Cleaning method for internal cavity of GIS equipment and cleaning robot

Abstract

The invention relates to the technical field of cleaning robots, and discloses a cleaning method for an internal cavity of GIS equipment and a cleaning robot. The method comprises the following steps: acquiring image information of an internal cavity of GIS equipment, separating a foreign matter area of the image information from a background area, generating a binary mask image, and calculating three-dimensional coordinates of the foreign matter area; a joint angle sequence corresponding to discrete points on the smooth track is converted into a driving signal of a servo motor through an inverse kinematics solver, and the flexible joint arm is driven to approach the foreign matter area point by point; after the tail end of the mechanical arm reaches a target position, the roll type dust-free cloth and the flexible gripper work in parallel to clean a foreign matter area; and generating new image information, inputting the new image information into the U-Net network, carrying out intersection-to-union ratio calculation on the output new mask and the initial mask, recalculating the complementary scanning track and continuing cleaning until the intersection-to-union ratio is greater than or equal to a preset threshold value. The cleaning efficiency and precision of the internal cavity of the GIS equipment are improved.

Description

Technical Field

[0001] The present application relates to the technical field of cleaning robots, and in particular to a method for cleaning the internal cavity of GIS equipment and a cleaning robot. Background Art

[0002] With the continuous development of automation technology and intelligent equipment, cleaning robots have been widely used in industries such as industry, healthcare, and home. The demand for automated cleaning equipment is increasing, especially in environments that require high-precision and efficient cleaning. This is especially true for cleaning GIS (Gas Insulated Switchgear) equipment, where the internal cavities are often narrow and complex. Traditional manual cleaning is not only inefficient but also carries high risks and the potential for incomplete cleaning.

[0003] Existing cleaning methods mainly rely on manual operation or simple automated equipment, and often face technical problems such as inaccurate foreign body identification, lack of intelligent optimization of the cleaning process, inaccurate path planning and motion control, and difficulty in verifying cleaning effects.

[0004] With the continuous advancement of deep learning, computer vision, path planning, and adaptive control technologies, new technical approaches have begun to be proposed to address these issues. For example, U-Net convolutional neural networks have been widely used in image segmentation, effectively and accurately identifying foreign object regions. Simultaneously, methods such as deep reinforcement learning are being used for dynamic path planning and cleaning strategies. Adaptive control systems can automatically adjust cleaning force and suction power based on the physical properties of foreign objects and environmental changes.

[0005] Despite this, existing technical solutions still have some limitations. In particular, when dealing with complex cleaning tasks, they have not yet achieved high-precision foreign object recognition, efficient path planning, and self-verification and optimization of cleaning results. Therefore, developing a fully automated cleaning method that can automatically identify foreign objects, dynamically adjust cleaning strategies, and ensure efficient and damage-free cleaning remains a major technical challenge in this field. Summary of the Invention

[0006] The present application provides a method for cleaning the internal cavity of a GIS device and a cleaning robot, which are used to improve the cleaning efficiency and accuracy of the internal cavity of the GIS device.

[0007] In a first aspect, the present application provides a method for cleaning the internal cavity of a GIS device, the method comprising: The cleaning robot uses a binocular vision system to collect image information of the internal cavity of the GIS equipment, uses a U-Net network to perform semantic segmentation on the image information, separates the foreign object area from the background area of ​​the image information, generates a binary mask image, and calculates the three-dimensional coordinates of the foreign object area; A collision-free smooth trajectory is generated based on the three-dimensional coordinates, and a joint angle sequence corresponding to discrete points on the smooth trajectory is converted into a drive signal for a servo motor through an inverse kinematics solver, and the flexible joint arm is driven to approach the foreign object area point by point based on the drive signal; After the end of the robotic arm reaches the target position, the rolled dust-free cloth and the flexible gripper work in parallel to clean the foreign matter area; After the cleaning action is completed, the foreign matter area is photographed again to generate new image information, the new image information is input into the U-Net network, and the intersection-and-union ratio of the output new mask and the initial mask is calculated. If the intersection-and-union ratio is less than the preset threshold, the scanning trajectory is recalculated and cleaning is continued until the intersection-and-union ratio is greater than or equal to the preset threshold.

[0008] In combination with the first aspect, separating the foreign matter area from the background area of ​​the image information includes: Collecting a first image set of the internal cavity of a plurality of GIS devices, marking a region of any first image in the first image set and performing image preprocessing to generate a second image; Use a deep learning framework to build a U-Net network, take all the second images as input, take the region label corresponding to the binary classification result of each pixel as output, and use the cross entropy loss function to calculate the difference between the network output and the true label; Inputting a second image set corresponding to the second image into a U-Net network for training, and adjusting network hyperparameters based on the gap value; Input all the image information into the adjusted U-Net network, the U-Net network predicts any pixel point and outputs a probability value, sets a first preset value, and if the probability value is greater than the first preset value, marks the pixel point as a foreign object pixel, otherwise, marks it as a background pixel; The image information is segmented according to different labeling categories, all the foreign body pixels are combined and set as the foreign body area, all the background pixels are combined and set as the background area, and the probability value of any pixel point is binarized and combined to generate the binary mask image.

[0009] In combination with the first aspect, the calculating the three-dimensional coordinates of the foreign matter area includes: Extracting a left-view image and a right-view image corresponding to the same image stream from all the image information based on the binocular vision system; Based on the binary mask image, matching pixel points of the foreign object area are extracted from the left view image and the right view image respectively, and the disparity value between each pair of matching pixel points is calculated by a stereo matching algorithm; A parameter matrix of the binocular vision system is obtained, and the two-dimensional coordinates of the foreign matter area are converted into the three-dimensional coordinates based on the parameter matrix and the disparity value.

