Anti-corrosion control system and method for pipeline operation robot

Abstract

The present invention discloses an anti-corrosion control system and method for a pipeline operation robot, relating to the field of robot control technology. The system is composed of several functional modules, including: a multi-sensor fusion module, which acquires pipeline images in real time through a camera; and simultaneously utilizes sensors to dynamically collect internal environmental parameters of the pipeline; a three-dimensional modeling and positioning module, which constructs a three-dimensional model of the pipeline based on a SLAM algorithm and pipeline images, and locates the pipeline position of the robot in real time; an intelligent decision-making module, which utilizes YOLOv5 based on the pipeline image to identify the pipeline corrosion area and position, and performs corrosion analysis on the corrosion area, the corrosion analysis including corrosion thickness and morphological contour; utilizes electromagnetic eddy current detection to measure the corrosion thickness of the corrosion area, and utilizes deep learning model image recognition to extract the morphological contour of the corrosion area; and triggers an adaptive control strategy based on the internal environmental parameters of the pipeline and the pipeline position of the robot.

Description

Technical Field

[0001] The present invention relates to the technical field of robot control, and in particular to an anti-corrosion control system and method for a pipeline operation robot. Background Art

[0002] The reservoir water diversion tunnel consists of an inlet gate, a tunnel section and an outlet gate chamber. After a long period of water flow, the steel lining of the water diversion tunnel is severely corroded and needs anti-corrosion treatment; the tunnel inlet is at the bottom of the reservoir. Due to this special terrain restriction, the construction personnel can only rely on the outlet for entry and exit and ventilation; a certain reservoir was built in the 1970s, and the water diversion pipeline has long been eroded by high humidity and high sediment environment, and the corrosion problem is serious; under such harsh working conditions, if traditional manual anti-corrosion operation methods are used, not only the safety risks are extremely high and the efficiency is very low, but also the construction quality fluctuates greatly and is difficult to guarantee; in order to complete this tunnel repair task safely, efficiently and with quality assurance, after careful consideration on site, it was decided to abandon traditional practices, adopt unmanned operation mode, and introduce special anti-corrosion robots to carry out operations.

[0003] The traditional manual operation mode has long faced multiple severe challenges. The inherent defects of its working environment and operating procedures have seriously restricted construction efficiency and quality assurance. Construction workers need to complete their work in a highly restricted physical space. The working area is usually narrow, highly humid, and has complex characteristics such as high dust concentration. In addition, traditional manual operation relies on experience-based judgment and has a significant lack of standardization. Construction workers need to use handheld power tools to process the inner wall of the pipeline section by section. Their working speed is limited by human endurance and equipment power, and the effective working time per day is usually less than 6 hours. At the quality control level, the uncontrollable factors of the traditional model are more prominent. There is a lack of real-time monitoring methods, and it relies entirely on visual inspection and post-inspection, resulting in a significant lag in defect discovery. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0005] An anti-corrosion control system for a pipeline operation robot, comprising:

[0006] The multi-sensor fusion module uses cameras to obtain pipeline images in real time and uses sensors to dynamically collect internal pipeline environmental parameters.

[0007] The 3D modeling and positioning module builds a 3D model of the pipeline based on the SLAM algorithm and pipeline images, and locates the robot's position in the pipeline in real time;

[0008] The intelligent decision-making module uses YOLOv5 to identify pipeline corrosion areas and locations based on pipeline images and conducts corrosion analysis on the corroded areas, including corrosion thickness and morphological contours. It uses electromagnetic eddy current detection to measure the corrosion thickness of the corroded areas and uses deep learning model image recognition to extract the morphological contours of the corroded areas. It also triggers an adaptive control strategy based on the internal environmental parameters of the pipeline and the robot's location on the pipeline.

[0009] The operation execution and interaction module polishes and cleans the corroded area based on the location and corrosion information of the corroded area, performs corrosion analysis at the same time, and makes intelligent decisions on the corroded area until the corroded area is completely cleared.

[0010] Furthermore, the pipeline image acquisition process is as follows:

[0011] The cameras include visible light cameras and infrared cameras, which acquire pipeline images in real time. The camera image quality is monitored online, and the image clarity is evaluated by the PSNR peak signal-to-noise ratio. If image blur is detected, it automatically switches to thermal imaging mode; otherwise, the visible light camera is used first.

