A soft yoke mooring system singular posture recognition system and method
By using deep learning methods to monitor and identify the attitude of a soft rigid boom mooring system in real time, the problem of excessive structural stress caused by singular attitudes is solved, and accurate identification and early warning of the attitude of the soft rigid boom mooring system are achieved, ensuring platform safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NINGBO INST OF DALIAN UNIV OF TECH
- Filing Date
- 2022-12-26
- Publication Date
- 2026-07-24
AI Technical Summary
During long-term service, soft rigid arm mooring systems may exhibit unusual attitudes, causing the mooring structure to deviate from its equilibrium position. This can prevent the release of rotational degrees of freedom at the nodes, leading to excessive stress on local structures and connecting nodes, resulting in loss of mooring positioning function and potentially causing platform damage and personnel casualties.
A flexible-rigid system is adopted, which includes a video monitoring module, an image preprocessing module, a key point detection module, an attitude classification and attitude prediction module, and an attitude prediction module. Through video monitoring, image preprocessing, key point detection, and attitude recognition, and through multi-angle video analysis, and by using deep learning methods, an attitude recognition model is established to monitor and judge the attitude changes of the flexible-rigid arm mooring system in real time and provide timely early warning and forecast.
It enables real-time monitoring and accurate identification of the attitude of a soft rigid arm mooring system, providing reference and guidance for platform operators, avoiding the assumptions and wasted computational resources of traditional methods, and improving the accuracy of identification and computational efficiency.
Smart Images
Figure CN115830123B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine engineering technology, specifically relating to a singular attitude recognition system and method for a soft rigid arm mooring system. Background Technology
[0002] Soft rigid arm mooring systems are a common type of equipment for shallow-sea oil and gas development, widely used in the Bohai Sea region of my country, and are an important mooring method to ensure the safe operation of offshore platforms. A soft rigid arm mooring system consists of mooring supports, mooring legs, a mooring arm, and a single-point turret. These are connected by hinged joints to release rotational degrees of freedom, thus enabling the platform to have a good weather vane effect under the combined effects of wind, waves, and currents. my country's first offshore nuclear power platform uses a soft rigid arm mooring system as its offshore positioning method, providing energy for offshore oil extraction, remote islands, and seawater desalination.
[0003] However, during long-term service, soft rigid arm mooring systems may exhibit unusual attitudes, causing the mooring structure to deviate from its equilibrium position and the rotational degrees of freedom of the nodes to be unable to be released. This results in excessive stress on the local structure and connecting nodes, leading to the loss of mooring positioning function and ultimately causing platform damage and personnel casualties. Summary of the Invention
[0004] The purpose of this invention is to meet the above requirements by providing a singular attitude recognition system and method for a soft rigid boom mooring system; real-time monitoring and identification of attitude changes in the soft rigid boom mooring system and timely early warning and forecasting, providing reference and guidance for platform operators.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] A singular attitude recognition system for a soft rigid arm mooring system, comprising a video monitoring module, an image preprocessing module, a key point detection module, a key point connection module, an attitude classification module, and an attitude prediction module;
[0007] The video monitoring module synchronously collects video information from different angles of the soft rigid arm mooring system. The arrangement angle must ensure that the subsequent algorithm can reconstruct the spatial configuration of the system.
[0008] The image preprocessing module is used to preprocess the information obtained by the monitoring module;
[0009] The key point detection module is used to detect key nodes of the soft rigid arm mooring system using a deep neural network.
[0010] The key point connection module is used to construct the spatial configuration of the soft rigid arm mooring system.
[0011] The attitude classification module is used to classify the spatial configuration of the soft rigid arm mooring system and determine whether it is singular;
[0012] The attitude prediction module is used to predict short-term attitude changes of the soft rigid arm mooring system.
[0013] The video monitoring module includes camera positions at multiple angles. The camera positions are set to meet the different viewpoints of the soft rigid arm mooring system, while maintaining time synchronization and consistent shooting frame rate among the camera positions.
[0014] The key point detection module includes the following detection nodes: upper node of the left mooring leg, lower node of the left mooring leg, upper node of the right mooring leg, lower node of the right mooring leg, and single-point turret node.
[0015] A method for singular attitude recognition of a soft rigid arm mooring system includes the following steps:
[0016] S1 utilizes field equipment to acquire multi-source video streams of the soft rigid arm mooring system;
[0017] S2 performs noise reduction, frame extraction, and cropping on the video described in S1 to obtain a single frame of the target image.
[0018] S3 inputs the image described in S2 into the detection network for detection and outputs key point information of the soft rigid arm mooring system. The detection network is trained in advance using historical mooring system images and on-site images.
