A multifunctional remote operating tool for a subsea Christmas tree and an intelligent identification control method thereof

Abstract

This invention relates to a multifunctional remote operation tool for underwater production trees and its intelligent identification and control method, comprising: a display controller; a connecting component, the upper end of which is connected to the display controller and also to a survey vessel; and an operation component, the lower end of which is connected to the connecting component and communicatively connected to the display controller. The operation component includes an operation processing module, on which a robotic arm is mounted. The robotic arm is used for opening and closing valves of shallow-water underwater production trees, inspecting underwater production trees, assisting in underwater installation, and retrieving underwater debris. Compared to traditional underwater robots, this tool has a simpler structure, more intuitive operation, and exhibits greater flexibility and economy in shallow water. It can be widely used in shallow-water marine oil and gas development, underwater infrastructure construction, underwater salvage, and emergency maintenance.

Description

Technical Field

[0001] This invention relates to a multi-functional remote operation tool for underwater wellheads and its intelligent identification and control method, belonging to the technical field of marine oil and gas extraction equipment. Background Technology

[0002] With the continuous development of offshore oil and gas resources, subsea production trees, as key equipment in subsea production systems, require increasingly frequent installation, commissioning, maintenance, and troubleshooting. Among these, valve control of the subsea production tree is a crucial aspect of oil extraction, typically requiring manual operation to open and close the valves. In shallow waters such as the Bohai Sea in my country, the installation of subsea production trees and valve control have traditionally relied on human divers or underwater robots. Human diving operations are limited by water depth, operating window time, and environmental factors (icing in winter), and also pose safety risks. Traditional underwater robots, in shallow water environments, suffer from poor visibility, lack of flexibility and adaptability due to the presence of seabed sediment and the agitation caused by propellers, and are also costly. Furthermore, with the expansion of underwater operations, tasks are no longer limited to valve operation but also include visual observation, structural inspection, operational status diagnosis of the subsea production tree, and the detection and retrieval of commonly lost underwater tools or components. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a multi-functional remote operation tool for underwater production trees and its intelligent identification and control method. This tool features a compact structure and high functional integration, enabling it to perform multiple tasks such as valve operation, structural status inspection, target identification and positioning, lighting guidance, and underwater object detection and retrieval without relying on manual diving. This solves the problems of high risk associated with manual diving, high cost of underwater robots, inconvenient operation, fragmented functions, and poor environmental adaptability in existing technologies.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A multi-functional remote operation tool for underwater oil production trees includes: Display controller; A connecting component, the upper end of which is connected to the display controller, and the connecting component is also connected to the survey vessel; An operating component is connected to the lower end of the connecting component and communicates with the display controller. The operating component includes a thruster and an operating processing module. The operating processing module is equipped with a robotic arm, which is used to open and close the subsea tree frame and the subsea tree valves. The thruster is used to control the operating processing module to reach the target position.

[0006] Preferably, in the aforementioned multi-functional remote operation tool for underwater oil production trees, the robotic arm includes a rotating electromagnetic arm and a clamping robotic arm. The rotating electromagnetic arm is connected to the base plate of the operation processing module via an intelligent recognition camera, and the clamping robotic arm is mounted on a frame connected to the base plate of the operation processing module.

[0007] Preferably, in the aforementioned multi-functional remote operation tool for underwater oil production trees, the robotic arm further includes several sliding rail robotic arms, which slidably connect the base plate of the operation processing module to the frame.

[0008] Preferably, the frame of the underwater oil production tree multi-functional remote operation tool is a closed slide rail frame formed by connecting several slide rails end to end, and several slide rail robotic arms are arranged in one-to-one correspondence with several slide rails, and the connector of the slide rail robotic arm is slidably connected to the slide rail.

[0009] Preferably, each of the aforementioned underwater oil production tree multi-functional remote operation tools is equipped with a long strip-shaped first lighting lamp.

[0010] Preferably, the multi-functional remote operation tool for underwater oil production trees is further equipped with a sonar module and a second lighting lamp on the base plate of the operation processing module.

[0011] Preferably, the multi-functional remote operation tool for underwater production trees is further equipped with a metal detector and a pipe sleeve on the rotating mechanical electromagnetic arm.

[0012] Preferably, the multi-functional remote operation tool for underwater production trees includes a telescopic column, a flexible hose column, a tubing column, and a bend in the connection assembly. The tubing column includes a first tubing column and a second tubing column. The display controller is connected to the upper end of the telescopic column, and the lower end of the telescopic column is connected to the upper end of the first tubing column. The upper and lower ends of the flexible hose column are connected to the lower end of the first tubing column and the upper end of the second tubing column, respectively, via flanges. The vertical end of the bend is connected to the lower end of the second tubing column, and the horizontal end of the bend is connected to the operation processing module.

