AUV submarine petroleum pipeline inspection risk identification system based on deep learning

Abstract

The invention discloses an AUV submarine petroleum pipeline inspection risk identification system based on deep learning, and relates to the technical field of ocean engineering, and the system comprises an inspection risk management center, and the inspection risk management center is in communication connection with the following modules: an AUV autonomous inspection module, which is used for deploying an AUV to carry out an inspection task of a submarine petroleum pipeline, and the control instruction of the AUV is dynamically adjusted according to the seabed environment and the pipeline state, and the inspection path and speed are optimized. The abnormal pipeline identification model is constructed by combining the deep learning algorithm, the pipeline image data can be efficiently classified, the submarine petroleum pipeline area in the abnormal state can be identified, the identified abnormal state area is further analyzed, the pipeline risk trend index is calculated, and the potential risk level of the pipeline is evaluated. Potential safety hazards can be found and treated in time, accidents are prevented, and safe operation of the submarine petroleum pipeline is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of marine engineering technology, and in particular to an AUV submarine oil pipeline inspection risk identification system based on deep learning. Background Art

[0002] With the gradual reduction of terrestrial oil resources, offshore oil has become an important source of energy supplement. The expansion of offshore oil development has led to an increasing number and length of submarine oil pipelines. These pipelines not only undertake the task of transporting oil and natural gas, but also involve auxiliary production processes such as water injection and gas injection. The expansion of the pipeline scale increases the probability of its failure, so more efficient and comprehensive inspection methods are needed to ensure its safe operation.

[0003] At present, traditional submarine oil pipeline inspection often uses ROV (remotely operated vehicle). This ROV is connected to the surface mother ship by a cable, and its operation relies on human operation. It is also limited by its range and has limitations. This may lead to the risk of not being able to detect pipelines in time, increasing the possibility of pipeline failure. Therefore, how to use AUV equipped with high-definition cameras to inspect submarine pipelines, and combine image classification and pattern recognition technology to classify the images of submarine pipelines, identify areas with significant differences from the normal pipeline status, further mine the potential information in the images, and identify risk points, is the problem to be solved by the present invention. To this end, a deep learning-based AUV submarine oil pipeline inspection risk identification system is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide an AUV submarine oil pipeline inspection risk identification system based on deep learning to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A deep learning-based AUV submarine oil pipeline inspection risk identification system includes an inspection risk management center, which is communicatively connected to the following modules:

[0007] The AUV autonomous inspection module is used to deploy AUVs to inspect submarine oil pipelines. It dynamically adjusts AUV control instructions based on the submarine environment and pipeline status, optimizing inspection paths and speeds. Upon completion of the inspection mission, the AUV automatically surfaces and returns to its home port, achieving an integrated "deployment-inspection-recovery" operation.

[0008] The image acquisition and processing module is used to collect and pre-process the image data of the submarine oil pipeline during the inspection process of the AUV, perform feature analysis, extract pipeline defect features, and form a defect feature sequence table;

[0009] The image recognition and classification module is used to build an abnormal pipeline identification model by combining deep learning algorithms, classify pre-processed pipeline image data, and identify areas of submarine oil pipelines in abnormal conditions;

[0010] The risk assessment module is used to further analyze the identified abnormal status areas, calculate the pipeline risk trend index, and evaluate the potential risk level of the pipeline.

[0011] A further improvement of the technical solution of the present invention is that: the AUV autonomous inspection module includes an AUV deployment and submersion unit and a pipeline identification and tracking unit;

[0012] The AUV deployment and diving unit is used to deploy the AUV to a designated location through a surface deployment device and control it to autonomously dive to the seabed area, thereby realizing the automated deployment and diving of the AUV, reducing manual intervention and improving operation efficiency.

[0013] The pipeline identification and tracking unit is used to use the multi-angle high-definition camera on the AUV to identify submarine oil pipelines, conduct inspections along the pipelines, and adopt a reinforcement learning algorithm to dynamically adjust the AUV's control instructions according to the submarine environment and pipeline status to optimize the inspection path and speed.

[0014] A further improvement of the technical solution of the present invention is that the AUV deployment submersible unit specifically includes:

[0015] Before deployment, the operator conducts a comprehensive inspection of the AUV to ensure that all its systems are functioning properly and that it has sufficient power. The operator then presets the dive path and target area based on mission requirements and the high-precision positioning system. The operator then installs the AUV on the surface deployment device and smoothly drops it to the designated surface location.

[0016] After receiving the dive command, the AUV's autonomous control system starts, controlling the buoyancy control system to gradually discharge the gas in the buoyancy tank, increasing its own weight and causing it to slowly dive. At the same time, the AUV's propulsion system adjusts the propulsion direction and speed based on the preset dive path and information fed back by the attitude sensor. When the AUV approaches the target area on the seabed, its altitude sensor and terrain matching system further accurately control the dive depth and position, allowing it to accurately reach the predetermined seabed area.