[0010] In combination with the first aspect, generating a collision-free smooth trajectory based on the three-dimensional coordinates includes: Acquire the cleaning robot's posture data, which includes motion data and IMU data, and fuse all posture data through extended Kalman filtering to generate real-time position and posture data; Calculating a relative posture error between the robot base and the foreign object area based on the three-dimensional coordinates, the real-time position and the posture data; Combining the relative pose error and using a trajectory planning algorithm to generate a first path for the robotic arm from the current position to the foreign object area; The first path is smoothed by quintic polynomial interpolation to generate the smooth trajectory.

[0011] In combination with the first aspect, when the flexible joint arm approaches the foreign object area, the vacuum pump adjusts the speed according to the expected negative pressure value output by the adsorption force model. The adsorption force model is based on the current load mass, sealing gap height and friction coefficient, and establishes a ternary mapping relationship through the Bernoulli equation and the gap flow equation, and controls the adsorption force of the cleaning robot based on the ternary mapping relationship.

[0012] In combination with the first aspect, calculating the intersection-over-union ratio of the output new mask and the initial mask includes: In the new mask and the initial mask, determine whether the mask value corresponding to any pixel point is the same, calculate the intersection area and union area between the new mask and the initial mask based on the determination result, and set the ratio of the intersection area to the union area as the intersection-to-union ratio.

[0013] In combination with the first aspect, recalculating the re-sweep trajectory and continuing cleaning includes: The residual area of ​​the foreign matter area is located based on the new mask, the cleaning trajectory of the robot arm is regenerated based on the coordinate position and shape of the residual area, and the cleaning trajectory is smoothed using quintic polynomial interpolation to generate the supplementary scanning trajectory.

[0014] In combination with the first aspect, after the cleaning robot continues cleaning, it also includes: Collect historical record data of all cleaning tasks, including task data, robot performance data, and cleaning effect data, and perform data preprocessing on the historical record data based on data type; Setting optimization goals, including path planning, cleaning strategy, and equipment status assessment, and selecting a machine learning algorithm based on the optimization goals to establish an optimization model; Extracting associated data sets corresponding to the same optimization goal from all processed historical record data, using the associated data sets to train the optimization model, and outputting the optimization results; All the optimization results are summarized based on the optimization target to generate a cleaning strategy, and the cleaning robot performs a new cleaning task based on the cleaning strategy.

[0015] In combination with the first aspect, the cleaning of the foreign matter area includes: Based on the real-time detection of the physical characteristics of foreign objects in the foreign object area by sensors, the support vector machine algorithm is used to classify and identify foreign objects and obtain the type of foreign objects; Adjusting the adsorption force and cleaning force based on the type of foreign matter and the physical characteristics, wherein the adsorption force model adjusts the negative pressure value according to the current load mass, the sealing gap height, and the friction coefficient; The negative pressure is calculated from the flow velocity and pressure difference in the suction cup area using the Bernoulli equation, the flow rate of the fluid passing through the small gap between the suction cup and the device surface is calculated according to the gap flow equation, and a mathematical expression is constructed based on the negative pressure and the flow rate, and is set as the ternary mapping relationship; During the operation of the flexible articulated arm, the adsorption force is adaptively adjusted based on the ternary mapping relationship, and the foreign matter area is cleaned based on the adjusted adsorption force.

[0016] In a second aspect, the present application provides a cleaning robot, comprising: An acquisition module is used for the cleaning robot to collect image information of the internal cavity of the GIS equipment using a binocular vision system, perform semantic segmentation on the image information using a U-Net network, separate the foreign object area from the background area of ​​the image information, generate a binary mask image, and calculate the three-dimensional coordinates of the foreign object area; a drive module, configured to generate a collision-free smooth trajectory based on the three-dimensional coordinates, convert a sequence of joint angles corresponding to discrete points on the smooth trajectory into drive signals for a servo motor through an inverse kinematics solver, and drive the flexible joint arm to approach the foreign object area point by point based on the drive signals; The cleaning module is used to clean the foreign matter area after the end of the robotic arm reaches the target position. The rolled dust-free cloth and the flexible gripper work in parallel. The verification module is used to photograph the foreign matter area again after the cleaning action is completed to generate new image information, input the new image information into the U-Net network, and calculate the intersection-and-union ratio of the output new mask and the initial mask. If the intersection-and-union ratio is less than the preset threshold, the supplementary scanning trajectory is recalculated and cleaning is continued until the intersection-and-union ratio is greater than or equal to the preset threshold.

[0017] In the technical solution provided by this application, first, by combining the binocular vision system with the U-Net network, accurate identification of foreign body areas is achieved. Using a deep learning framework to perform high-precision segmentation on the image, the foreign body area can be effectively separated from the background area, thereby accurately locating the foreign body position and generating an accurate cleaning path. Then, based on the foreign body classification and recognition technology of the support vector machine algorithm, the physical properties of the foreign body are detected in real time, and the robot's adsorption force and cleaning strength are intelligently adjusted to ensure that the adsorption force and cleaning strength can be optimized and adjusted during the cleaning process of different types of foreign bodies, avoiding the equipment damage or incomplete cleaning caused by excessive or insufficient adsorption force in traditional cleaning methods. Then, by combining the U-Net network with intersection-over-union calculation, real-time feedback and verification of the cleaning effect are achieved. After the cleaning task is completed, the robot re-photographs the clean area and calculates the intersection-over-union ratio to evaluate the cleaning effect in real time. The closed-loop feedback mechanism ensures the efficiency and accuracy of the cleaning process, avoiding the defects of the traditional cleaning method that the cleaning effect is difficult to control and verify. Finally, it is able to perform adaptive optimization based on data from historical cleaning tasks, thereby continuously improving the efficiency and accuracy of subsequent tasks. The cleaning robot can make optimization decisions based on historical data, automatically select the best path planning, cleaning strategy and equipment status assessment method, and improve the robot's long-term operating performance.