[0013] The present invention has the following advantages due to the adoption of the above technical solutions: 1. The multi-functional remote operation tool for underwater production trees in shallow nearshore waters provided by this invention enables remote operation, avoiding the high risks associated with manual diving operations. This is particularly beneficial in shallow waters, cold seas, or harsh sea conditions, significantly improving the safety and controllability of operations. Furthermore, it achieves multi-functional integrated operation: this invention integrates valve operation, structural inspection, underwater target identification, lighting guidance, and object detection and retrieval into a single device, reducing the frequency of equipment switching and the difficulty of system coordination, thus greatly improving the efficiency of underwater operations.

[0014] 2. By improving the original YOLO11 network's feature propagation, multi-scale fusion, and detection confidence formation mechanisms, which rely entirely on visual information, sonar spatial structure information is introduced as a constraint, thereby enhancing the robustness, safety, and reliability of underwater target detection in complex environments. This approach does not simply overlay sonar information into the YOLO11 network; rather, it uses sonar information to constrain and modify key decision variables within the YOLO11 network, forming an improved YOLO11 detection framework. Attached Figure Description

[0015] Figure 1 This is a working diagram of a multi-functional remote operation tool for underwater oil production trees provided in an embodiment of the present invention; Figure 2 This is a top view of the multi-functional remote operation tool for underwater oil production trees provided in this embodiment of the present invention; Figure 3 This is a lower-middle schematic diagram of the multi-functional remote operation tool for underwater oil production trees provided in this embodiment of the present invention; Figure 4 This is a schematic diagram of the operation processing module in the multi-functional remote operation tool for underwater oil production trees provided in this embodiment of the present invention; Figure 5 This is a partial schematic diagram of the rotating mechanical electromagnetic arm in the multi-functional remote operation tool for underwater oil production trees provided in this embodiment of the present invention; Figure 6 This is a flowchart illustrating the intelligent identification and operation of the multi-functional remote operation tool for underwater oil production trees provided in this embodiment of the invention. Figure 7 This is a schematic diagram of the improved YOLO11 vision-sonar cooperative underwater target detection according to the present invention; The attached figures are labeled as follows: 1- Multifunctional remote tool for underwater production trees; 2- Underwater production tree frame; 11-Display controller; 12-Telescopic column; 13-Thruster; 14-Snap ring; 15-Flange; 16-Hose column; 17-Pipe column; 18-Bend; 19-Operation processing module; 171 - First tubing string; 172 - Second tubing string; 191-Operation processing module base plate; 192-Slide rail; 193-First lighting lamp; 194-Slide rail robotic arm; 195-Clamping robotic arm; 196-Sonar module; 197-Second lighting lamp; 198-Intelligent recognition camera; 199-Rotating mechanical electromagnetic arm; 1991-Metal detector; 1992-Pipe sleeve. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0017] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," "third," "fourth," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0018] For ease of description, spatial relative terms may be used in the text to describe the relationship of one element or feature relative to another element or feature as shown in the figure. These relative terms include, for example, "inside," "outside," "middle," "outer," "below," "above," etc. Such spatial relative terms are intended to include different orientations of the device in use or operation, other than those depicted in the figure.

[0019] Currently, the installation of subsea production trees and the control of valves rely on human divers or underwater robots. Human diving operations are limited by water depth, working window time, and environmental factors (ice formation in winter), and also pose safety risks. Traditional underwater robots suffer from poor visibility, lack flexibility and adaptability in such environments, and are costly. Furthermore, with the expansion of underwater operations, these tasks are no longer limited to valve operation but also include visual inspection of subsea production trees, structural checks, operational status diagnostics, and the detection and retrieval of commonly dropped underwater tools or components.

[0020] To address the aforementioned technical problems, this invention provides a multi-functional remote operation tool for underwater production trees. This tool features a compact structure and high functional integration, enabling it to perform multiple tasks such as valve operation, structural status inspection, target identification and positioning, lighting guidance, and underwater object detection and retrieval without relying on manual diving. This solves the problems of high risks associated with manual diving, high costs of underwater robots, inconvenient operation, fragmented functions, and poor environmental adaptability in existing technologies.