[0017] After reaching the seabed, the AUV's positioning system communicates with the seabed positioning base station, and uses acoustic positioning technology to determine its own position on the seabed and compare it with the preset target position. If there is a deviation, the AUV automatically adjusts its position until it completely reaches the target area. At this time, the AUV enters standby mode, and its high-definition camera and sensors begin to warm up and initialize, ready to carry out pipeline inspection tasks.

[0018] A further improvement of the technical solution of the present invention is that the pipeline identification and tracking unit specifically includes:

[0019] After the AUV dives to the target area on the seabed, its multi-angle high-definition camera begins working, performing a 360-degree full-scale scan to capture images of the seabed environment. Using image recognition technology, it compares and analyzes the characteristics of the submarine oil pipeline, identifying its location and direction. At this point, the AUV's positioning system, combined with the initial position information of the pipeline, determines its relative position to the pipeline. Based on the preset inspection task requirements and the initial distribution of the pipeline, it plans an initial inspection route that covers the key parts of the pipeline.

[0020] During the inspection along the initial path, the AUV continuously senses information about the seabed environment, including water flow speed and direction, seabed topography, and the real-time status of the pipeline. A reinforcement learning algorithm is then used to analyze this information. Based on the current environment and pipeline status, the AUV's control instructions are evaluated to dynamically adjust and optimize the inspection path and speed.

[0021] The process of dynamically adjusting the control instructions of the AUV is as follows:

[0022] As the AUV conducts its inspection along the initial inspection route, it continuously collects information about the seabed environment and pipeline status through its various sensors, including water velocity and direction, acquired through flow sensors, seabed topography, scanned through a sonar system, and the real-time status of the pipeline, captured and imaged by a high-definition camera, which is then recognized and transmitted to the AUV's central processing unit;

[0023] After receiving information about the seabed environment and pipeline status collected from sensors, the system runs a reinforcement learning algorithm to evaluate the potential impact of different control instructions on inspection results based on the current water speed, direction, and seabed topography. By simulating multiple control strategies, the system analyzes the safety, efficiency, and stability indicators under each control strategy, and then calculates the expected return under each strategy. Using historical data and a preset reward mechanism, the system learns control instructions that achieve optimal inspection results in complex environments while avoiding collisions and pipeline deviations to adapt to the ever-changing seabed environment.

[0024] Based on the analysis results of the reinforcement learning algorithm, the AUV control instructions are dynamically adjusted, including the speed and direction of the thrusters, to optimize the inspection path and speed. Through dynamic adjustment, efficient inspections can be achieved in complex seabed environments while ensuring the safety and stability of the inspection process.

[0025] A further improvement of the technical solution of the present invention is that: the image acquisition and processing module includes an image acquisition unit and an image feature extraction unit;

[0026] The image acquisition unit is used to collect image data of the submarine oil pipeline through the high-definition camera carried by the AUV, and pre-process the image data, including denoising, contrast enhancement and distortion correction operations;

[0027] The image feature extraction unit is used to perform feature analysis on the pre-processed image data, extract pipeline defect features related to the health status of the submarine oil pipeline, and integrate them to obtain a defect feature sequence table.

[0028] A further improvement of the technical solution of the present invention is that the image acquisition unit specifically includes:

[0029] After the AUV arrives at the submarine oil pipeline inspection area, it conducts a self-inspection of the high-definition camera on board to ensure that all parameters are normal, the lens is unobstructed and there are no faults. The brightness of the camera's strong light source is automatically adjusted according to the submarine ambient light conditions. When the strong light source is turned on, it provides sufficient illumination for the camera to cope with the low-light environment on the submarine. Then, the high-definition camera captures all-round and multi-angle images of the submarine oil pipeline at the preset shooting frequency and angle, and records the shooting location and time information in real time.

[0030] The collected image data is preprocessed, including denoising, contrast enhancement and distortion correction operations, and then the preprocessed image data is stored in the AUV's storage system.

[0031] A further improvement of the technical solution of the present invention is that the image feature extraction unit specifically includes:

[0032] Extract the pre-processed image data and perform image segmentation using an edge detection algorithm to separate the relevant areas of the submarine oil pipeline. Combined with the basic submarine oil pipeline data built into the system, the boundaries of the submarine oil pipeline are further located.

[0033] Based on the relevant areas of the submarine oil pipeline segmented from the image, the geometric and shape features of the submarine oil pipeline are analyzed, and defect features related to the health status of the submarine oil pipeline are extracted, including pipeline diameter, pipeline curvature, crack length, crack width, hole area, hole diameter, corrosion depth, and corrosion area ratio;

[0034] The extracted defect features are integrated, and according to the inspection requirements of submarine oil pipelines, the baseline value of each defect feature is determined, that is, the maximum allowable value of each defect feature, to form a defect feature sequence table. For each defect area, the extracted geometric features and shape features are summarized and their location information on the pipeline is marked.

[0035] A further improvement of the technical solution of the present invention is that the image recognition and classification module specifically includes:

[0036] Collect a large amount of labeled submarine oil pipeline image data, including images of normal conditions as well as images of various abnormal conditions such as cracks, corrosion, and deformation. Extract pipeline defect features related to the health status of submarine oil pipelines from them to form a comprehensive dataset, which is then divided into a training set and a validation set.