[0018] This application also applies extended Kalman filtering to position data fusion, providing real-time accurate position information of the robot in complex environments. Combined with real-time feedback from cleaning tasks, the robot can dynamically adjust its path planning and cleaning strategy to adapt to different cleaning tasks and environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 This is a schematic diagram of an embodiment of a method for cleaning the internal cavity of a GIS device in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of the overall architecture of the cleaning robot in the embodiment of the present application; Figure 3 This is a schematic diagram of an embodiment of the foreign body identification process in the embodiment of the present application; Figure 4 This is a schematic diagram of an embodiment of the distribution of cleaning robot modules in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The embodiments of the present application provide a method for cleaning the internal cavity of a GIS device and a cleaning robot. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for cleaning the internal cavity of a GIS device includes: Step S101: The cleaning robot uses a binocular vision system to collect image information of the internal cavity of the GIS equipment, uses a U-Net network to perform semantic segmentation on the image information, separates the foreign body area from the background area of ​​the image information, generates a binary mask image, and calculates the three-dimensional coordinates of the foreign body area.

[0023] It is understandable that the execution subject of the present application can be a cleaning device for the internal cavity of the GIS equipment, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0024] Specifically, Figure 2This is the overall architecture diagram of the cleaning robot. The binocular vision system consists of two cameras, which capture two slightly different image information by simultaneously shooting the same scene from different perspectives. U-Net is a deep learning network specifically used for image segmentation. Through an encoding-decoding structure, the network efficiently processes the input image information at the pixel level, separating the target area (such as the foreign object area) from the background area. The foreign object area refers to the part where the foreign objects to be cleaned are present, and the background area refers to the part inside the cavity of the GIS equipment. By marking the foreign object area as "1" and the background area as "0" at each pixel point, a binary mask image can be generated. Three-dimensional coordinates refer to the spatial coordinates of the foreign object area.

[0025] Step S102: Generate a collision-free smooth trajectory based on the three-dimensional coordinates, and convert the joint angle sequence corresponding to the discrete points on the smooth trajectory into a drive signal for the servo motor through an inverse kinematics solver. Drive the flexible joint arm to approach the foreign object area point by point based on the drive signal.

[0026] Specifically, the smooth trajectory refers to the running route of the cleaning robot to the foreign body area. Based on the generated smooth trajectory, the inverse kinematics method is used to calculate the angle of each joint of the robotic arm, that is, the joint angle sequence. The inverse kinematics solver converts each discrete point in the trajectory into the joint angle of the robotic arm, and controls the movement of each joint of the robotic arm through the servo motor. The joint angle obtained by the inverse kinematics solution will be converted into a PWM duty cycle signal, which is the drive signal, and then the precise movement of the robotic arm is achieved through the motor drive control system. The calculation and control instructions of each joint angle will be sent to the robotic arm control system in real time, so that the cleaning robot approaches the foreign body area point by point.

[0027] Step S103: After the end of the robotic arm reaches the target position, the rolled dust-free cloth and the flexible gripper work in parallel to clean the foreign matter area.

[0028] Specifically, during cleaning tasks, the end of the robotic arm is equipped with two cleaning tools: a dust-free roll cloth and a flexible gripper. The dust-free cloth is used to gently wipe the surface, while the flexible gripper is suitable for sucking up larger and heavier foreign objects.

[0029] Step S104: After the cleaning action is completed, the foreign body area is photographed again to generate new image information, and the new image information is input into the U-Net network. The output new mask is compared with the initial mask to calculate the intersection and union ratio. If the intersection and union ratio is less than the preset threshold, the scanning trajectory is recalculated and cleaning is continued until the intersection and union ratio is greater than or equal to the preset threshold.

[0030] Specifically, Intersection over Union (IoU) is a commonly used evaluation metric in object detection and segmentation tasks, measuring the degree of overlap between two regions. If the IoU is less than a preset threshold, the cleaning effect is insufficient and a rescan trajectory needs to be recalculated and cleaning continued. The cleaning task is complete when the IoU is greater than or equal to the preset threshold.

[0031] In a specific embodiment, separating the foreign matter area from the background area of ​​the image information includes: (1) A first image set of the internal cavity of various GIS devices is collected. After marking the area of ​​any first image in the first image set, image preprocessing is performed to generate a second image.

[0032] (2) Use the deep learning framework to build a U-Net network, take all second images as input, take the region label corresponding to the binary classification result of each pixel as output, and use the cross entropy loss function to calculate the difference between the network output and the true label.

[0033] (3) Input the second image set corresponding to the second image into the U-Net network for training, and adjust the network hyperparameters based on the gap value.

[0034] (4) All image information is input into the adjusted U-Net network. The U-Net network predicts any pixel point and outputs a probability value. A first preset value is set. If the probability value is greater than the first preset value, the pixel point is marked as a foreign object pixel. Otherwise, it is marked as a background pixel.