[0012] Furthermore, the environmental parameter collection process is as follows:

[0013] Through sensors, the internal environmental parameters of the pipeline are dynamically collected; environmental parameters include temperature and humidity, harmful gas values, moving distance and magnetic field strength values.

[0014] Furthermore, the specific process of constructing the pipeline three-dimensional model is as follows:

[0015] Based on RGB-D vision and combined with pipeline images, a three-dimensional model of the pipeline is constructed:

[0016] S201: Set the initial position and direction and load the pipeline prior map;

[0017] S202: Construct a factor graph to capture the depth information of the scene and generate local point cloud data; fuse the local point cloud data into a global point cloud model and use Poisson reconstruction to generate a triangular mesh model;

[0018] S203: Refine the 3D model and perform semantic annotation on the environment, perform surface smoothing on the generated triangular mesh model, classify the point cloud data using a deep learning model, and combine the optimized 3D model with real-time sensor data to construct a dynamic environment model.

[0019] Furthermore, the process of real-time positioning of the pipeline position of the robot is as follows:

[0020] The distance between the robot and the fixed base station is obtained using UWB ranging technology, combined with the angular velocity and acceleration data of the IMU inertial measurement unit. An RGB-D camera is used to collect depth and color images of the inner wall of the pipe in real time, extract ORB and SIFT feature points, and perform feature matching with the prior map to calculate the robot's current pose, position, and orientation. The local point cloud data is aligned with the global point cloud model, and the accumulated drift caused by sensor noise is corrected by minimizing the pose error. When the robot passes through a mapped area, feature matching is used to correct the accumulated error and close the map gap.

[0021] Furthermore, the process of identifying the corrosion area and location of the pipeline is as follows:

[0022] Identify pipeline corrosion areas and locations:

[0023] S301: Use the improved YOLOv5 model to detect pitting by introducing the BiFPN feature pyramid and CBAM attention mechanism;

[0024] S302: Input the pipeline image to the YOLOv5 model and output the bounding box and category of the corrosion area;

[0025] S303: Refine the contour of the corrosion area in combination with semantic segmentation, generate a pixel-level corrosion morphology map, and classify the corrosion type.

[0026] Furthermore, the morphological profile and the corrosion thickness include:

[0027] The probe scans the corrosion area and records the signal changes on the impedance plane. Using the calibration data, the thickness-impedance relationship is fitted to calculate the actual corrosion thickness.

[0028] Based on deep learning images, a semantic segmentation model is used to extract the pixel-level contours of the corrosion area, and the morphological parameters are calculated by combining the geometric analysis algorithm.

[0029] Furthermore, the process of the adaptive control strategy is:

[0030] Through a multi-dimensional scoring model, the corrosion depth weight, morphological complexity and environmental deterioration factor are output; using a dynamic risk matrix, combined with corrosion location, medium flow rate and historical corrosion rate, a risk heat map is generated.

[0031] Furthermore, the process of making intelligent decisions on the corrosion area is as follows:

[0032] During the polishing process, electromagnetic eddy current detection is continuously used to verify the corrosion depth. At the same time, YOLOv5 is used to compare the images before and after polishing in real time. If the corroded area is not completely removed, the processing time is extended until it is completely removed.

[0033] A corrosion control method for a pipeline operation robot comprises the following steps:

[0034] Step 1: Use the camera to obtain the pipeline image in real time; use the sensor to dynamically collect the internal environmental parameters of the pipeline;

[0035] Step 2: Based on the SLAM algorithm and pipeline image, a 3D model of the pipeline is constructed, and the robot's position in the pipeline is located in real time;

[0036] Step 3: Based on the pipeline image, use YOLOv5 to identify the pipeline corrosion area and location, and perform corrosion analysis on the corrosion area, including corrosion thickness and morphological contours. Use electromagnetic eddy current detection to measure the corrosion thickness of the corrosion area, and use deep learning model image recognition to extract the morphological contours of the corrosion area. Based on the internal environmental parameters of the pipeline and the position of the robot in the pipeline, trigger the adaptive control strategy.

[0037] Step 4: Based on the location and corrosion information of the corrosion area, the corrosion area is polished and cleaned, and corrosion analysis is performed to make intelligent decisions on the corrosion area until the corrosion area is completely cleared.