[0037] A convolutional neural network (CNN) was selected as the underlying deep learning algorithm architecture to construct an abnormal pipeline identification model. Based on the characteristics of submarine oil pipeline images, the CNN model was adjusted. Images from the training set were fed into the model, and the model parameters were continuously adjusted using a backpropagation algorithm and optimizer to minimize the error between the predicted results and the true labels. During training, the model performance was monitored in real time using a validation set. After multiple rounds of iterations, the model achieved stable performance on the validation set and met the requirements. Finally, the abnormal pipeline identification model was constructed.

[0038] The trained abnormal pipeline recognition model is deployed and the pipeline image data to be inspected is input into the abnormal pipeline recognition model. The model then identifies abnormal pipeline defect features based on the baseline values ​​of each pipeline defect feature. Based on these abnormal pipeline defect features and the model's probability prediction of the image belonging to different state categories, the model determines the image's state category, i.e., normal or abnormal. For images judged to be in an abnormal state, the specific location and range of the abnormal area on the pipeline are further determined, ultimately identifying the abnormal submarine oil pipeline area.

[0039] A further improvement of the technical solution of the present invention is that the risk assessment module specifically includes:

[0040] Receive the abnormal state area information output by the image recognition and classification module, perform quantitative analysis on the pipeline defect characteristics in the abnormal state area, compare and analyze each pipeline defect characteristic with the pre-set pipeline defect characteristic benchmark value, and determine that the pipeline defect characteristic is a risk factor if the actual value of a pipeline defect characteristic exceeds or reaches its benchmark value;

[0041] After determining the risk factors, the degree of deviation of each risk factor from its corresponding benchmark value is analyzed. Combined with the pipeline's design specifications, historical operating data, and safety requirements, different weights are assigned to each risk factor to reflect its contribution to the overall pipeline risk. The quantitative value of each risk factor is then combined with its corresponding weight to calculate the pipeline risk trend index, which reflects the changing trend of the pipeline's potential risk.

[0042] Based on the calculated pipeline risk trend index and the pre-set risk level classification standards, the potential risks of the pipeline are divided into different risk levels, namely low risk level, medium risk level and high risk level. Among them, the risk level classification standards are pre-set according to the pipeline design specifications, operation history and safety requirements, and then the risk threshold of each risk level is determined. After the assessment is completed, a risk assessment report is generated, including detailed information on the abnormal area, risk factor analysis, calculation results of the pipeline risk trend index and risk level determination. At the same time, corresponding recommended measures are proposed for different risk levels. After the AUV is recovered, the risk assessment report is transmitted to relevant personnel.

[0043] A further improvement of the technical solution of the present invention is that the calculation process of the pipeline risk trend index is:

[0044] Identify the number of pipeline defect characteristics involved in the risk assessment, that is, the total number of risk factors. For each risk factor, obtain its quantitative value, corresponding benchmark value, and pre-set weight;

[0045] For each risk factor, subtract the benchmark value from its quantified value, divide the result by the benchmark value, and take the absolute value to calculate the relative deviation value of each risk factor. Analyze the relative deviation between the actual value of the risk factor and the benchmark value.

[0046] For each risk factor, calculate the ratio of its quantitative value to the benchmark value, add 1 to the result, take the natural logarithm, and calculate the logarithmic term of each risk factor;

[0047] Multiply the weight, relative deviation value and logarithmic term of each risk factor to obtain the contribution value of each risk factor. Then add the contribution values ​​of all risk factors to obtain the total contribution value. Take the square root of the total contribution value to obtain the pipeline risk trend index. The larger the value of the pipeline risk trend index, the higher the potential risk of the pipeline.

[0048] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0049] 1. The present invention provides an AUV submarine oil pipeline inspection risk identification system based on deep learning. By combining a deep learning algorithm to build an abnormal pipeline identification model, it can efficiently classify pipeline image data and identify submarine oil pipeline areas with abnormal conditions. It can then further analyze the identified abnormal areas, calculate the pipeline risk trend index, and assess the potential risk level of the pipeline. This helps to promptly discover and address potential safety hazards, prevent accidents, and ensure the safe operation of submarine oil pipelines.

[0050] 2. The present invention provides an AUV submarine oil pipeline inspection risk identification system based on deep learning. Through the AUV autonomous inspection module, it realizes automated and intelligent inspection of submarine oil pipelines. Compared with the traditional ROV inspection method, it greatly reduces manual intervention and improves inspection efficiency. At the same time, the multi-angle high-definition camera equipped by the AUV can capture pipeline images from all directions and angles, ensuring the comprehensiveness and omission of the inspection, effectively improving the inspection quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0052] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0053] Figure 2 Schematic diagram of the workflow of the image recognition and classification module of the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. 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.