[0035] (5) Segment the image information according to different label categories, combine all foreign object pixels and set them as foreign object areas, combine all background pixels and set them as background areas, binarize the probability value of any pixel point and combine them to generate a binary mask image.

[0036] Specifically, Figure 3 This is a flow chart for foreign object identification. The first image set refers to a collection of first images collected from the internal cavities of multiple different GIS devices by the visual system (such as a binocular vision system) equipped by the cleaning robot. The first image contains various objects and foreign matter inside the device, such as dust, metal particles, and debris. Region marking is the process of manually annotating the foreign object areas in the image. In each image, the foreign object areas (such as dust and debris) and background areas (such as the surface of the device) are manually marked. After marking is completed, the necessary image preprocessing operations (such as resizing, denoising, standardization, etc.) are performed, which is the second image, which will be used as label data for network training.

[0037] U-Net, through its encoder-decoder structure, can accurately separate the target (foreign object area) from the background area in the image. The cross-entropy loss function is often used in binary classification problems to measure the difference between the network output and the true label. For example, the calculation formula for the difference value L output by the cross-entropy loss function is: ,in, is the i-th true label, is the i-th probability predicted by the U-Net network, and N is the number of labels. In this application, the output is the classification result (foreign object or background) for each pixel in the image, and the true label is the manually annotated foreign object area and background area.

[0038] During the training process, the model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) are adjusted according to the error calculated by the loss function so that the model can better fit the training data and ultimately achieve higher accuracy and generalization ability.

[0039] The trained U-Net network can predict the information of each input image and output whether each pixel is a "foreign object pixel" or a "background pixel" through a probability value.

[0040] The final binary mask image is generated through the segmentation results of the U-Net network, in which the pixel value of the foreign body area is 1 and the pixel value of the background area is 0.

[0041] In a specific embodiment, calculating the three-dimensional coordinates of the foreign matter area includes: (1) Based on the binocular vision system, the left-view image and the right-view image corresponding to the same image stream are extracted from all image information.

[0042] (2) Based on the binary mask image, matching pixels of the foreign object area are extracted in the left view image and the right view image respectively, and the disparity value between each pair of matching pixels is calculated by the stereo matching algorithm.

[0043] (3) Obtain the parameter matrix of the binocular vision system, and convert the two-dimensional coordinates of the foreign object area into three-dimensional coordinates based on the parameter matrix and disparity value.

[0044] Specifically, the binocular vision system consists of two cameras, which shoot at the same scene from a fixed perspective, namely the left perspective image and the right perspective image respectively. By comparing the difference between the two images (parallax value), the depth information of objects in the scene can be calculated.

[0045] The disparity value is the offset between corresponding pixels in two images along the X-axis, indicating the distance of the object from the camera. A larger disparity value indicates that the object is closer to the camera, and vice versa. Stereo matching algorithms include but are not limited to block matching or other matching methods.

[0046] The parameter matrix includes the intrinsic parameter matrix of the cameras in the binocular vision system (focal length, optical center coordinates, etc.) and the extrinsic parameter matrix between the cameras (the position and rotation angle of the two cameras). Based on the pinhole imaging principle and triangulation, the two-dimensional image coordinates are converted into three-dimensional space coordinates. For example, the formula for the pinhole imaging principle is: , where f is the focal length of the camera, B is the baseline distance between the two cameras in the binocular vision system, d is the disparity value of the pixel point, and Z is the depth information of the object in the scene in three-dimensional space (i.e., the distance between the camera and the object). Use the triangulation principle to restore the three-dimensional coordinates of the object, and convert the two-dimensional coordinates of the pixel points in the image into Mapping to 3D coordinates in 3D space ,in, , The precise position of each foreign object pixel in three-dimensional space can be determined.

[0047] In a specific embodiment, generating a collision-free smooth trajectory based on three-dimensional coordinates includes: (1) Obtain the posture data of the cleaning robot, which includes motion data and IMU data. All posture data are fused through extended Kalman filtering to generate real-time position and posture data.

[0048] (2) Calculate the relative posture error between the robot base and the foreign object area based on the three-dimensional coordinates, real-time position and posture data.

[0049] (3) Combine the relative pose error and use the trajectory planning algorithm to generate the first path of the robot arm from the current position to the foreign object area.

[0050] (4) The first path is smoothed by quintic polynomial interpolation to generate a smooth trajectory.

[0051] Specifically, motion data refers to the displacement, velocity, and acceleration information obtained by the robot through motion sensors (such as encoders). IMU data is used to measure the acceleration and angular velocity of the robot in space and estimate its rotation angle and linear velocity. The extended Kalman filter (EKF) is a classic filtering algorithm specifically designed to handle state estimation problems in nonlinear systems. It can provide more accurate pose estimation based on data from multiple sensors. The fused data can provide the robot's real-time pose (including real-time position and attitude data), that is, the robot's position, angle, and motion state in space. For example, the fusion formula is: ,in, is the pose estimate at the current moment, is the state transfer function (calculating the pose through the odometer and IMU data), is the noise figure.

[0052] Relative pose error refers to the positional error between the cleaning robot's current position and the target foreign object area. It represents the distance and angle differences between the robot and the foreign object area. This includes both position error and pose error. Position error is the distance difference between the robot's current position and the target position, while pose error is the angular difference between the robot's current orientation and the target direction.