[0038] The present invention provides an anti-corrosion control system and method for a pipeline operation robot, which has the following beneficial effects:

[0039] (1) The present invention uses an unmanned operation mode, and the robot can independently complete high-risk operations such as rust removal and spraying, avoiding personnel from entering narrow, high-humidity, and high-dust environments, significantly reducing safety risks; combined with RGB-D vision, UWB positioning, IMU inertial navigation and other technologies, the robot can achieve real-time and precise positioning with an accuracy of centimeters, construct a three-dimensional model of the pipeline, and provide a high-precision spatial reference for operation path planning. At the same time, based on YOLOv5 corrosion detection, electromagnetic eddy current thickness measurement and multi-dimensional scoring model, the grinding parameters such as speed and pressure processing strategies are dynamically adjusted to ensure that the corroded area is completely removed without damaging the healthy area.

[0040] (2) The present invention dynamically collects internal environmental parameters of the pipeline through sensors such as temperature and humidity, harmful gases, and magnetic field strength, and combines them with corrosion heat map generation to provide early warning of high-risk areas, correct cumulative posture errors, close map gaps, ensure that the three-dimensional model is consistent with the actual pipeline status, support long-term operation and maintenance data comparison, and ensure the quality of corrosion area treatment through dual verification of electromagnetic eddy current detection and image recognition, reducing the rework rate. At the same time, the system adopts a modular design, and the power tractor, grinding vehicle, and spraying vehicle operate independently, which can quickly adapt to different pipe diameters and operation requirements, thereby improving the equipment reuse rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the system flow of the present invention;

[0042] Figure 2 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0043] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Example 1

[0045] See also Figure 1 , Embodiment 1 of the present application provides an anti-corrosion control system for a pipeline operation robot, the system comprising:

[0046] The multi-sensor fusion module uses cameras to obtain pipeline images in real time and uses sensors to dynamically collect internal pipeline environmental parameters.

[0047] Anti-corrosion robot composition:

[0048] The robot is 6.8 meters long and consists of a power tractor, a grinding and rust removal vehicle, a vacuum cleaning vehicle, a spray anti-corrosion vehicle, a wireless remote control module, and a control box.

[0049] Get pipeline image:

[0050] The camera is located inside the power tractor and transmits real-time images via a wireless bridge, with no delay within a range of 500 meters. The 10-inch monitor on the control box can simultaneously display four video channels. The cameras include visible light cameras and infrared cameras, which acquire image data inside the pipeline in real time.

[0051] Visible light camera: 4-megapixel high-definition camera;

[0052] Infrared camera: used to penetrate interference such as smoke and oil;

[0053] The camera continuously captures images of the inner wall of the pipeline at a fixed frame rate, and uses LED lights to ensure clear images in dark environments. Infrared or thermal imaging cameras are used to penetrate oil stains.

[0054] The camera image quality is also monitored online, with image clarity assessed using the PSNR peak signal-to-noise ratio or SSIM structural similarity metrics. If image blur or environmental interference, such as excessive smoke concentration, is detected, the system automatically switches to thermal imaging mode.

[0055] Internal environmental parameters of the state collection pipeline:

[0056] Temperature and humidity sensor: through Send initialization commands to configure the sensor's operating mode; for example, Default C address is 0x38, need to send , the soft reset command ensures that the sensor enters normal working state; first send a read command, wait for the sensor to complete the measurement, and then read the raw data returned by the sensor, usually a 16-bit temperature value and a 16-bit humidity value. Multiply the 16-bit raw value by the calibration coefficient to obtain the temperature or humidity; perform a secondary correction on the data according to the calibration parameters provided in the sensor manual; and transmit the final result through the wireless module;

[0057] Harmful gas detection sensor: 、 、 Etc., response time < 1s, send initialization command, configure gas detection range, such as CO concentration range 0–500ppm, Range 0–5000ppm; send a read command, trigger the measurement, read the raw data returned by the sensor, and directly output the ppm value, such as SGP30 The concentration range is 400–8192ppm, outputs gas concentration values, and sets threshold alarms. For example, an alarm is triggered when the CO concentration is greater than 50ppm.

[0058] Ultrasonic ranging array: Connect each ultrasonic module to the Trig pin, use multiple GPIOs to control the Trig pins of 8 channels, share one Echo pin or assign an independent Echo pin to each channel; initialize and configure the GPIO pins as output and input, set a timer to measure the high level time of the Echo signal; send an Echo signal to the Trig pin. A high-level pulse starts ultrasonic emission. The high-level duration of the Echo pin is captured by a timer. Through multi-channel polling, the measurement of the eight channels is triggered in sequence, and the distance data of the robot moving in the channel is recorded.