[0055] Example 1, as Figure 1 、 Figure 2 As shown, the present invention provides an AUV submarine oil pipeline inspection risk identification system based on deep learning, including an inspection risk management center, which is communicatively connected to the following modules, wherein:

[0056] The inspection risk management center, as the control center of the system, is responsible for coordinating the operation of each module, receiving and processing data from each module, and issuing risk warnings;

[0057] The AUV autonomous inspection module is used to deploy AUVs to inspect submarine oil pipelines. It dynamically adjusts AUV control instructions based on the submarine environment and pipeline status, optimizes the inspection path and speed, and automatically surfaces and returns to the vessel after the inspection mission, achieving an integrated "deployment-inspection-retrieval" operation. The AUV autonomous inspection module includes an AUV deployment and submersible unit and a pipeline identification and tracking unit.

[0058] Among them, the AUV deployment and diving unit is used to deploy the AUV to the designated location through the surface deployment device, and control it to dive autonomously to the seabed area, realizing the automatic deployment and diving of the AUV, reducing manual intervention and improving operational efficiency. Before deployment, the operator conducts a comprehensive inspection of the AUV to ensure that its various systems are functioning normally and that the battery is sufficient. The operator presets the diving path and target area based on the mission requirements and the high-precision positioning system, and then installs the AUV on the surface deployment device, adjusts the angle and position of the device to make it in the optimal deployment posture. The surface deployment device is equipped with a positioning system and a communication module. The operator inputs the deployment command through the console, and the surface deployment device starts working, and the AUV is smoothly deployed to the designated surface position. At this time, the buoyancy adjustment system of the AUV starts working, keeping it on standby on the surface, ready to start the diving mission. After the AUV receives the diving command, its autonomous control system starts, controlling the buoyancy adjustment system to gradually discharge the gas in the buoyancy tank, increasing its own weight, so that it begins to slowly Slowly dive. At the same time, the AUV's propulsion system adjusts the propulsion direction and speed according to the preset diving path and information fed back by the attitude sensor to ensure the stability and accuracy of the diving process. During the diving process, the AUV's sonar system is used to monitor the seabed topography and surrounding environment in real time. When the AUV approaches the target area on the seabed, its altitude sensor and terrain matching system are used to further accurately control the diving depth and position, so that it can accurately reach the predetermined seabed area. After reaching the seabed, the AUV's positioning system communicates with the seabed positioning base station and determines its position on the seabed through acoustic positioning technology. It compares it with the preset target position. If there is a deviation, the AUV automatically adjusts its position until it completely reaches the target area. At this time, the AUV enters standby mode, and its high-definition camera and sensors begin to warm up and initialize, ready to carry out pipeline inspection tasks. At the same time, the AUV's communication system transmits its own status information and position data back to the surface mother ship in real time to ensure that the operator can grasp the working status of the AUV at any time.

[0059] The pipeline identification and tracking unit is used to use the multi-angle high-definition camera on the AUV to identify submarine oil pipelines and conduct inspections along the pipelines. It also uses a reinforcement learning algorithm to dynamically adjust the AUV's control instructions according to the submarine environment and pipeline status, optimize the inspection path and speed, improve the AUV's adaptability and inspection efficiency in complex submarine environments, and ensure the stability and safety of the inspection process. After the AUV dives to the target area on the submarine, its multi-angle high-definition camera starts working, capturing images of the submarine environment through 360-degree full-scale scanning, and using image recognition technology to compare and analyze the characteristics of the submarine oil pipeline to identify the submarine oil pipeline. At this time, the AUV's positioning system combines the initial position information of the pipeline to determine its relative position relationship with the pipeline, and then plans an initial inspection path based on the preset inspection task requirements and the preliminary distribution of the pipeline. The inspection path covers the key parts of the pipeline. During the inspection along the initial path, the AUV continuously perceives the seabed environment information, including water speed and direction, seabed topography and the real-time status of the pipeline. Then, a reinforcement learning algorithm is used for analysis. According to the current environment and pipeline status, the impact of different control instructions on the inspection effect is evaluated, and the AUV's control instructions are dynamically adjusted to optimize the inspection path and speed.

[0060] In addition, the process of dynamically adjusting the control instructions of the AUV is:

[0061] As the AUV patrols along the initial inspection route, it continuously collects information about the seabed environment and pipeline status through its various onboard sensors. This includes water flow speed and direction (obtained through flow sensors), seabed topography (scanned through a sonar system), and the real-time status of the pipeline. High-definition cameras capture and perform image recognition, and transmit the information to the AUV's central processing unit. After receiving the seabed environment and pipeline status information collected from the sensors, a reinforcement learning algorithm is run to evaluate the potential impact of different control instructions on the inspection results based on the current water flow speed, direction, and seabed topography. By simulating multiple control strategies, the safety, efficiency, and stability indicators under each control strategy are analyzed, and the expected return under each strategy is calculated. Using historical data and a preset reward mechanism, the AUV learns control instructions that achieve optimal inspection results in complex environments while avoiding collisions and deviations from the pipeline to adapt to the ever-changing seabed environment. Based on the analysis results of the reinforcement learning algorithm, the AUV's control instructions are dynamically adjusted, including the speed and direction of the thrusters, to optimize the inspection path and speed. Through dynamic adjustments, efficient inspections are achieved in complex seabed environments while ensuring the safety and stability of the inspection process.