[0053] Trajectory planning algorithms (such as the RRT algorithm) are the process of calculating the path between the robot's starting position and the target position. The RRT algorithm quickly explores a tree in the configuration space (i.e., the space where the robot may move) through random sampling, thereby generating a path from the starting point to the target point, which is the first path. The first path is collision-free and is calculated based on the real-time environment and obstacle avoidance.

[0054] Quintic polynomial interpolation is a commonly used method for smoothing paths. It can make paths smoother and avoid sharp turns or discontinuities in the path.

[0055] In a specific embodiment, when the flexible joint arm approaches the foreign object area, the vacuum pump adjusts the speed according to the expected negative pressure value output by the adsorption force model. The adsorption force model establishes a ternary mapping relationship based on the current load mass, sealing gap height and friction coefficient through the Bernoulli equation and the gap flow equation, and controls the adsorption force of the cleaning robot based on the ternary mapping relationship.

[0056] Specifically, the suction force model is a mathematical model that calculates the required negative pressure based on the actual conditions of the robot's suction system (such as load mass, sealing gap height, and friction coefficient). Negative pressure control is a core component of the cleaning robot's suction process, as excessive suction force can damage the surface, while insufficient suction force may prevent stable adhesion. Negative pressure is a key variable in the cleaning robot's suction system, directly affecting the suction force between the robot and the surface. The vacuum pump is a core component of the cleaning robot's suction system, responsible for generating the required negative pressure. By adjusting the vacuum pump's speed, the negative pressure can be varied in real time, thereby adjusting the suction force. The ternary mapping relationship refers to the relationship between "structural parameters - leakage flow - negative pressure." The Bernoulli equation describes the principle of energy conservation in fluids. In the cleaning robot's suction system, the Bernoulli equation can be used to describe the pressure changes in the air flowing between the suction cup and the surface, thereby deriving the relationship between negative pressure and flow. The gap flow equation is primarily used to describe the flow behavior of fluids in small gaps (such as the space between the suction cup and the surface). The fluid flow rate and negative pressure are closely related and can be described using fluid mechanics equations. By combining the Bernoulli equation and the gap flow equation, a ternary mapping relationship between structural parameters, leakage flow, and negative pressure was established. Specifically, the suction cup's structural parameters (such as size, sealing gap, and friction coefficient) affect the air flow velocity and pressure, and thus the magnitude of the negative pressure. Through experiments and numerical analysis, the relationship between the suction cup's structural parameters, leakage flow, and negative pressure was determined. Ultimately, a mapping relationship was established. This mapping relationship enables the system to adjust the suction force according to the target negative pressure, thereby precisely controlling the contact between the robot and the device surface.

[0057] Through a ternary mapping relationship, the system can optimize the adsorption force in real time based on the relationship between structural parameters, leakage flow, and negative pressure. If different types of foreign matter are encountered during the cleaning task, the system will automatically adjust the adsorption force (negative pressure) according to their different physical properties.

[0058] In a specific embodiment, calculating the intersection-over-union ratio between the output new mask and the initial mask includes: In the new mask and the initial mask, determine whether the mask value corresponding to any pixel point is the same. Based on the determination result, calculate the intersection area and union area between the new mask and the initial mask, and set the ratio of the intersection area to the union area as the intersection-to-union ratio.

[0059] Specifically, the new mask refers to the binary image generated by the image retaken by the camera at the end of the cleaning robot after the cleaning task is completed and processed by the U-Net network, indicating whether there is a foreign body part in the cleaning area. The "1" in the new mask indicates the foreign body area and the "0" indicates the background area. The initial mask is a binary image generated by processing the pre-cleaning image through the U-Net network before the cleaning task begins, which identifies the foreign body area in the pre-cleaning image. Each pixel in the mask image represents the classification result of the corresponding position in the image, that is, whether the position belongs to the foreign body area (1) or the background area (0). In order to calculate the intersection-over-union ratio, it is necessary to compare the values ​​of each pixel in the new mask and the initial mask to see if they are consistent. The intersection area refers to the area corresponding to the pixel points marked as "1" in both the new mask and the initial mask, that is, the area of ​​the foreign body area that overlaps after cleaning and before cleaning. The union area refers to the area in which at least one mask is marked as "1" in the new mask and the initial mask, that is, the total area of ​​the foreign body area before and after cleaning. The intersection-over-union (IoU) ratio is a common metric for measuring the overlap between two regions. It is usually used in image segmentation tasks to evaluate the degree of match between the model output and the true label. The higher the value, the better the removal effect of the foreign matter area after cleaning and the higher the cleaning quality.

[0060] In one embodiment, recalculating the touch-up trajectory and continuing cleaning includes: The residual area of ​​the foreign body area is located based on the new mask, and the cleaning trajectory of the robot arm is regenerated based on the coordinate position and shape of the residual area. The cleaning trajectory is smoothed using quintic polynomial interpolation to generate a supplementary sweeping trajectory.

[0061] Specifically, residual areas are areas where foreign matter remains after the initial cleaning. By comparing the mask images before and after cleaning (initial mask and new mask), we can identify areas that remain incompletely cleaned after cleaning, known as residual areas. The cleaning trajectory is the path the robot arm must follow during cleaning. When calculating the re-sweep trajectory, the path is replanned based on the location and shape of the residual area. Similarly, a quintic polynomial interpolation algorithm is used to smooth the path and generate the re-sweep trajectory.

[0062] In a specific embodiment, after the cleaning robot continues cleaning, the method further includes: (1) Collect historical data of all cleaning tasks, including task data, robot performance data, and cleaning effect data, and perform data preprocessing on the historical data based on the data type.