[0019] S4 calculates the spatial line configuration of the YOKE rigid arm, mooring leg and mooring support based on the information obtained from S3;
[0020] S5 classifies the attitude of the soft rigid arm mooring system based on the spatial line configuration obtained in S4. Using the Softmax classifier based on long-term monitoring database samples and the relative angles of each structural component, it determines whether the current mooring system attitude is a singular attitude.
[0021] S6 inputs the key point information of the mooring system monitored over a period of time into the prediction network. This prediction network is pre-trained using historical monitoring data. Using the key point information of the mooring system over a previous period as input, it outputs the short-term future changes in the positions of the key points of the mooring system, predicting whether the mooring system attitude is about to exhibit an unusual attitude in the short term.
[0022] In the above process, the input to the detection network in step S3 is a set of images taken at the same time. The detection targets include the upper node of the left mooring leg, the lower node of the left mooring leg, the upper node of the right mooring leg, the lower node of the right mooring leg, and the single-point turret node. The output information includes the detection box label, the coordinates of the upper left corner, and the width and height of the detection box for each detection target in each image in the set of images. After each detection, the detection results can be analyzed. Detection results that do not meet the IOU requirements are corrected and supplemented as the training set for the detection network to improve the accuracy of the detection model.
[0023] In step S3, the coordinates of each node required to calculate the spatial lines are the coordinates of the detection box center, i.e.
[0024] in To calculate the node coordinates required for spatial lines, and These are the coordinates of the top-left corner of the detection box output by the detection network, as well as the width and height of the detection box;
[0025] The relative positions of the nodes are calculated based on the coordinates of each node on images from different angles, that is, the two-dimensional mapping information of each node of the soft rigid arm mooring system at different angles. Taking the single-point turret node as the center, the three-dimensional spatial coordinates of each node on the system are restored.
[0026] This method, employing the above technical solutions, offers the following advantages: The system and method for identifying singular attitudes of a flexible rigid arm mooring system includes a video monitoring module, an image preprocessing module, a key point detection module, a key point connection module, an attitude classification module, and an attitude prediction module. This method acquires multi-source video streams of the flexible rigid arm mooring system; preprocesses the flexible rigid arm mooring system in the video, labeling the structures to be detected with bounding boxes; constructs an image recognition neural network algorithm to track and identify the flexible rigid arm mooring system; calculates the spatial line configuration of the YOKE rigid arm, mooring legs, and mooring support based on the recognition results; classifies the constructed spatial line configuration of the flexible rigid arm mooring system using a deep learning algorithm; and identifies singular attitudes of the flexible rigid arm mooring system based on long-term monitoring database samples.
[0027] This recognition system and method employs a deep learning-based approach to establish an attitude recognition model. The model is trained using system motion response data, and the trained model is then used to handle the complex problem of singular attitude recognition. This avoids the drawbacks of traditional numerical methods, such as the introduction of numerous assumptions and implementation difficulties. Simultaneously, it can accurately identify singular attitudes in soft rigid boom mooring systems. The deep learning model, combined with feature processing, exhibits strong generalization ability, saves computational resources, and improves computational efficiency and attitude recognition accuracy. This recognition system and method are well-suited to real-world field conditions and are applicable to attitude recognition problems in soft rigid boom mooring systems.
[0028] This invention can intelligently track the motion attitude of a soft rigid arm mooring system on an offshore platform, thereby extracting the spatial configuration of the soft rigid arm mooring system's attitude in real time. Furthermore, it can identify singular attitudes of the arm mooring system and provide real-time early warning and forecasting through classification algorithms, providing reference and guidance for platform operators. Attached Figure Description
[0029] Figure 1 This is a module included in a singular attitude recognition system for a soft rigid arm mooring system.
[0030] Figure 2 This is a schematic diagram of each node of the soft rigid arm.
[0031] Figure 3 This is a flowchart of a singular attitude recognition method for a soft rigid arm mooring system.
[0032] Figure 4 This is a specific implementation diagram of a singular attitude recognition method for a soft rigid arm mooring system.
[0033] Figure 5 This is a schematic diagram of a singular attitude recognition system for a soft rigid arm mooring system.
[0034] Figure 6 This example demonstrates how to detect loss and accuracy changes during network pre-training. Detailed Implementation
[0035] The present invention will now be described in detail with reference to the accompanying drawings. However, it should be understood that the drawings are provided only to better understand the present invention and should not be construed as limiting the present invention.
[0036] Example 1
[0037] Figure 1 and Figure 2 The image illustrates a singular attitude recognition system for a soft rigid arm mooring system, comprising a video monitoring module, an image preprocessing module, a key point detection module, a key point connection module, an attitude classification module, and an attitude prediction module.