[0021] like Figure 1 , 2 As shown in Figures 1 and 3, the multi-functional remote operation tool for underwater production trees involved in this invention includes: Display controller 11 is used to display operation information in real time. Display controller 11 is connected to operation processing module 19 via a connection assembly. The connection assembly includes a telescopic column 12, a pipe column 17 (including a first pipe column 171 and a second pipe column 172), a flexible hose column 16, and a bend 18 connected in series. Display controller 11 is connected to the top of telescopic column 12. The lower end of telescopic column 12 is connected to the upper end of first pipe column 171 via a retaining ring 14. Two pushers 13 are connected to first pipe column 171 via retaining ring 14. The upper and lower ends of flexible hose column 16 are connected to the lower end of first pipe column 171 and the upper end of second pipe column 172 via flanges 15. The lower end of second pipe column 172 is connected to the vertical end of bend 18 via retaining ring 14. The horizontal end of bend 18 is connected to operation processing module 19. Figure 1 , 2 As shown in Figure 3, the telescopic column 12 has four telescopic sections, allowing for seamless docking with various survey vessel types. The thruster 13 propels the operating assembly 19 to better reach the target position. The flexible hose column 16 rises and falls according to sea level movements, effectively compensating for ocean heave.

[0022] Furthermore, such as Figure 4 , 5 As shown, the operation processing module 19 includes an operation processing module base plate 191 and a robotic arm. The robotic arm includes a slide rail robotic arm 194, a clamping robotic arm 195, and a rotary electromagnetic arm 199. The clamping robotic arm 195 is mounted on a frame connected to the operation processing module base plate 191. The frame is a closed slide rail frame formed by connecting several slide rails 192 end to end. The slide rail robotic arms 194 slidably connect the operation processing module base plate 191 to the frame. The number of slide rail robotic arms 194 corresponds one-to-one with the number of slide rails 192, and the connectors of the slide rail robotic arms 194 are slidably connected to the slide rails 192. The rotating mechanical electromagnetic arm 199 is fixed to the lower part of the intelligent recognition camera 198. It is electrically controlled to extend, retract, and rotate to open and close the underwater production tree valve. The rotating mechanical electromagnetic arm 199 integrates a metal detector 1992, a pipe sleeve 1992, and electromagnetic components. The metal detector 1992 and electromagnetic components can efficiently perform the task of locating and recovering underwater objects. The pipe sleeve 1992 can be used to tighten or loosen the bolts on the underwater production tree frame 2.

[0023] Specifically, such as Figure 4As shown, in a specific embodiment of the present invention, the operation processing module base plate 191 is square, and the frame is a closed slide rail frame formed by four slide rails 192 connected end to end. Four mechanical arms 195 are clamped and positioned at the connection points of adjacent slide rails 192. Four slide rail mechanical arms 194 are also provided, with one end fixed to each of the four corners of the operation processing module base plate 191, and the other end slidably connected to each of the four slide rails 192. Connectors are provided at the ends of the slide rail mechanical arms 194, and these connectors are slidably connected to the slide rails 192. An intelligent recognition camera 198 is positioned at the center of the operation processing module base plate 191, and a rotating mechanical electromagnetic arm 199 is fixed to the lower part of the intelligent recognition camera 168. Through the sliding of the four slide rail mechanical arms 194, the rotating mechanical electromagnetic arm 199 can reach any position within the slide rail plane. The gripping robotic arm 195 is used to grasp the underwater production tree frame 2, and the slide rail 192 is equipped with a bright, long strip-shaped first lighting lamp 193 around its perimeter, providing lighting and directional guidance for the installation and operation of the production tree by the tool of the present invention.

[0024] Furthermore, such as Figure 4 As shown, the operation processing module base plate 191 integrates a sonar module 196, a second lighting lamp 197, and an intelligent recognition camera 198. The sonar module 196 can detect and locate underwater targets; the second lighting lamp 197 can adjust its brightness according to changes in ambient light to ensure clear visibility of the work area; the intelligent recognition camera 198 can capture images of the work site in real time and transmit them to the display controller 11, while simultaneously capturing the position of the underwater production tree valve to be processed and feeding it back to the operation processing module 191. The operation processing module 191 controls the movement of the slide rail robotic arm 194, enabling the rotating mechanical electromagnetic arm 169 to accurately reach any position within the range of the slide rail 192.

[0025] The working principle of the device of this invention is as follows: After the multi-functional remote tool 1 for the underwater production tree is lowered to the working position of the underwater production tree frame 2, the second lighting 192, the high-brightness first lighting 193, and the intelligent recognition camera 198 are first turned on to ensure that the working area is bright and that real-time images are fed back to the display controller 11. Next, the gripping robotic arm 195 is manipulated to steadily grasp the underwater production tree frame 2 to ensure a stable operation. Based on the real-time data provided by the intelligent recognition camera 198, the sliding rail robotic arm 194 is controlled to move to the designated switch valve position of the underwater production tree. Subsequently, using the rotating mechanical electromagnetic arm 199, the extension and retraction are driven by the electronic control system to precisely grasp the valve and perform precise valve opening and closing operations.