[0059] Through the collaborative operation of multiple ultrasonic sensors, a non-contact measurement system is implemented to measure the distance, position, and shape of target objects. Distance is calculated based on the time difference between ultrasonic emission and reception, and multi-sensor data fusion is used to improve measurement accuracy and environmental perception capabilities. A single ultrasonic sensor transmits high-frequency ultrasonic waves (usually 40kHz) and receives reflected waves, calculating distance based on the round-trip time difference and the speed of sound.

[0060] For example: The typical workflow of the HC-SR04 module is to trigger the TRIG pin to send High level → transmit 8 40kHz square waves → receiver detects echo → ECHO pin outputs high level duration;

[0061] Ultrasonic ranging arrays, by deploying multiple ultrasonic sensors at different locations, can cover a wider detection range and determine the three-dimensional position of a target through triangulation or time difference analysis. Industrial robots use ultrasonic arrays to perceive their surroundings in real time and, combined with path planning algorithms, achieve autonomous obstacle avoidance.

[0062] Electromagnetic field strength detector: Use The interface is connected to the detector. If it is an analog output, the voltage must be converted into a digital signal through ADC; Send configuration commands, set the measurement range and sampling rate, and record the zero offset value in an environment without electromagnetic interference. Get the magnetic field strength value returned by the sensor, convert the ADC reading to a voltage value, and convert the digital value to Unit, use sliding average or low-pass filter to reduce high-frequency noise, output real-time magnetic field strength value, and mark abnormal fluctuations;

[0063] Toxic and harmful gases may exist in closed or semi-closed pipeline environments, such as 、 、 By monitoring the concentration of these gases in real time, the safety of the robot and operators can be ensured to avoid the risk of poisoning or explosion; and through the ultrasonic ranging array, the robot can be prevented from colliding with the internal structure of the pipeline or other obstacles during movement.

[0064] The 3D modeling and positioning module builds a 3D model of the pipeline based on the SLAM algorithm and pipeline images, and locates the robot's position in the pipeline in real time;

[0065] Build a 3D pipeline model:

[0066] By acquiring real-time visual information and environmental parameters inside the pipeline, combined with RGB-D visual SLAM and UWB positioning technology, the robot can accurately locate itself and construct a map of its surroundings in real time using pipeline images captured by cameras and dynamically acquired environmental parameters. This not only accurately determines the robot's real-time position within the pipeline, but also builds a complete 3D model of the pipeline based on the real-time data, providing high-precision spatial reference and environmental perception support for subsequent operations.

[0067] Initial robot position and orientation, using known map information to help initialize, extract features from sensor data and perform matching;

[0068] S201: Set the initial position and orientation, load the pipeline prior map (usually a two-dimensional map), extract feature points such as ORB and SIFT from the environment image, and calculate the robot's posture change through feature matching;

[0069] ORB: uses the FAST algorithm to detect corner feature points in the image, assigns a rotation direction to each feature point using the grayscale centroid method, and generates a binary feature vector in combination with the BRIEF descriptor;

[0070] SIFT: It detects scale-space extreme points through a Difference of Gaussian pyramid and calculates the gradient direction histogram for each key point to achieve rotation invariance, ultimately generating a 128-dimensional floating-point descriptor.

[0071] Feature matching: Use Hamming distance to compare the feature vectors of the two images, select the matching pairs with the highest similarity, remove outliers (i.e., false matches) through Lowe ratio test or RANSAC algorithm, and retain the feature point pairs with high confidence;

[0072] Pose calculation: Based on the matched feature point pairs, the basic matrix is ​​used to estimate the motion parameters of the camera or robot relative to the environment. The rotation matrix and translation vector are extracted by decomposing the matrix, and combined with the prior map or initial pose, the 3D pose change of the robot in the environment is finally determined.

[0073] S202: Construct a factor graph, minimize the pose error, fuse the local point cloud data into a global point cloud model, and generate a triangular mesh model using Poisson reconstruction or Marching Cubes algorithm;

[0074] Factor graph optimization: Local point cloud data collected by the robot at different locations is used as nodes in the factor graph. The relative pose constraints of adjacent nodes are calculated through feature matching to establish a graph structure. The factor graph is optimized using nonlinear least squares to correct accumulated pose errors caused by sliding, vibration, or sensor noise, ensuring the spatial consistency of the point cloud data in the global coordinate system.