[0063] (2) Set optimization goals, which include path planning, cleaning strategy, and equipment status evaluation. Select a machine learning algorithm based on the optimization goals to establish an optimization model.

[0064] (3) Extract the associated data sets corresponding to the same optimization objectives from all the processed historical records, use the associated data sets to train the optimization model, and output the optimization results.

[0065] (4) All optimization results are summarized based on the optimization objective to generate a cleaning strategy, and the cleaning robot performs a new cleaning task based on the cleaning strategy.

[0066] Specifically, task data includes task duration, cleaning area dimensions, and cleaning targets. Robot performance data includes the robot's trajectory, cleaning time, speed, energy consumption, and robotic arm motion data. Cleaning performance data includes final evaluation data for cleaning tasks, such as cleaning accuracy (calculated using intersection-over-union), task completion, and residual foreign matter. To efficiently train machine learning models, the collected historical data undergoes data preprocessing, including cleaning, standardization, and normalization.

[0067] Path planning involves optimizing the robot's path during the cleaning process to ensure efficient and collision-free completion. Cleaning strategies refer to the specific strategies used to optimize the robot's cleaning tasks, including the selection of cleaning tools and the adjustment of cleaning force and speed. Equipment condition assessment uses machine learning algorithms to evaluate the equipment's condition and provide optimization recommendations to improve its long-term performance and stability. Different machine learning algorithms are used to construct optimization models based on different optimization objectives. For example, reinforcement learning algorithms can be used in path planning and cleaning strategy optimization to gradually optimize the strategy through a reward mechanism. Clustering algorithms are used to analyze historical data and automatically identify patterns across different equipment, cleaning tasks, and equipment conditions. Supervised learning algorithms (such as regression analysis and decision trees) are used to predict the relationship between cleaning performance and task data.

[0068] If the optimization objective is path planning, task data related to path planning, such as path points, cleaning areas, and cleaning times, needs to be extracted. Using a selected machine learning algorithm and the extracted associated dataset, the optimization model is trained to learn the relationships between various factors in the historical data, thereby optimizing the cleaning task. The output of the trained optimization model is the optimized cleaning task parameters based on the historical data and the training objective, such as the optimal path, optimal cleaning strategy, and appropriate equipment status assessment.

[0069] The cleaning strategy includes parameters such as the path, cleaning method, and cleaning intensity that the cleaning robot needs to follow when performing tasks, enabling the cleaning robot to achieve efficient and accurate operations in subsequent cleaning tasks.

[0070] In a specific embodiment, cleaning the foreign matter area includes: (1) Based on the real-time detection of the physical characteristics of foreign objects in the foreign object area by sensors, the support vector machine algorithm is used to classify and identify foreign objects and obtain the type of foreign objects.

[0071] (2) The adsorption force and cleaning force are adjusted based on the type and physical properties of the foreign matter. The adsorption force model adjusts the negative pressure value according to the current load mass, sealing gap height, and friction coefficient.

[0072] (3) The negative pressure is calculated by the Bernoulli equation based on the flow velocity and pressure difference in the suction cup area. The flow rate of the fluid passing through the small gap between the suction cup and the surface of the device is calculated according to the gap flow equation. A mathematical expression is constructed based on the negative pressure and flow rate, and is set as a ternary mapping relationship.

[0073] (4) During the operation of the flexible joint arm, the adsorption force is adaptively adjusted based on the ternary mapping relationship, and the foreign matter area is cleaned based on the adjusted adsorption force.

[0074] Specifically, foreign matter refers to the material in the foreign matter area that needs to be cleaned. Its physical properties include hardness, viscosity, coefficient of friction, surface roughness, etc. Appropriate sensors (such as force sensors and pressure sensors) are used to collect data corresponding to the foreign matter's physical properties in real time. Using pre-calibrated training data, the support vector machine (SVM) algorithm can classify foreign matter into different types (such as metal, glass, plastic, dust, etc.), which are then referred to as foreign matter types.

[0075] The suction force model calculates the required negative pressure based on the physical properties of the foreign object (such as mass, seal gap height, friction coefficient, etc.), thereby controlling the suction cup's suction force. The suction force needs to be adjusted appropriately for different foreign objects.

[0076] Adjust the suction force (i.e., negative pressure) based on the detected foreign object's mass and friction coefficient. For example, heavier foreign objects require a stronger suction force to ensure stable suction, while lighter foreign objects do not require as much suction force. Adjust the suction force (i.e., negative pressure) based on the detected foreign object's mass and friction coefficient. For example, heavier foreign objects require a stronger suction force to ensure stable suction, while lighter foreign objects do not require as much suction force. Control the cleaning force by adjusting the suction cup's contact pressure and the movement speed of the robot's flexible articulated arm. Reduce the pressure for soft foreign objects; increase the cleaning force for hard foreign objects.

[0077] Similarly, the required negative pressure for the suction cup is calculated based on the Bernoulli equation and the fluid flow between the robot's suction cup and the surface. A higher negative pressure strengthens the suction force, while a lower negative pressure weakens it. The fluid flow rate (i.e., leakage flow) is calculated based on the gap between the suction cup and the device surface and the fluid's viscosity. Based on the relationship between flow rate and negative pressure, the negative pressure is adjusted to control the suction force.

[0078] By combining the Bernoulli equation with the gap flow equation, a mathematical model is established to establish the relationship between structural parameters (such as suction cup size and gap height), leakage flow, and negative pressure. This mapping relationship enables the control system to adjust the suction force based on the current load mass, friction coefficient, and sealing gap height.