[0026] After the valve control operation is completed, if further inspection or diagnosis of the subsea production tree frame 2 is required, the control system can acquire real-time images and video data through the intelligent recognition camera 198 to assess and detect the valve status. All monitoring data during this process will be fed back to the display controller 11 in real time, facilitating the operator to make appropriate decisions. In addition, if tools or parts fall during the operation, the metal detector 1991, electromagnetic components, and sonar module 196 can quickly detect and locate the fallen object, and combine it with the rotating mechanical electromagnetic arm 199 for precise retrieval, ensuring that lost parts are recovered in a timely manner and avoiding resource waste or equipment damage.

[0027] The slide rail robotic arm 194 of this invention can drive the rotating mechanical electromagnetic arm 199 to any position within the slide rail plane. The clamping robotic arm 195 is fixed around the slide rail 192 and can grip the underwater tree frame 2. The slide rail 192 is equipped with high-brightness first lighting lamps 193 around its perimeter to provide illumination and directional guidance for the installation and operation of the tree. The rotating mechanical electromagnetic arm 199 is fixed to the lower part of the intelligent recognition camera 198 and is electrically controlled to extend, retract, grip, rotate, and open / close the underwater tree valves. Simultaneously, the rotating mechanical electromagnetic arm 199 integrates a metal detector 1991 and electromagnetic components, which can efficiently perform the task of locating and recovering underwater objects. The pipe sleeve 1992 can be used to tighten or loosen bolts on the underwater tree frame 2.

[0028] Specifically, such as Figure 6 As shown, the intelligent recognition method of the tool of the present invention includes the following steps: Step 1: Image Acquisition: The intelligent recognition camera 198 acquires images of the underwater operation scene under the supplementary lighting provided by the first illumination lamp 193 and the second illumination lamp 197. At the same time, the sonar module 196 acquires spatial point cloud information to assist in 3D modeling and positioning. The operation processing module 19 uses a built-in multi-sensor data synchronization and fusion system and timestamp synchronization technology to ensure the spatiotemporal alignment of visual data and sonar data.

[0029] The image acquisition method of this invention combines an improved YOLO11 network with sonar information for underwater target detection, thereby improving the robustness and reliability of target detection in complex underwater environments.

[0030] The specific steps for image acquisition are as follows: Step 1: Acquire visual image information and sonar echo information of the underwater operation scene; Step 2: Use the improved YOLO11 network to extract features from the visual image and generate multi-level visual features; Step 3: Map the sonar echo information to the spatial region corresponding to the visual features, evaluate the credibility of the visual features, and obtain the sonar evaluation results; Step 4: During the feature propagation process of the improved YOLO11 network, adjust the contribution of visual features based on the sonar evaluation results; Step 5: In the multi-scale feature fusion stage of the improved YOLO11 network, sonar constraints are introduced to dynamically adjust the fusion weights of features at different scales to obtain the fused features. Step 6: Perform target detection based on the fused features and output the initial detection results; Step 7: Combine the spatial consistency information provided by sonar to correct the confidence level of the initial detection results and obtain the final underwater target detection results.

[0031] The detection network used in this invention is based on the YOLO11 network framework. The overall structure still employs an end-to-end architecture consisting of a backbone feature extraction network, a feature fusion network, and a detection head. A detailed schematic diagram of this network is shown below. Figure 7 As shown.

[0032] Based on this, the present invention systematically improves the original detection paradigm of YOLO11, enabling sonar information to participate in the detection decision-making process in the following three key stages: 1. Backbone Feature Extraction Stage: Sonar-constrained visual feature credibility; 2. Multi-scale feature fusion stage (Neck): Sonar-constrained multi-scale feature fusion weights; 3. Detection result output stage (Head): Confidence level of sonar constraint detection results.

[0033] Through the above improvements, sonar information is no longer used as external reference information, but as an internal constraint, deeply involved in the detection decision-making process of the improved YOLO11 network.

[0034] The specific improvements to the YOLO11 network include the following three aspects: 1. An Improved YOLO11 Network Feature Reliability Control Method Based on Sonar Constraints In complex underwater environments, visual images are susceptible to scattering noise, suspended particles, and light attenuation, causing some non-target areas to exhibit abnormally enhanced responses in the feature space. To improve the reliability of visual features in subsequent detection processes, this method introduces sonar information to evaluate the spatial reliability of visual features and adjusts the feature propagation process accordingly.