[0075] Point cloud fusion: Based on the optimized pose information, the local point cloud data is aligned to the global coordinate system through rigid transformation, and voxel filtering or octree structure is used to remove duplicates and merge overlapping areas to generate a continuous global point cloud model;

[0076] Mesh reconstruction: Poisson reconstruction: By solving the Poisson equation, the normal vector information of the point cloud is converted into an implicit function field, and then a smooth and watertight triangular mesh model is generated through isosurface extraction. By dividing the 3D voxel space, the point cloud density distribution is interpolated, and isosurfaces are extracted to generate a triangular mesh, which is suitable for processing regularly sampled or low-noise point cloud data.

[0077] S203: Refine the 3D model and perform environmental semantic annotation. Smooth the surface of the generated triangular mesh model. Use a deep learning model to classify the point cloud data. Combine the optimized 3D model with real-time sensor data to construct a dynamic environmental model to reflect the corrosion status of the pipeline over time.

[0078] Real-time positioning of the robot in the pipeline:

[0079] UWB ranging technology is used to obtain the precise distance between the robot and the fixed base station, with centimeter-level accuracy. Combined with the angular velocity and acceleration data from the IMU (Inertial Measurement Unit), multi-source data is fused through extended Kalman filtering or particle filtering algorithms to eliminate errors from single sensors, such as IMU drift and UWB multipath interference. In complex pipeline environments, UWB provides a global positioning reference, while the IMU compensates for short-term motion trajectories, achieving continuous and stable positioning.

[0080] The RGB-D camera collects real-time depth and color images of the inner wall of the pipeline, extracts feature points such as ORB and SIFT, and performs feature matching with the prior map (the two-dimensional map in S201) to calculate the robot's current pose, position, and orientation. The local point cloud data is aligned with the global point cloud model, minimizing pose errors and correcting accumulated drift caused by sensor noise or motion uncertainty. When the robot passes through a mapped area, feature matching, such as loop closure detection, is used to identify repeated scenes, correct accumulated errors, and close map gaps.

[0081] For example:

[0082] The water diversion tunnel of a certain reservoir is a 1300mm diameter steel pipe. Due to long-term sediment erosion, the inner wall is severely corroded. During the repair project, a robot equipped with an RGB-D camera collects real-time depth and color images of the inner wall of the pipe, and combines them with the prior 2D map for positioning and mapping.

[0083] The RGB-D camera captures 60 frames of images per second with a depth accuracy of ±2cm. ORB feature points and SIFT feature points are extracted from the color image. When the robot moves in the straight section of the pipeline, ORB is used to quickly detect the threaded joint feature points on the pipe wall. In the elbow area, SIFT identifies irregular spots caused by corrosion and uses them as a basis for robust matching. The feature points of the current frame are matched with the prior two-dimensional map, and high-confidence matching pairs are selected using the Hamming distance. After removing outliers using the RANSAC algorithm, the PnP algorithm is used to calculate the rotation matrix and translation vector of the camera relative to the prior map to determine the current robot pose. If the accumulated pose error causes the robot to mistakenly deviate from the pipeline axis, the pose is corrected using the matched elbow feature points.

[0084] The RGB-D camera generates a local point cloud every 1m of movement, covering the entire 360° circumference of the pipeline. The ICP algorithm is used to align the local point cloud with the global point cloud model, correcting sensor noise or motion drift by minimizing the Euclidean distance between the point clouds. For example, in the pipeline outlet lock chamber section, the robot slipped, causing the local point cloud to shift by 2cm. After ICP alignment, it was restored to the designed axis position. When the robot returns to the mapped area, the bag-of-words model is used to quickly match the current frame feature points with the historical map feature library to identify repeated scenes. Through factor graph optimization, the current pose is constrained to the prior pose, correcting accumulated drift and closing map gaps. After closure, the global point cloud model automatically merges repeated areas to generate a continuous and seamless three-dimensional pipeline model.

[0085] The intelligent decision-making module uses YOLOv5 to identify pipeline corrosion areas and locations based on pipeline images and conducts corrosion analysis on the corroded areas, including corrosion thickness and morphological contours. It uses electromagnetic eddy current detection to measure the corrosion thickness of the corroded areas and uses deep learning model image recognition to extract the morphological contours of the corroded areas. It also triggers an adaptive control strategy based on the internal environmental parameters of the pipeline and the robot's location on the pipeline.