[0079] Based on the ternary mapping relationship, the robot can adaptively adjust the suction force (negative pressure) in real time. If the physical properties of the foreign object change (such as mass, friction coefficient, surface smoothness, etc.), the system will automatically adjust the negative pressure to ensure that the suction force remains within the optimal range, avoiding over-adsorption or under-adsorption.

[0080] The above describes the cleaning method of the internal cavity of the GIS equipment in the embodiment of the present application. The following describes a cleaning robot in the embodiment of the present application. Figure 4 In one embodiment of the present application, a cleaning robot includes: The acquisition module 201 is used for the cleaning robot to use the binocular vision system to collect image information of the internal cavity of the GIS equipment, use the U-Net network to perform semantic segmentation on the image information, separate the foreign body area from the background area of ​​the image information, generate a binary mask image, and calculate the three-dimensional coordinates of the foreign body area.

[0081] The driving module 202 is used to generate a collision-free smooth trajectory based on the three-dimensional coordinates, and convert the joint angle sequence corresponding to the discrete points on the smooth trajectory into the driving signal of the servo motor through the inverse kinematics solver, and drive the flexible joint arm to approach the foreign object area point by point based on the driving signal.

[0082] The cleaning module 203 is used to clean the foreign matter area by using the rolled dust-free cloth and the flexible gripper to work in parallel after the end of the robotic arm reaches the target position.

[0083] Verification module 204 is used to photograph the foreign matter area again after the cleaning action is completed to generate new image information, input the new image information into the U-Net network, and calculate the intersection-and-union ratio of the output new mask and the initial mask. If the intersection-and-union ratio is less than the preset threshold, the scanning trajectory is recalculated and cleaning is continued until the intersection-and-union ratio is greater than or equal to the preset threshold.

[0084] Through the collaborative efforts of these components, the system first achieves precise identification of foreign object areas by combining a binocular vision system with a U-Net network. Using a deep learning framework for high-precision image segmentation, it effectively separates foreign object areas from background areas, pinpointing the location of foreign objects and generating an accurate cleaning path. Then, foreign object classification and recognition technology, based on a support vector machine algorithm, intelligently adjusts the robot's suction and cleaning forces by detecting the physical properties of foreign objects in real time. This ensures optimal adjustment of suction and cleaning force for different types of foreign objects, avoiding the problems of excessive or insufficient suction force that can damage equipment or cause incomplete cleaning in traditional cleaning methods. Next, the U-Net network, combined with intersection-over-union (IoU) calculations, enables real-time feedback and verification of cleaning results. After completing a cleaning task, the robot re-images the cleaned area and calculates the IoU ratio for real-time evaluation of cleaning results. This closed-loop feedback mechanism ensures efficient and accurate cleaning, avoiding the drawbacks of traditional cleaning methods that make it difficult to control and verify cleaning results. Finally, it is able to perform adaptive optimization based on data from historical cleaning tasks, thereby continuously improving the efficiency and accuracy of subsequent tasks. The cleaning robot can make optimization decisions based on historical data, automatically select the best path planning, cleaning strategy and equipment status assessment method, and improve the robot's long-term operating performance.

[0085] This application also applies extended Kalman filtering to position data fusion, providing real-time accurate position information of the robot in complex environments. Combined with real-time feedback from cleaning tasks, the robot can dynamically adjust its path planning and cleaning strategy to adapt to different cleaning tasks and environments.

[0086] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0087] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0088] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for cleaning the internal cavity of a GIS device, characterized in that: The method for cleaning the internal cavity of the GIS equipment includes: The cleaning robot uses a binocular vision system to collect image information of the internal cavity of the GIS equipment, uses a U-Net network to perform semantic segmentation on the image information, separates the foreign object area from the background area of ​​the image information, generates a binary mask image, and calculates the three-dimensional coordinates of the foreign object area; A collision-free smooth trajectory is generated based on the three-dimensional coordinates, and a joint angle sequence corresponding to discrete points on the smooth trajectory is converted into a drive signal for a servo motor through an inverse kinematics solver, and the flexible joint arm is driven to approach the foreign object area point by point based on the drive signal; After the end of the robotic arm reaches the target position, the rolled dust-free cloth and the flexible gripper work in parallel to clean the foreign matter area; After the cleaning action is completed, the foreign matter area is photographed again to generate new image information, the new image information is input into the U-Net network, and the intersection-and-union ratio of the output new mask and the initial mask is calculated. If the intersection-and-union ratio is less than the preset threshold, the scanning trajectory is recalculated and cleaning is continued until the intersection-and-union ratio is greater than or equal to the preset threshold.

2. The method for cleaning the internal cavity of GIS equipment according to claim 1, characterized in that: The separating of the foreign matter area from the background area of ​​the image information includes: Collecting a first image set of the internal cavity of a plurality of GIS devices, marking a region of any first image in the first image set and performing image preprocessing to generate a second image; Use a deep learning framework to build a U-Net network, take all the second images as input, take the region label corresponding to the binary classification result of each pixel as output, and use the cross entropy loss function to calculate the difference between the network output and the true label; Inputting a second image set corresponding to the second image into a U-Net network for training, and adjusting network hyperparameters based on the gap value; Input all the image information into the adjusted U-Net network, the U-Net network predicts any pixel point and outputs a probability value, sets a first preset value, and if the probability value is greater than the first preset value, marks the pixel point as a foreign object pixel, otherwise, marks it as a background pixel; The image information is segmented according to different labeling categories, all the foreign body pixels are combined and set as the foreign body area, all the background pixels are combined and set as the background area, and the probability value of any pixel point is binarized and combined to generate the binary mask image.