[0035] Assuming the input visual image is Sonar echoes or sonar images are First, the sonar echoes are structurally encoded to obtain a sonar structure reliability map. This is used to describe the probability of a stable physical structure existing at various spatial locations. The sonar structure reliability map can be derived from the sonar intensity consistency... Local structural continuity and local stability The results were obtained by weighting the indicators as follows:

[0036]

[0037]

[0038]

[0039] in, This represents the spatial gradient of the sonar echo. It is a very small positive number, used to avoid numerical anomalies where the denominator is zero during the normalization process. This is the gradient suppression scaling parameter, used to control the degree to which the gradient magnitude affects the evaluation of structural continuity. This is the local variance suppression scaling parameter, used to control the impact of local statistical instability on credibility assessment. This represents the variance statistics within a local window. These are the weighting coefficients. The sonar structure credibility map... Scale mapping and spatial alignment are used to map to the corresponding visual feature scale. Let's say a visual feature at a certain scale... Represented as:

[0040] Sonar confidence map at corresponding scale for:

[0041] in Indicates downsampling / interpolation. For the first Spatial height of layer visual features For the first Spatial width of layer visual features For the first The number of channels in the layer visual features is used to further construct a visual region confidence weight map. :

[0042] in For the Sigmoid mapping function, As a confidence threshold, This is the adjustment coefficient.

[0043] Based on the credibility weight map, spatial-level modulation of visual features is performed, i.e. :

[0044] in, This represents element-wise multiplication. This is the weighted average, used to stabilize the overall amplitude. This is the spatial modulation intensity coefficient, used to control the degree of adjustment of sonar credibility to the spatial response of visual features.

[0045] Furthermore, the visual feature channel dimensions can be adaptively adjusted. First, global statistics are performed on the adjusted features, i.e. :

[0046] Ultimately, the output is a reliable visual feature. It can be represented as:

[0047] By employing the above methods, the reliability of sonar in detecting visual features can be controlled at both the spatial and channel levels, thereby suppressing the interference of noise region features on the detection process.

[0048] 2. An Improved YOLO11 Multi-Scale Feature Fusion Method Based on Sonar Constraints The YOLO11 network achieves the detection of targets of different sizes through multi-scale feature fusion, but its original fusion method mainly relies on the visual features themselves, making it difficult to distinguish the degree of noise interference on features of different scales.

[0049] To fully utilize the expressive power of features at different scales on the target structure, while avoiding excessive influence of noise scale on the fusion result, this method introduces sonar information to dynamically constrain the fusion weights of multi-scale features.

[0050] Let the multi-scale visual feature set be :

[0051] The corresponding scale sonar structure credibility map for:

[0052] Calculate the scale stability index for each scale. :

[0053] in, For balance coefficient, For the first The height of the scale feature map, For the first The width of the scale feature map is used to suppress the interference of local unstable regions on stability estimation.

[0054] Further construct scale fusion weights :

[0055] in This is a temperature parameter used to control the smoothness of the weight distribution. This parameter is not the physical ambient temperature or medium temperature, but rather originates from the probability normalization control factor in statistical modeling.

[0056] To achieve precise spatial control, a spatial consistency factor is introduced in addition to scale weighting. :

[0057] in, The adjustment coefficient is used. Features at each scale are weighted and fused to obtain the fused features. , means as follows:

[0058] in, This represents the scale alignment mapping function.

[0059] By employing the above method, multi-scale features are simultaneously constrained by scale stability and spatial structure consistency during the fusion process, thereby obtaining robust fused feature representations.

[0060] 3. An Improved YOLO11 Detection Confidence Correction Method Based on Sonar Consistency In the YOLO11 network, the confidence level of the detection results is mainly determined by the visual classification probability and regression results, lacking verification of the reasonable existence of the target in real physical space. To improve the reliability of the detection results in real physical space, this method introduces sonar spatial consistency constraints in the target detection output stage to adaptively correct the detection confidence level.

[0061] Assuming the output is the first The bounding boxes of the detected candidate targets are Its class probability is The probability of targeting is Its initial detection confidence level is defined as :

[0062] Map the detection box region to the sonar structure credibility map Calculate the structural support within the frame :

[0063] Further calculations were performed to determine the consistency index between the detection frame boundary and the sonar structure edge. :

[0064] Construct a comprehensive consistency score:

[0065] in, These are the weighting coefficients. For the first Each detected candidate object is located within a bounding box region in the image space. This is the average response value of the sonar structure confidence map within the bounding box region, used to characterize the degree of structural support for the detected target in sonar space.

[0066] Mapping consistency scores to confidence correction factors :

[0067] Final detection confidence level It can be defined as:

[0068] in, This is the lower bound coefficient, used to avoid excessive confidence decay in extreme cases. This is the sonar consistency threshold, used to distinguish whether the detected target is adequately supported by the sonar structure. This is a confidence lower limit protection coefficient, used to avoid excessive suppression of detection confidence in extreme cases, and to avoid excessive decay of confidence in extreme cases.