[0086] Identify pipeline corrosion areas and locations:

[0087] S301: Using an improved YOLOv5 model, the BiFPN feature pyramid and CBAM attention mechanism are introduced to improve the detection accuracy of pitting corrosion. The training dataset contains annotated images covering various corrosion types.

[0088] S302: Input the pipeline image to the YOLOv5 model, and output the bounding box and category of the corrosion area, such as cracks, holes, and corrosion patches;

[0089] S303: Combine semantic segmentation to refine the contours of the corrosion area and generate a pixel-level corrosion morphology map. Corrosion type classification is also performed, and the detected area is classified according to seven types of corrosion labels, such as fracture, surface corrosion, and deformation.

[0090] Corrosion thickness:

[0091] The eddy current probe emits an alternating magnetic field, inducing eddy currents on the metal surface. The material thickness is inferred based on the change in eddy current impedance. The probe scans the corroded area and records the signal changes on the impedance plane. Using calibration data, such as benchmark samples of different thicknesses, the thickness-impedance relationship is fitted to calculate the actual corrosion thickness with an error of ≤0.1mm.

[0092] For example:

[0093] The eddy current probe emits a high-frequency alternating magnetic field through an electromagnetic coil. This magnetic field penetrates the surface of the metal material and induces eddy currents related to the material properties. The thickness and conductivity of the pipe will affect the flow path and intensity of the eddy currents, causing the impedance of the probe coil to change. The impedance value change is recorded by a high-precision sensor. The eddy current probe is moved to scan the corrosion area point by point to generate an impedance plane diagram, such as a phase-amplitude diagram. The signal difference in different areas reflects the unevenness of the material thickness. Experiments are performed using standard samples of known thickness, such as metal test blocks of different thicknesses, and their corresponding impedance values ​​are measured to establish a mathematical model of thickness and impedance, such as linear regression or polynomial fitting. The impedance data of the scanned corrosion area is substituted into the calibration model to infer the actual thickness value of each point, so that the error of the detection system is controlled within 0.1 mm to meet the needs of high-precision detection.

[0094] Morphological outline:

[0095] Based on deep learning images, a semantic segmentation model is used to extract the pixel-level contours of the corrosion area, and combined with a geometric analysis algorithm, morphological parameters such as the corrosion volume are calculated;

[0096] Map the bounding box of YOLOv5 and the thickness data of electromagnetic eddy current to the same coordinate system to generate the thickness and morphological features of the 3D corrosion model;

[0097] Adaptive control strategy:

[0098] Through the multi-dimensional scoring model, the corrosion depth weight, morphological complexity and environmental deterioration factor are output;

[0099] Corrosion depth weight: thickness loss ≥30% of the wall thickness is high risk, 10%~30% is medium risk, and <10% is low risk;

[0100] Morphological complexity: Irregular corrosion, such as cracks and pits, should be treated first, while regular corrosion, such as uniform thinning, can be postponed;

[0101] Environmental degradation factor: High Concentration, high SRB bacteria content or high temperature and high pressure environment require higher treatment priority;

[0102] Generate a risk heat map using a dynamic risk matrix, combined with corrosion location, medium flow rate, and historical corrosion rate;

[0103] Table 1: Strategy classification and parameter mapping table:

[0104]

[0105] Dynamic adjustment logic: Force control based on corrosion thickness: Dynamically adjust the pressure of the robotic arm through the PID algorithm. For example, when the thickness loss is 20%, the pressure increases linearly from 0.5MPa to 1.5MPa.

[0106] Morphology-based path optimization: "Z"-shaped trajectory coverage is used for groove corrosion, and fixed-point multiple impacts are used for pitting corrosion;

[0107] Environmental adaptive compensation: reduce the grinding speed in high temperature (>60℃) or high humidity environment to avoid overheating damage. Increase the frequency of corrosion inhibitor spraying when the concentration is >100,000 mg / L;

[0108] The operation execution and interaction module polishes and cleans the corroded area based on its location and corrosion information, and performs corrosion analysis to make intelligent decisions on the corroded area until the corroded area is completely cleared.