[0062] The calculation expression of the expected return of security under each strategy is:

[0063] R(s,a)=α·Safety(s,a)+β·Efficiency(s,a)+γ·Stability(s,a);

[0064]

[0065] Where R(s, a) is the expected return of performing action a in state s, α, β, and γ are weight coefficients used to balance the impact of safety, efficiency, and stability, ranging from 0 to 1, and α+β+γ=1, Safety(s, a) is the safety index, Efficiency(s, a) is the efficiency index, Stability(s, α) is the stability index, and d(s, a) is the distance to the obstacle when performing action a in state s. The smaller the distance, the lower the safety. d is the standard deviation of the distance, which is used to adjust the sensitivity of the safety index. v(s, a) is the inspection speed when performing action a in state s. The higher the speed, the higher the efficiency. max is the maximum inspection speed, Δv(s, a) is the speed change rate when executing action a in state s. The smaller the speed change rate, the higher the stability. v is the standard deviation of the speed change rate, which is used to adjust the sensitivity of the stability index. When d(s, a) increases, Safety(s, a) tends to 1, indicating high safety. When d(s, a) decreases, Safety(s, a) tends to 0, indicating low safety. When v(s, a) increases, Efficiency(s, a) tends to 1, indicating high efficiency. When v(s, a) decreases, Efficiency(s, a) tends to 0, indicating low efficiency. When Δv(s, a) When Δv(s, a) decreases, Stability(s, a) tends to 1, indicating high stability. When Δv(s, a) increases, Stability(s, a) tends to 0, indicating low stability. When Safety(s, a), Efficiency(s, a), and Stability(s, a) are all high, R(s, a) is close to 1, indicating that the overall effect of performing action a in state s is good. When any one of the indicators is low, R(s, a) will decrease, indicating a poor overall effect.

[0066] The image acquisition and processing module is used to collect and pre-process the image data of the submarine oil pipeline during the inspection process of the AUV, perform feature analysis, extract pipeline defect features, and form a defect feature sequence table. The image acquisition and processing module includes an image acquisition unit and an image feature extraction unit;

[0067] Among them, the image acquisition unit is used to collect image data of the submarine oil pipeline through the high-definition camera carried by the AUV, and pre-process the image data, including denoising, contrast enhancement and distortion correction operations, to improve the clarity and recognizability of the image and improve the image quality. Among them, the camera is equipped with a strong light source to cope with the weak light environment of the submarine. After the AUV arrives at the submarine oil pipeline inspection area, the high-definition camera carried is self-checked to ensure that its various parameters are normal, the lens is unobstructed and there is no fault, and the brightness of the strong light source equipped with the camera is automatically adjusted according to the light conditions of the submarine environment. After the strong light source is turned on, it provides sufficient lighting for the camera to cope with the weak light environment of the submarine, and then the submarine oil pipeline is all-round and multi-angled through the high-definition camera according to the preset shooting frequency and angle. The system collects images at different resolutions and records the shooting position and time information in real time. It also pre-processes the collected image data, including denoising, contrast enhancement and distortion correction. The image is denoised to effectively remove noise interference in the image, making the image cleaner and clearer. The denoised image is contrast enhanced by using histogram equalization, adaptive contrast enhancement and other technologies to improve the contrast between different areas in the image, making the outline, details and other features of the pipeline more obvious. After contrast enhancement, the image is distortion corrected by using pre-calibrated lens distortion parameters to transform the image, eliminate barrel distortion and pincushion distortion in the image, and restore the image to its normal geometric shape. The pre-processed image data is then stored in the AUV's storage system.

[0068] An image feature extraction unit is used to perform feature analysis on the preprocessed image data, extract pipeline defect features related to the health status of the submarine oil pipeline, integrate them to obtain a defect feature sequence list, extract the preprocessed image data, and use an edge detection algorithm (Canny edge detection) to perform image segmentation on it to separate the relevant areas of the submarine oil pipeline. Combined with the basic data of the submarine oil pipeline built into the system (pipeline shape, size and position information), the boundary of the submarine oil pipeline is further located. Based on the relevant areas of the submarine oil pipeline segmented from the image, the geometric features and shape features of the submarine oil pipeline are analyzed to extract defect features related to the health status of the submarine oil pipeline, including pipeline diameter, pipeline curvature, crack length, crack width, hole area, hole diameter, corrosion depth and corrosion area ratio. The extracted defect features are integrated and, based on the inspection requirements of the submarine oil pipeline, a baseline value of each defect feature is determined, that is, the maximum allowable value of each defect feature, to form a defect feature sequence list. For each defect area, the extracted geometric features and shape features are summarized and its position information on the pipeline is annotated.

[0069] The image recognition and classification module is used to build an abnormal pipeline identification model by combining deep learning algorithms, classify pre-processed pipeline image data, and identify areas of submarine oil pipelines in abnormal conditions;

[0070] The risk assessment module is used to further analyze the identified abnormal status areas, calculate the pipeline risk trend index, and evaluate the potential risk level of the pipeline.