[0069] Through the aforementioned confidence correction mechanism, the detection results can simultaneously satisfy the constraints of visual consistency and sonar spatial consistency, thereby improving the stability and security of detection decisions in engineering application scenarios.

[0070] Step 2: Image preprocessing: The images collected by the intelligent recognition camera 198 are preprocessed using a strategy that combines physical models and deep learning: (1) Intelligent restoration: The underwater images are dehazed, color corrected and style transferred using a dark channel prior improvement algorithm combined with a recurrent generative adversarial network, fundamentally eliminating the effects of underwater scattering and color shift; (2) Feature enhancement: The multi-scale Retinex algorithm and adaptive histogram equalization are used to significantly highlight the geometric edges and texture details of valves and small falling objects; (3) Multi-sensor pixel-level fusion: The depth map generated by the sonar point cloud is fused with the RGB image using RGB-D data, and the depth information is used to guide image segmentation, greatly improving the scene resolution capability in low light and turbid water environments.

[0071] Step 3: Target recognition and detection: Construct a recognition model based on a deep convolutional neural network with an attention mechanism. The construction and training process of the deep convolutional neural network is as follows: (1) Construct an enhanced dataset: Collect samples of multiple categories and expand the training set using data augmentation techniques (such as random occlusion and adversarial perturbation); (2) Refined annotation: Perform instance segmentation-level annotation on the data; (3) Network architecture upgrade: Select a lightweight CNN with integrated convolutional block attention module as the base network so that the model can automatically focus on key feature areas such as valve handles and bolts and suppress background noise; (4) Optimized training: Use the Focal Loss loss function to solve the sample imbalance problem and use backpropagation and AdamW optimizer to minimize the loss; (5) Model output: Obtain a multi-class high-precision recognition model with small sample generalization ability.

[0072] The automatic identification and positioning process is as follows: (1) Classification: accurately distinguish oil well valves, pipe fittings, tools and fallen objects; (2) Positioning: adopt vision-sonar tight coupling technology. The specific process is as follows: extract semantic feature points from visual images, obtain depth constraints from sonar point clouds, fuse the pose estimation of the two through nonlinear optimization algorithms, construct a dense three-dimensional point cloud map with semantic information, not only output target coordinates, but also realize semantic understanding of structured environment.

[0073] Step 4: Decision Making and Path Planning: Based on the identification and localization results, the operation processing module 19 executes high-order path planning. A dual-delay deep deterministic policy gradient algorithm based on deep reinforcement learning is adopted, combined with an artificial potential field-guided RRT. This trajectory planning and control system maps the water flow field, robotic arm dynamics constraints, and obstacle information to the state space. By maximizing the reward function (considering energy consumption, time, and safety), it directly generates a smooth, collision-free, and dynamically optimal trajectory in the continuous action space, possessing real-time obstacle avoidance capabilities for dynamic obstacles.

[0074] Step 5: Task Execution: Operation processing module 19 drives the robotic arm based on an impedance control and force / position hybrid control architecture. Position control ensures speed during free-space movement; during contact operations (such as valve turning, gripping, and handling), it seamlessly switches to impedance control mode. This mode endows the robotic arm's end effector with a "virtual spring" characteristic, allowing it to operate smoothly in accordance with the physical constraints of the valve, effectively avoiding equipment damage or installation jamming due to excessive rigidity.

[0075] Step 6: Feedback and Optimization: Display controller 11 achieves full-state monitoring through force / displacement sensors and visual servo. Active disturbance rejection control and nonlinear model predictive control are introduced. Closed-loop process: (1) Disturbance observation: The extended state observer in active disturbance rejection control estimates and compensates for total internal and external disturbances such as water flow impact and cable pulling in real time; (2) Predictive control: Nonlinear model predictive control predicts the future state based on the system model and solves the optimal control quantity to correct the trajectory error; (3) Visual servo: Image-based visual servo technology is adopted to directly map the image feature error into the robot arm motion command, forming a high-frequency response and high-precision dual closed-loop control system to ensure that the operation accuracy reaches the sub-millimeter level in complex water flow environment.

[0076] Through the above steps, intelligent identification, precise positioning, and stable operation of wellhead valves and falling objects can be achieved in complex underwater environments, ensuring the reliability and safety of operations.