[0109] During the polishing process, electromagnetic eddy current testing is continuously used to verify the thickness restoration effect. At the same time, YOLOv5 is used to compare the images before and after polishing in real time. If the corroded area is not completely removed, the processing time is extended;

[0110] Safety threshold protection: Maximum processing time limit: single grinding should not exceed 30 seconds to avoid excessive wear of healthy areas;

[0111] If abnormal pipeline vibration or sudden temperature rise is detected, the operation will be suspended immediately;

[0112] The corrosion area is judged to be completely cleared and multi-source data verification is performed; electromagnetic eddy current detection is used to confirm that the thickness of the corrosion area has recovered to more than 95% of the original value, and the grinding of the corrosion area is stopped; through image recognition, YOLOv5 is used to detect that there are no residual corrosion patches and the grinding is stopped.

[0113] Example 2

[0114] See also Figure 2 Based on Example 1, Example 2 of the present application further provides an anti-corrosion control method for a pipeline operation robot, comprising the following specific steps:

[0115] Step 1: Use the camera to obtain the pipeline image in real time; use the sensor to dynamically collect the internal environmental parameters of the pipeline;

[0116] Step 2: Based on the SLAM algorithm and pipeline image, a 3D model of the pipeline is constructed, and the robot's position in the pipeline is located in real time;

[0117] Step 3: Based on the pipeline image, use YOLOv5 to identify the pipeline corrosion area and location, and perform corrosion analysis on the corrosion area, including corrosion thickness and morphological contours. Use electromagnetic eddy current detection to measure the corrosion thickness of the corrosion area, and use deep learning model image recognition to extract the morphological contours of the corrosion area. Based on the internal environmental parameters of the pipeline and the position of the robot in the pipeline, trigger the adaptive control strategy.

[0118] Step 4: Based on the location and corrosion information of the corrosion area, the corrosion area is polished and cleaned, and corrosion analysis is performed to make intelligent decisions on the corrosion area until the corrosion area is completely cleared.

[0119] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0120] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0121] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An anti-corrosion control system for a pipeline operation robot, characterized in that: The system includes: The multi-sensor fusion module uses cameras to obtain pipeline images in real time and uses sensors to dynamically collect internal pipeline environmental parameters. The 3D modeling and positioning module builds a 3D model of the pipeline based on the SLAM algorithm and pipeline images, and locates the robot's position in the pipeline in real time; The intelligent decision-making module uses YOLOv5 to identify pipeline corrosion areas and locations based on pipeline images and conducts corrosion analysis on the corroded areas, including corrosion thickness and morphological contours. It uses electromagnetic eddy current detection to measure the corrosion thickness of the corroded areas and uses deep learning model image recognition to extract the morphological contours of the corroded areas. It also triggers an adaptive control strategy based on the internal environmental parameters of the pipeline and the robot's location on the pipeline. The process of identifying pipeline corrosion areas and locations is as follows: Identify pipeline corrosion areas and locations: S301: Use the improved YOLOv5 model to detect pitting by introducing the BiFPN feature pyramid and CBAM attention mechanism; S302: Input the pipeline image to the YOLOv5 model and output the bounding box and category of the corrosion area; S303: Refine the contour of the corrosion area by combining semantic segmentation, generate a corrosion morphology map, and classify the corrosion type; The morphological profile and the corrosion thickness include: The probe scans the corrosion area and records the signal changes on the impedance plane. Using the calibration data, the thickness-impedance relationship is fitted to calculate the actual corrosion thickness. Based on deep learning images, a semantic segmentation model is used to extract the pixel-level contours of the corrosion area, and the morphological parameters are calculated by combining the geometric analysis algorithm. The process of the adaptive control strategy is: A multi-dimensional scoring model is used to output corrosion depth weights, morphological complexity, and environmental deterioration factors. A dynamic risk matrix is ​​used to combine corrosion location, medium flow rate, and historical corrosion rates to generate a risk heat map. The operation execution and interaction module polishes and cleans the corroded area based on its location and corrosion information, performs corrosion analysis, and makes intelligent decisions on the corroded area until the corroded area is completely cleared. The process of making intelligent decisions on corrosion areas is as follows: During the polishing process, electromagnetic eddy current detection is continuously used to verify the corrosion depth. At the same time, YOLOv5 is used to compare the images before and after polishing in real time. If the corroded area is not completely removed, the processing time is extended until it is completely removed.