[0071] Example 2, as Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, the image recognition and classification module specifically includes:

[0072] A large amount of labeled submarine oil pipeline image data is collected, including images of normal states and various abnormal states such as cracks, corrosion, and deformation. Pipeline defect features related to the health status of submarine oil pipelines are extracted from them to form a comprehensive data set, and the comprehensive data set is divided into a training set and a validation set. The training set is used for model learning, and the validation set is used to evaluate model performance to ensure reasonable data distribution. A convolutional neural network (CNN) is selected as the basic architecture of the deep learning algorithm to build an abnormal pipeline recognition model. According to the characteristics of submarine oil pipeline images, the convolutional neural network model is adjusted, and the images of the training set are input into the model. The images pass through the convolution layer and the pooling layer in turn. The convolution layer automatically extracts the pipeline defect features in the image through the convolution kernel, and the pooling layer reduces the dimension of the feature map to reduce the amount of calculation. The extracted features are classified and predicted by the fully connected layer, and the output image belongs to the probability of normal or different abnormal states. The back propagation algorithm and optimizer are used to continuously adjust the model parameters to make the prediction results consistent with the true label error. The difference is reduced. During the training process, the validation set is used to monitor the model performance in real time, and the generalization ability is evaluated based on the accuracy and recall rate indicators. If overfitting occurs, the model is optimized by adding L1 and L2 regularization terms, or using the Dropout method to randomly discard some neurons to ensure good performance on the training set and validation set. After multiple rounds of iterations, until the model performance on the validation set is stable and meets the requirements, the abnormal pipeline recognition model is finally constructed and deployed. The pipeline image data to be detected is input into the abnormal pipeline recognition model, and the abnormal pipeline defect features are identified by combining the baseline values ​​of each pipeline defect feature. Then, according to the abnormal pipeline defect features and the model's probability prediction of the image belonging to different state categories, the state category of the image is judged, that is, normal state or abnormal state. For images judged to be in abnormal state, the specific location and range of the abnormal area on the pipeline are further determined, and finally the submarine oil pipeline area with abnormal state is identified;

[0073] The risk assessment module specifically includes:

[0074] Receive the abnormal state area information output by the image recognition and classification module, perform quantitative analysis on the pipeline defect characteristics of the abnormal state area, compare and analyze each pipeline defect characteristic with the pre-set pipeline defect characteristic benchmark value, and if the actual value of a pipeline defect characteristic exceeds or reaches its benchmark value, then determine that the pipeline defect characteristic is a risk factor. After determining the risk factor, analyze the degree of deviation of each risk factor from its corresponding benchmark value. Combined with the design specifications, operation history data and safety requirements of the pipeline, assign different weights to each risk factor to reflect its contribution to the overall risk of the pipeline. Then, combine the quantitative value of each risk factor with its corresponding weight to calculate the pipeline risk trend index, which reflects the changing trend of the potential risk of the pipeline. Risk trend index: According to the pre-set risk level classification standards, the potential risks of the pipeline are divided into different risk levels, namely low risk level, medium risk level and high risk level. Among them, the risk level classification standards are pre-set according to the pipeline design specifications, operation history and safety requirements, and then the risk threshold of each risk level is determined. After the assessment is completed, a risk assessment report is generated, including detailed information of the abnormal area, risk factor analysis, calculation results of the pipeline risk trend index and risk level determination. At the same time, corresponding recommended measures are proposed for different risk levels. For low-risk areas, regular monitoring is recommended. For high-risk and extremely high-risk areas, immediate maintenance or replacement measures are recommended. After the AUV is recovered, the risk assessment report is transmitted to relevant personnel;

[0075] In addition, the calculation process of pipeline risk trend index is:

[0076] Clarify the number of pipeline defect characteristics involved in the risk assessment, that is, the total number of risk factors. For each risk factor, obtain its quantitative value, corresponding benchmark value, and pre-set weight. For each risk factor, subtract the benchmark value from its quantitative value, divide the result by the benchmark value, and take the absolute value. Calculate the relative deviation value of each risk factor and analyze the relative deviation between the actual value of the risk factor and the benchmark value. For each risk factor, calculate the ratio of its quantitative value to the benchmark value, add 1 to the result, and take the natural logarithm. Calculate the logarithmic term of each risk factor, considering the relative size relationship between the actual value of the risk factor and the benchmark value. As the ratio increases, the value of the logarithmic function also increases. Multiply the weight, relative deviation value, and logarithmic term of each risk factor to obtain the contribution value of each risk factor. Then add the contribution values ​​of all risk factors to obtain the total contribution value. Take the square root of the total contribution value to obtain the pipeline risk trend index. The larger the value of the pipeline risk trend index, the higher the potential risk of the pipeline.