[0077] The multi-functional remote-controlled underwater drilling tool 1 of this invention significantly improves operational efficiency, reduces reliance on professional divers, and effectively avoids potential risks and accidents during diving operations. Compared to traditional underwater robots, this tool has a simpler structure, is more intuitive to operate, and exhibits greater flexibility and economy in shallow water. It can be widely used in offshore oil and gas development, underwater infrastructure construction, underwater salvage, and emergency maintenance.

[0078] The multi-functional remote-controlled underwater wellhead tool 1 disclosed in this invention employs a remote operation method, avoiding the high risks associated with manual diving operations, especially in shallow waters, cold seas, or harsh sea conditions, significantly improving the safety and controllability of operations. Furthermore, it achieves multi-functional integrated operation: this invention integrates multiple functions such as valve operation, structural inspection, underwater target identification, lighting guidance, and falling object detection and retrieval into a single device, reducing the frequency of equipment switching and the difficulty of system coordination, greatly improving the efficiency of underwater operations.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-functional remote operation tool for underwater oil production trees, characterized in that, include: Display controller (11); A connecting component, the upper end of which is connected to the display controller (11), and the connecting component is also connected to the survey vessel; An operating component is connected to the lower end of the connecting component and communicates with the display controller (11). The operating component includes a thruster (13) and an operating processing module (19). A robotic arm is provided on the operating processing module (19). The thruster (13) is used to control the operating processing module (19) to reach the target position. The robotic arm is used to open and close the subsea tree frame (2) and the subsea tree valve.

2. The multi-functional remote operation tool for underwater production trees according to claim 1, characterized in that, The robotic arm includes a rotary electromagnetic arm (199) and a clamping robotic arm (195). The rotary electromagnetic arm (199) is connected to the operation processing module base plate (191) of the operation processing module (16) via an intelligent recognition camera (198). The clamping robotic arm (195) is mounted on a frame connected to the operation processing module base plate (191).

3. The multi-functional remote operation tool for underwater production trees according to claim 2, characterized in that, The robotic arm also includes several slide rail robotic arms (194), which slide the operation processing module base plate (191) together with the frame.

4. The multi-functional remote operation tool for underwater production trees according to claim 3, characterized in that, The frame is a closed slide rail frame formed by connecting several slide rails (192) end to end. Several slide rail robotic arms (194) are arranged in a one-to-one correspondence with several slide rails (192), and the connector of the slide rail robotic arm (194) is slidably connected to the slide rail (192).

5. The multi-functional remote operation tool for underwater production trees according to claim 4, characterized in that, Each of the slide rails (192) is equipped with a long strip-shaped first lighting lamp (193).

6. The multi-functional remote operation tool for underwater production trees according to claim 5, characterized in that, The operation processing module base plate (191) is also equipped with a sonar module (196) and a second lighting lamp (197).

7. The multi-functional remote operation tool for underwater production trees according to claim 6, characterized in that, The rotating mechanical electromagnetic arm (199) is also equipped with a metal detector (1991) and a sleeve (1992).

8. The multi-functional remote operation tool for underwater production trees according to claim 1, characterized in that, The connecting components include a telescopic column (12), a flexible hose column (16), a pipe column (17), and a bend (18). The pipe column (17) includes a first pipe column (171) and a second pipe column (172). The display controller (11) is connected to the upper end of the telescopic column (12). The lower end of the telescopic column (12) is connected to the upper end of the first pipe column (171). The upper and lower ends of the flexible hose column (16) are connected to the lower end of the first pipe column (171) and the upper end of the second pipe column (172) respectively through flanges (15). The vertical end of the bend (18) is connected to the lower end of the second pipe column (17). The horizontal end of the bend (18) is connected to the operation processing module (19).

9. A method for intelligent identification and control of a multi-functional remote operation tool for underwater production trees according to claim 7, characterized in that, Includes the following steps: Image acquisition: The intelligent recognition camera (198) acquires underwater operation scene images, while the sonar module (196) acquires sonar echo information. The timestamp synchronization technology is used to ensure the spatiotemporal alignment of the underwater operation scene images and the spatial point cloud information. Image preprocessing: The images captured by the intelligent recognition camera (198) are preprocessed to obtain preprocessed images; Target recognition and detection: A multi-class recognition model is obtained by training the preprocessed image based on a deep convolutional neural network. The multi-class recognition model is used to realize the automatic recognition and location of valves and fallen objects. Decision-making and path planning: Based on the identification and positioning results, the operation processing module (19) performs path planning; Task execution: The operation processing module (19) drives the clamping robotic arm (195), the rotating robotic arm (199) and the pipe sleeve (1992) according to the trajectory instructions to complete operations including valve rotation, gripping and handling and component installation; Feedback and optimization: The display controller (11) achieves full-state monitoring through force / displacement sensors and visual servo, ensuring that the operation accuracy reaches the sub-millimeter level in complex water flow environments. The intelligent recognition and control method according to claim 9 is characterized in that the image acquisition is based on an improved YOLO11 network, and the acquisition process is as follows: Acquire visual image information and sonar echo information of underwater operation scenarios; An improved YOLO11 network is used to extract features from visual image information and generate multi-scale visual features. Sonar echo information is mapped to spatial regions corresponding to multi-scale visual features, and the credibility of the multi-scale visual features is evaluated to obtain sonar evaluation results. During the feature propagation process of the improved YOLO11 network, the contribution of multi-scale visual features is adjusted according to the sonar evaluation results. In the multi-scale visual feature fusion stage of the improved YOLO11 network, sonar constraints are introduced to dynamically adjust the fusion weights of multi-scale visual features to obtain the fused features. Target detection is performed based on the fused features, and the initial detection results are output. By combining the spatial consistency information provided by sonar, the confidence level of the initial detection results is corrected to obtain the final underwater target detection results.