3. The method for cleaning the internal cavity of GIS equipment according to claim 2, characterized in that: The calculating of the three-dimensional coordinates of the foreign matter area includes: Extracting a left-view image and a right-view image corresponding to the same image stream from all the image information based on the binocular vision system; Based on the binary mask image, matching pixel points of the foreign object area are extracted from the left view image and the right view image respectively, and the disparity value between each pair of matching pixel points is calculated by a stereo matching algorithm; A parameter matrix of the binocular vision system is obtained, and the two-dimensional coordinates of the foreign matter area are converted into the three-dimensional coordinates based on the parameter matrix and the disparity value.

4. The method for cleaning the internal cavity of GIS equipment according to claim 1, characterized in that: Generating a collision-free smooth trajectory based on the three-dimensional coordinates includes: Acquire the cleaning robot's posture data, which includes motion data and IMU data, and fuse all posture data through extended Kalman filtering to generate real-time position and posture data; Calculating a relative posture error between the robot base and the foreign object area based on the three-dimensional coordinates, the real-time position and the posture data; Combining the relative pose error and using a trajectory planning algorithm to generate a first path for the robotic arm from the current position to the foreign object area; The first path is smoothed by quintic polynomial interpolation to generate the smooth trajectory.

5. The method for cleaning the internal cavity of GIS equipment according to claim 1, characterized in that: When the flexible joint arm approaches the foreign object area, the vacuum pump adjusts the speed according to the expected negative pressure value output by the adsorption force model. The adsorption force model establishes a ternary mapping relationship between the Bernoulli equation and the gap flow equation based on the current load mass, sealing gap height and friction coefficient, and controls the adsorption force of the cleaning robot based on the ternary mapping relationship.

6. The method for cleaning the internal cavity of GIS equipment according to claim 1, characterized in that: The step of calculating the intersection-over-union ratio between the output new mask and the initial mask includes: In the new mask and the initial mask, determine whether the mask value corresponding to any pixel point is the same, calculate the intersection area and union area between the new mask and the initial mask based on the determination result, and set the ratio of the intersection area to the union area as the intersection-to-union ratio.

7. The method for cleaning the internal cavity of GIS equipment according to claim 6, characterized in that: The recalculation of the sweep trajectory and the continuation of cleaning include: The residual area of ​​the foreign matter area is located based on the new mask, the cleaning trajectory of the robot arm is regenerated based on the coordinate position and shape of the residual area, and the cleaning trajectory is smoothed using quintic polynomial interpolation to generate the supplementary scanning trajectory.

8. The method for cleaning the internal cavity of GIS equipment according to claim 1, characterized in that: After the cleaning robot continues cleaning, it also includes: Collect historical record data of all cleaning tasks, including task data, robot performance data, and cleaning effect data, and perform data preprocessing on the historical record data based on data type; Setting optimization goals, including path planning, cleaning strategy, and equipment status assessment, and selecting a machine learning algorithm based on the optimization goals to establish an optimization model; Extracting associated data sets corresponding to the same optimization goal from all processed historical record data, using the associated data sets to train the optimization model, and outputting the optimization results; All the optimization results are summarized based on the optimization target to generate a cleaning strategy, and the cleaning robot performs a new cleaning task based on the cleaning strategy.

9. The method for cleaning the internal cavity of GIS equipment according to claim 5, characterized in that: The cleaning of the foreign matter area includes: Based on the real-time detection of the physical characteristics of foreign objects in the foreign object area by sensors, the support vector machine algorithm is used to classify and identify foreign objects and obtain the type of foreign objects; Adjusting the adsorption force and cleaning force based on the type of foreign matter and the physical characteristics, wherein the adsorption force model adjusts the negative pressure value according to the current load mass, the sealing gap height, and the friction coefficient; The negative pressure is calculated from the flow velocity and pressure difference in the suction cup area using the Bernoulli equation, the flow rate of the fluid passing through the small gap between the suction cup and the device surface is calculated according to the gap flow equation, and a mathematical expression is constructed based on the negative pressure and the flow rate, and is set as the ternary mapping relationship; During the operation of the flexible articulated arm, the adsorption force is adaptively adjusted based on the ternary mapping relationship, and the foreign matter area is cleaned based on the adjusted adsorption force.

10. A cleaning robot, characterized in that: include: An acquisition module is used for the cleaning robot to collect image information of the internal cavity of the GIS equipment using a binocular vision system, perform semantic segmentation on the image information using a U-Net network, separate the foreign object area from the background area of ​​the image information, generate a binary mask image, and calculate the three-dimensional coordinates of the foreign object area; a drive module, configured to generate a collision-free smooth trajectory based on the three-dimensional coordinates, convert a sequence of joint angles corresponding to discrete points on the smooth trajectory into drive signals for a servo motor through an inverse kinematics solver, and drive the flexible joint arm to approach the foreign object area point by point based on the drive signals; The cleaning module is used to clean the foreign matter area after the end of the robotic arm reaches the target position. The rolled dust-free cloth and the flexible gripper work in parallel. The verification module is used to photograph the foreign matter area again after the cleaning action is completed to generate new image information, input the new image information into the U-Net network, and calculate the intersection-and-union ratio of the output new mask and the initial mask. If the intersection-and-union ratio is less than the preset threshold, the supplementary scanning trajectory is recalculated and cleaning is continued until the intersection-and-union ratio is greater than or equal to the preset threshold.

Images