[0077] The calculation formula of pipeline risk trend index is:

[0078]

[0079] Where, RTI is the pipeline risk trend index, which is used to intuitively reflect the changing trend of pipeline potential risks, n is the total number of risk factors, that is, the number of pipeline defect characteristics involved in risk assessment, i is the sequence number of the risk factor, which is used to traverse each risk factor, and w i is the weight of the i-th risk factor, reflecting the contribution of this risk factor to the overall risk of the pipeline. i is the quantitative value of the i-th risk factor, that is, the specific value of the pipeline defect characteristic obtained through analysis, B i is the benchmark value corresponding to the i-th risk factor. When the quantitative values ​​of all risk factors are equal to their benchmark values, RTI = 0, indicating that the pipeline is in a risk-free state. As the deviation between the quantitative value of the risk factor and the benchmark value increases, the RTI value will also increase. When the quantitative value of a risk factor deviates more from the benchmark value, The larger the value of Will also follow increases with the increase of The value of increases, which eventually leads to an increase in the RTI value, indicating that the pipeline has a higher potential risk;

[0080] Multiple risk levels correspond to multiple risk thresholds one by one, and the corresponding relationship is as follows:

[0081] The risk threshold for low risk level is: 0 <RTI≤R L ;

[0082] The risk threshold for medium risk level is: R L <RTI≤R M ;

[0083] The risk threshold for high risk level is: RTI>R M ;

[0084] Among them, RTI is the pipeline risk trend index, R L is the upper threshold of low risk level and the lower threshold of medium risk level, R M It is the upper threshold of medium risk level and the lower threshold of high risk level.

[0085] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A deep learning-based AUV submarine oil pipeline inspection risk identification system, including an inspection risk management center, characterized in that: The inspection risk management center is connected to the following modules: The AUV autonomous inspection module is used to deploy AUVs to inspect submarine oil pipelines and dynamically adjust AUV control instructions based on the submarine environment and pipeline status to optimize inspection paths and speeds. The image acquisition and processing module is used to collect and pre-process the image data of the submarine oil pipeline during the inspection process of the AUV, perform feature analysis, extract pipeline defect features, and form a defect feature sequence table; The image recognition and classification module is used to build an abnormal pipeline identification model by combining deep learning algorithms, classify pre-processed pipeline image data, and identify areas of submarine oil pipelines in abnormal conditions; The risk assessment module is used to further analyze the identified abnormal status areas, calculate the pipeline risk trend index, and evaluate the potential risk level of the pipeline.

2. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 1 is characterized by: The AUV autonomous inspection module includes an AUV deployment and submersion unit and a pipeline identification and tracking unit; The AUV deployment and submergence unit is used to deploy the AUV to a designated location through a surface deployment device and control it to autonomously dive to the seabed area; The pipeline identification and tracking unit is used to use the multi-angle high-definition camera on the AUV to identify submarine oil pipelines, conduct inspections along the pipelines, and adopt a reinforcement learning algorithm to dynamically adjust the AUV's control instructions according to the submarine environment and pipeline status to optimize the inspection path and speed.

3. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 2 is characterized by: The AUV deployment submersible unit specifically includes: Before deployment, the operator inspects the AUV and presets the diving path and target area based on the mission requirements and the high-precision positioning system. The operator then installs the AUV on the surface deployment device and smoothly releases the AUV to the designated surface location. After receiving the dive command, the AUV's autonomous control system starts, controlling the buoyancy control system to gradually discharge the gas in the buoyancy tank, increasing its own weight and causing it to slowly dive. At the same time, the AUV's propulsion system adjusts the propulsion direction and speed based on the preset dive path and information fed back by the attitude sensor. When the AUV approaches the target area on the seabed, its altitude sensor and terrain matching system further accurately control the dive depth and position. After reaching the seabed, the AUV's positioning system communicates with the seabed positioning base station, and uses acoustic positioning technology to determine its own position on the seabed and compare it with the preset target position. If there is a deviation, the AUV automatically adjusts its position until it completely reaches the target area. At this time, the AUV enters standby mode, and its high-definition camera and sensors begin to warm up and initialize, ready to carry out pipeline inspection tasks.

4. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 2 is characterized by: The pipeline identification and tracking unit specifically includes: After the AUV descends to the target seabed area, its onboard multi-angle high-definition camera begins working, performing a 360-degree full-scale scan to capture images of the seabed environment. Using image recognition technology, it compares and analyzes the characteristics of the submarine oil pipeline, identifying its location and direction. The AUV's positioning system, combined with the initial position information of the pipeline, determines its relative position to the pipeline, and then plans an initial inspection route based on the preset inspection task requirements and the initial distribution of the pipeline. During the inspection along the initial path, the AUV continuously perceives information about the seabed environment, including water flow speed and direction, seabed topography, and the real-time status of the pipeline. A reinforcement learning algorithm is then used for analysis. Based on the current environment and pipeline status, the impact of different control instructions on the inspection effect is evaluated, and the AUV's control instructions are dynamically adjusted to optimize the inspection path and speed.

5. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 2 is characterized by: The image acquisition and processing module includes an image acquisition unit and an image feature extraction unit; The image acquisition unit is used to collect image data of the submarine oil pipeline through the high-definition camera carried by the AUV and pre-process the image data; The image feature extraction unit is used to perform feature analysis on the pre-processed image data, extract pipeline defect features related to the health status of the submarine oil pipeline, and integrate them to obtain a defect feature sequence table.

6. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 5 is characterized by: The image acquisition unit specifically includes: After the AUV arrives at the submarine oil pipeline inspection area, it conducts a self-inspection of the high-definition camera on board and automatically adjusts the brightness of the camera's strong light source according to the submarine ambient light conditions. The high-definition camera then captures images of the submarine oil pipeline at a preset shooting frequency and angle, while also recording the shooting location and time information in real time. The collected image data is preprocessed, including denoising, contrast enhancement and distortion correction operations, and then the preprocessed image data is stored in the AUV's storage system.

7. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 5 is characterized by: The image feature extraction unit specifically includes: Extract the pre-processed image data and perform image segmentation using an edge detection algorithm to separate the relevant areas of the submarine oil pipeline. Combined with the basic submarine oil pipeline data built into the system, the boundaries of the submarine oil pipeline are further located. Based on the relevant areas of the submarine oil pipeline segmented from the image, the geometric and shape features of the submarine oil pipeline are analyzed, and defect features related to the health status of the submarine oil pipeline are extracted, including pipeline diameter, pipeline curvature, crack length, crack width, hole area, hole diameter, corrosion depth, and corrosion area ratio; The extracted defect features are integrated, and according to the inspection requirements of submarine oil pipelines, the baseline value of each defect feature is determined, that is, the maximum allowable value of each defect feature, to form a defect feature sequence table. For each defect area, the extracted geometric features and shape features are summarized and their location information on the pipeline is marked.

8. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 7 is characterized by: The image recognition and classification module specifically includes: Collect annotated submarine oil pipeline image data, including images of normal and abnormal states, and extract pipeline defect features related to the health status of submarine oil pipelines to form a comprehensive dataset. The comprehensive dataset is then divided into a training set and a validation set. A convolutional neural network was selected as the underlying deep learning algorithm architecture to construct an abnormal pipeline identification model. Based on the characteristics of submarine oil pipeline images, the convolutional neural network model was adjusted. Images from the training set were fed into the model, and the model parameters were continuously adjusted using a backpropagation algorithm and optimizer. During training, the model performance was monitored in real time using a validation set. After multiple rounds of iterations, the model achieved stable performance on the validation set and met the requirements. Finally, the abnormal pipeline identification model was constructed. The trained abnormal pipeline recognition model is deployed and the pipeline image data to be inspected is input into the abnormal pipeline recognition model. The model then identifies abnormal pipeline defect features based on the baseline values ​​of each pipeline defect feature. Based on these abnormal pipeline defect features and the model's probability prediction of the image belonging to different state categories, the model determines the image's state category, i.e., normal or abnormal. For images judged to be in an abnormal state, the specific location and range of the abnormal area on the pipeline are further determined, ultimately identifying the abnormal submarine oil pipeline area.

9. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 8, characterized in that: The risk assessment module specifically includes: Receive the abnormal state area information output by the image recognition and classification module, perform quantitative analysis on the pipeline defect characteristics in the abnormal state area, compare and analyze each pipeline defect characteristic with the pre-set pipeline defect characteristic benchmark value, and determine that the pipeline defect characteristic is a risk factor if the actual value of a pipeline defect characteristic exceeds or reaches its benchmark value; After determining the risk factors, the degree of deviation of each risk factor from its corresponding benchmark value is analyzed. Based on the pipeline design specifications, historical operation data, and safety requirements, different weights are assigned to each risk factor. The quantitative value of each risk factor is then combined with its corresponding weight to calculate the pipeline risk trend index, which reflects the changing trend of the pipeline's potential risks. Based on the calculated pipeline risk trend index and the pre-set risk level classification standards, the potential risks of the pipeline are divided into different risk levels, namely low risk level, medium risk level and high risk level. Among them, the risk level classification standards are pre-set according to the pipeline design specifications, operation history and safety requirements, and then the risk threshold of each risk level is determined. After the assessment is completed, a risk assessment report is generated, including detailed information on the abnormal area, risk factor analysis, calculation results of the pipeline risk trend index and risk level determination. At the same time, corresponding recommended measures are proposed for different risk levels. After the AUV is recovered, the risk assessment report is transmitted to relevant personnel.

10. The deep learning-based AUV submarine oil pipeline inspection risk identification system according to claim 9 is characterized by: The calculation process of the pipeline risk trend index is as follows: Identify the number of pipeline defect characteristics involved in the risk assessment, that is, the total number of risk factors. For each risk factor, obtain its quantitative value, corresponding benchmark value, and pre-set weight; For each risk factor, subtract the benchmark value from its quantified value, divide the result by the benchmark value, and take the absolute value to calculate the relative deviation value of each risk factor. Analyze the relative deviation between the actual value of the risk factor and the benchmark value. For each risk factor, calculate the ratio of its quantitative value to the benchmark value, add 1 to the result, take the natural logarithm, and calculate the logarithmic term of each risk factor; Multiply the weight, relative deviation value and logarithmic term of each risk factor to obtain the contribution value of each risk factor. Then add the contribution values ​​of all risk factors to obtain the total contribution value. Take the square root of the total contribution value to obtain the pipeline risk trend index.

Images