10. The intelligent identification and control method according to claim 9, characterized in that, The YOLO11 network has the following improvements: Structural encoding is performed on the sonar echo information to obtain a sonar structure reliability map. The credibility map of the sonar structure The visual features are mapped to the corresponding visual feature scale through scale mapping and spatial alignment. Then, a visual region confidence weight map is constructed. Based on the confidence weight map, the visual features are spatially controlled and the channel dimension is adaptively controlled, and finally, a reliable visual feature is output. A multi-scale visual feature set is established, and then the sonar structure confidence map at the corresponding scale is determined. The scale stability index is calculated for each scale, and scale fusion weights are further constructed. A spatial consistency factor is introduced on the basis of the scale fusion weights, and then the visual features at each scale are weighted and fused to obtain the fused features. Assuming the initial detection confidence of a certain candidate target is output, the detection box region is mapped to the sonar structure confidence map. The system calculates the structural support within the detection box and the consistency index between the detection box boundary and the sonar structure edge, constructs a comprehensive consistency score, maps the consistency score to a confidence correction factor, and finally determines the detection confidence.

11. The intelligent identification and control method according to claim 9, characterized in that, The specific process of image preprocessing is as follows: (1) Intelligent restoration: The underwater image is dehazed, color corrected and style transferred by using the dark channel prior improvement algorithm combined with the recurrent generative adversarial network, which fundamentally eliminates the influence of underwater scattering and color deviation; (2) Feature enhancement: The multi-scale Retinex algorithm and adaptive histogram equalization are used to significantly highlight the geometric edges and texture details of valves and small falling objects; (3) Multi-sensor pixel-level fusion: The depth map generated by the sonar point cloud is fused with the RGB image in RGB-D data, and the depth information is used to guide image segmentation, which greatly improves the scene resolution capability in low light and turbid water environments; The specific process of target recognition and detection is as follows: The construction and training process based on deep convolutional neural networks is as follows: (1) Constructing an enhanced dataset: Collect samples of multiple categories and expand the training set using data augmentation techniques; (2) Refined annotation: Perform instance segmentation-level annotation on the data; (3) Network architecture upgrade: Select a lightweight CNN with integrated convolutional block attention module as the base network, so that the model can automatically focus on key feature areas including valve handles and bolts, and suppress background noise; (4) Optimized training: Use the Focal Loss loss function to solve the sample imbalance problem, and use backpropagation and AdamW optimizer to minimize the loss; (5) Model output: Obtain a multi-class high-precision recognition model with small sample generalization ability; The specific process of decision-making and path planning is as follows: The dual-delay deep deterministic policy gradient algorithm based on deep reinforcement learning is adopted, combined with the RRT guided by the artificial potential field method. The trajectory planning and control system maps the water flow field, the dynamic constraints of the robotic arm and the obstacle information to the state space. By maximizing the reward function, a smooth, collision-free and dynamically consistent optimal trajectory is directly generated in the continuous action space, which has the ability to avoid obstacles in real time. The specific process of feedback and optimization is as follows: Introducing active disturbance rejection control and nonlinear model predictive control, closed-loop process: (1) Disturbance observation, the extended state observer in active disturbance rejection control estimates and compensates for the total internal and external disturbances such as water flow impact and cable pulling in real time; (2) Predictive control, nonlinear model predictive control predicts the future state based on the system model, solves the optimal control quantity to correct the trajectory error; (3) Visual servo: adopting image-based visual servo technology, the image feature error is directly mapped to the robotic arm motion command, forming a high-frequency response and high-precision dual closed-loop control system, ensuring that the operation accuracy reaches the sub-millimeter level in complex water flow environment.

Images