2. The anti-corrosion control system for a pipeline operation robot according to claim 1, characterized in that: The pipeline image acquisition process is as follows: The cameras include visible light cameras and infrared cameras, which acquire pipeline images in real time. The camera image quality is monitored online, and the image clarity is evaluated by the PSNR peak signal-to-noise ratio. If image blur is detected, it automatically switches to thermal imaging mode; otherwise, the visible light camera is used first.

3. The anti-corrosion control system for a pipeline operation robot according to claim 1, characterized in that: The process of collecting the environmental parameters is as follows: Through sensors, the internal environmental parameters of the pipeline are dynamically collected; environmental parameters include temperature and humidity, harmful gas values, moving distance and magnetic field strength values.

4. The anti-corrosion control system for a pipeline operation robot according to claim 3, characterized in that: The specific process of constructing the pipeline three-dimensional model is as follows: Based on RGB-D vision and combined with pipeline images, a three-dimensional model of the pipeline is constructed: S201: Set the initial position and direction and load the pipeline prior map; S202: Construct a factor graph to capture the depth information of the scene and generate local point cloud data; fuse the local point cloud data into a global point cloud model and use Poisson reconstruction to generate a triangular mesh model; S203: Refine the 3D model and perform semantic annotation on the environment, perform surface smoothing on the generated triangular mesh model, classify the point cloud data using a deep learning model, and combine the optimized 3D model with real-time sensor data to construct a dynamic environment model.

5. The anti-corrosion control system for a pipeline operation robot according to claim 4, characterized in that: The process of real-time positioning of the pipeline position of the robot is as follows: The distance between the robot and the fixed base station is obtained using UWB ranging technology, combined with the angular velocity and acceleration data of the IMU inertial measurement unit. An RGB-D camera is used to collect depth and color images of the inner wall of the pipe in real time, extract ORB and SIFT feature points, and perform feature matching with the prior map to calculate the robot's current pose, position, and orientation. The local point cloud data is aligned with the global point cloud model, and the accumulated drift caused by sensor noise is corrected by minimizing the pose error. When the robot passes through a mapped area, feature matching is used to correct the accumulated error and close the map gap.

6. A corrosion control method for a pipeline operation robot, characterized in that: The steps include: Step 1: Use the camera to obtain the pipeline image in real time; use the sensor to dynamically collect the internal environmental parameters of the pipeline; Step 2: Based on the SLAM algorithm and pipeline image, a 3D model of the pipeline is constructed, and the robot's position in the pipeline is located in real time; Step 3: Based on the pipeline image, use YOLOv5 to identify the pipeline corrosion area and location, and perform corrosion analysis on the corrosion area, including corrosion thickness and morphological contours. Use electromagnetic eddy current detection to measure the corrosion thickness of the corrosion area, and use deep learning model image recognition to extract the morphological contours of the corrosion area. Based on the internal environmental parameters of the pipeline and the position of the robot in the pipeline, trigger the adaptive control strategy. The process of identifying pipeline corrosion areas and locations is as follows: Identify pipeline corrosion areas and locations: S301: Use the improved YOLOv5 model to detect pitting by introducing the BiFPN feature pyramid and CBAM attention mechanism; S302: Input the pipeline image to the YOLOv5 model and output the bounding box and category of the corrosion area; S303: Refine the contour of the corrosion area by combining semantic segmentation, generate a corrosion morphology map, and classify the corrosion type; The morphological profile and the corrosion thickness include: The probe scans the corrosion area and records the signal changes on the impedance plane. Using the calibration data, the thickness-impedance relationship is fitted to calculate the actual corrosion thickness. Based on deep learning images, a semantic segmentation model is used to extract the pixel-level contours of the corrosion area, and the morphological parameters are calculated by combining the geometric analysis algorithm. The process of the adaptive control strategy is: A multi-dimensional scoring model is used to output corrosion depth weights, morphological complexity, and environmental deterioration factors. A dynamic risk matrix is ​​used to combine corrosion location, medium flow rate, and historical corrosion rates to generate a risk heat map. Step 4: Based on the location and corrosion information of the corroded area, the corroded area is polished and cleaned, and corrosion analysis is performed to make intelligent decisions on the corroded area until the corroded area is completely cleared; The process of making intelligent decisions on corrosion areas is as follows: During the polishing process, electromagnetic eddy current detection is continuously used to verify the corrosion depth. At the same time, YOLOv5 is used to compare the images before and after polishing in real time. If the corroded area is not completely removed, the processing time is extended until it is completely removed.

Images