[0038] The video monitoring module is used to acquire video of the on-site mooring system. The video monitoring module includes camera positions at multiple angles, and the camera positions are set to meet the acquisition of different perspectives such as the front view, left view, and top view of the soft rigid arm mooring system. At the same time, the time synchronization and shooting frame rate are consistent between the camera positions.
[0039] The image preprocessing module is used to perform frame extraction and cropping on video information, which serves as the dataset for subsequent detection.
[0040] The keypoint detection module is used to input images into a trained detection network, which outputs bounding boxes for keypoints. The keypoint detection module's detection nodes include nodes for the upper left mooring leg, lower left mooring leg, upper right mooring leg, lower right mooring leg, and a single-point turret node. The node distribution is as follows: Figure 2 As shown.
[0041] The key point connection module calculates the spatial configuration of the mooring system based on key point information from multiple angles.
[0042] The attitude classification module is used to identify whether the attitude of the mooring system is singular;
[0043] The attitude prediction module uses a prediction network to predict short-term attitude changes in the mooring system.
[0044] Example 2
[0045] See Figure 3 , Figure 4 and Figure 5 This method for identifying singular attitudes in a soft rigid arm mooring system includes the following steps:
[0046] S1 utilizes field equipment to acquire multi-source video streams from the soft rigid arm mooring system. In this embodiment, a numerical model is used to simulate the field mooring system to apply loads under different operating conditions to verify the recognition effect in multiple scenarios. In this embodiment, video information from the left, right, and directly above is used.
[0047] S2 performs frame extraction and cropping on the video described in S1 to obtain a single-frame target image; in this embodiment, two frames are extracted per second from the video and used as input to the detection network.
[0048] S3 inputs the image from S2 into the detection network for detection and outputs key point information of the soft rigid arm mooring system. Each detection result can be supplemented into the training set for retraining after human judgment to improve the recognition accuracy.
[0049] In this embodiment, the detected targets include the upper node of the left mooring leg, the lower node of the left mooring leg, the upper node of the right mooring leg, the lower node of the right mooring leg, and the single-point turret node. The output information includes the detection box label, the upper left corner coordinates, and the width and height of the detection box for each detected target in a set of images.
[0050] This example uses the MASK-RCNN network. The network process is as follows: A 1024*1024 pixel image is input into the backbone feature network to obtain a feature map. Each grid on the feature map contains three prior boxes. The backbone feature extraction network uses ResNet101, and the feature layers contain feature extraction results with two, three, four, and five compressions (256*256*256, 128*128*256, 64*64*256, 32x*32x*256, and 16x*16*256). Based on the results from the backbone feature extraction network, a feature pyramid (FPN) is constructed. The feature map is then input into the Region Proposal Network (RPN). The RPN adjusts the prior boxes in the feature map and initially determines whether the boxes contain the target to be detected. Boxes containing the target are considered proposal boxes. A classification network is used to classify and adjust the proposal boxes to obtain the final predicted box positions. Non-maximum suppression is applied based on the scores to avoid duplicate detection. Figure 6As shown, the detection network was pre-trained using 500 sets of image data. The training set included real images of the mooring system and simulated effect images, and was trained for 50 epochs. The batch size was set to 10, and the final accuracy was approximately 97%.
[0051] In this embodiment, the single-point turret node of the soft rigid arm is taken as the origin of the three-dimensional coordinate system, and the spatial coordinates of the remaining points are calculated using the following formula based on the coordinate information in the three directions.
[0052]
[0053] In the formula, u and v are the horizontal and vertical coordinates in the two-dimensional plane of the image; x, y, z are the three-dimensional spatial coordinates; R is the rotation amount; t is the offset amount; and A is the camera intrinsic parameter matrix.
[0054] S4 calculates the spatial line configuration of the YOKE rigid arm, mooring leg, and mooring support based on the information obtained from S3.
[0055] S5 classifies the attitude of the flexible rigid arm mooring system based on the spatial line configuration obtained in S4, and determines whether the current mooring system attitude is singular based on long-term monitoring database samples. In this embodiment, the judgment condition is based on the angle between each flexible rigid arm deviating from the equilibrium position by 20°. A Softmax classifier is trained by combining the combination of relative angles of each structural component under different mooring system attitudes in historical monitoring data to determine whether the mooring system is singular.
