the current waypoint according to the motion control parameters and to collect robot posture and position data and main inspection sensor data. The online inspection quality index calculation and evaluation module is used to calculate and evaluate the quality index of the main inspection sensor data online in order to obtain real-time quality score and quality status indicator; The adaptive control decision module is used to make adaptive control decisions based on the quality status flag, the current motion control parameters, and the current robot posture and position data to obtain adaptive control commands.

8. The PLC and robot collaborative wind turbine periodic inspection control system according to claim 7, characterized in that, The online inspection quality indicator calculation and evaluation module includes: The signal-to-noise ratio calculation unit is used to calculate the signal-to-noise ratio of the ultrasonic waveform data as the real-time quality score in response to the main inspection sensor data being ultrasonic waveform data. The gradient energy aggregation value calculation unit is used to calculate the gradient energy aggregation value of the image data as the real-time quality score in response to the main inspection sensor data being image data. The quality status indicator determination unit is used to determine the quality status indicator based on the comparison between the real-time quality score and the inspection quality threshold.

9. The PLC and robot collaborative wind turbine periodic inspection control system according to claim 7, characterized in that, The gradient energy aggregation value calculation unit also includes: The distortion Jacobian map determination unit is used to determine the distortion Jacobian map based on robot posture and position data, camera intrinsic parameters, and blade 3D model. The parallel computation unit for geometric fidelity weights and gradient features is used to perform pixel-level parallel computation of geometric fidelity weights and gradient features on image data based on the distortion Jacobian map to obtain geometric fidelity weight maps and pixel gradient maps. The gradient energy aggregation value determination unit is used to determine the gradient energy aggregation value of the image data based on the geometric fidelity weight map and the pixel gradient map.

10. The PLC and robot collaborative wind turbine periodic inspection control system according to claim 7, characterized in that, The adaptive control decision module includes: Action intent determination unit, used to determine action intent based on quality status indicators; The action command generation unit is used to generate action commands based on the action intention, the current motion control parameters, and the current robot posture and position data; The adaptive instruction synthesis unit is used to synthesize adaptive instructions from action intentions and action instructions to obtain adaptive control instructions.

Images