[0056] S6 inputs the key point information of the mooring system monitored over a period of time into the prediction network to predict the attitude changes of the mooring system in the short term and determine whether a singular attitude is about to occur. In this embodiment, an LSTM network is used. The LSTM network introduces a forget gate on the basis of the RNN network, which can solve the gradient vanishing and gradient exploding problems to a certain extent. In this embodiment, historical monitoring data is used for training, and the coordinates of each node of the mooring system in the previous hour are used as input to predict the movement of each node in half an hour. The results are then input into the classifier to determine whether a singular attitude has occurred. The prediction accuracy is as high as 100%, proving that the prediction method is accurate and effective.
Claims
1. A method for identifying singular attitudes in a soft rigid arm mooring system, characterized in that, The system used in this method includes a video monitoring module, an image preprocessing module, a key point detection module, a key point connection module, a pose classification module, and a pose prediction module. The video monitoring module synchronously collects video information from different angles of the soft rigid arm mooring system. The arrangement angle must ensure that the subsequent algorithm can reconstruct the spatial configuration of the system. The image preprocessing module is used to preprocess the information obtained by the monitoring module; The key point detection module is used to detect key nodes of the soft rigid arm mooring system using a deep neural network. The key point connection module is used to construct the spatial configuration of the soft rigid arm mooring system. The attitude classification module is used to classify the spatial configuration of the soft rigid arm mooring system and determine whether it is singular; The attitude prediction module is used to predict short-term attitude changes of the soft rigid arm mooring system. The video monitoring module includes camera positions at multiple angles. The camera positions are set to meet the different viewpoints of the soft rigid arm mooring system, while maintaining time synchronization and consistent shooting frame rate between the camera positions. The key point detection module includes the following detection nodes: upper node of the left mooring leg, lower node of the left mooring leg, upper node of the right mooring leg, lower node of the right mooring leg, and single-point turret node. The method includes the following steps: S1. Acquire multi-source video streams of the soft rigid arm mooring system using field equipment; S2. Perform frame extraction and cropping on the video to obtain a single frame of the target image; S3. Input the image into the detection network for detection. This detection network is pre-trained using historical mooring system images and on-site images, and outputs the relative position information of key points in the soft rigid arm mooring system, while defining the three-dimensional spatial coordinate information of each key point; specifically including: S3.1 The input to the detection network is a set of images taken at the same time. The detection targets include the upper node of the left mooring leg, the lower node of the left mooring leg, the upper node of the right mooring leg, the lower node of the right mooring leg, and the single-point turret node. The output results include the position and size of the detection box of each detection target in each image of the set of images. S3.
2. Obtain the coordinates of each node at different angles based on the detection network, calculate the relative position of the nodes, that is, the two-dimensional mapping information of each node of the soft rigid arm mooring system at different angles. The coordinates of each node required to calculate the spatial lines are the coordinates of the detection box center, i.e. in To calculate the node coordinates required for spatial lines, and These are the coordinates of the top-left corner of the detection box output by the detection network, as well as the width and height of the detection box; S3.
3. Then, taking the single-point turret node as the center, restore the three-dimensional spatial coordinates of each node in the system; Using the single-point turret node of the soft rigid arm as the origin of the three-dimensional coordinate system, the spatial coordinates of the remaining points are calculated using the following formula based on the coordinate information in the three directions. In the formula, u and v are the horizontal and vertical coordinates in the two-dimensional plane of the image; x, y, z are the three-dimensional spatial coordinates; R is the rotation amount; t is the offset amount; and A is the camera intrinsic parameter matrix. S4. Based on the obtained three-dimensional spatial coordinate information of the nodes, calculate the spatial line configuration of the YOKE rigid arm, mooring leg and mooring support according to the topological connection method of the mooring system; S5. Classify the attitude of the soft rigid arm mooring system based on the obtained spatial line configuration. Use the Softmax classifier based on statistical samples from the long-term monitoring database and combine the relative angles of each structural component to determine whether the current mooring system attitude is a singular attitude. S6. Input the key point information of the mooring system monitored over a period of time into the prediction network. The prediction network is trained in advance using historical monitoring data. It takes the key point information of the mooring system over a period of time as input and outputs the changes in the position of the key points of the mooring system in the short term, predicting whether the mooring system attitude will soon exhibit a singular attitude in the short term. The attitude of the soft rigid boom mooring system is determined based on the relative angles of each structural component. If the relative angle exceeds the included angle threshold, the attitude is determined to be singular; if the relative angle does not exceed the included angle threshold, the attitude is determined to be non-singular. Incorrect classification results are corrected and added to the data monitoring database.
2. The identification method according to claim 1, characterized in that: The detection network in step S3 and the prediction network in step S6 adopt a 3-4 layer neural network algorithm.
3. The identification method according to claim 2, characterized in that: In step S3, the detection network uses the MASK-RCNN network, and in step S6, the prediction network uses the LSTM network.