Artificial intelligence-based offshore booster station steel structure safety monitoring method and system

Abstract

The application discloses an offshore booster station steel structure safety monitoring method and system based on artificial intelligence, which collects stress, vibration, corrosion, crack and environment data of the steel structure through multi-modal sensors arranged at steel structure parts of the offshore booster station, and time-aligns the multi-modal data through a time synchronization protocol, wherein the multi-modal sensors include stress sensors, vibration sensors, corrosion sensors, ultrasonic sensors and environment sensors. Based on finite element analysis and real-time sensor data, a dynamically updatable digital twin model is constructed, which breaks through the limitations of traditional static monitoring, can reflect the stress distribution and damage state of the steel structure in real time, and provides a scientific basis for simulating the structural response under extreme loads, thereby significantly improving the accuracy and predictability of the steel structure safety evaluation under the environment.

Description

Technical Field

[0001] This invention relates to the field of marine engineering structure health monitoring technology, specifically to an artificial intelligence-based method and system for safety monitoring of steel structures in offshore substations. Background Technology

[0002] Offshore substations are a crucial component of offshore wind farms, primarily functioning to collect the electricity output from wind turbines and boost its voltage for transmission to the onshore power grid. In the marine environment, the steel structures of offshore substations must withstand prolonged wind loads, waves, salt spray corrosion, and cyclic fatigue loads. External factors can lead to stress concentration, crack propagation, and corrosion damage in the steel structures, threatening their long-term safety and service life.

[0003] Traditional manual inspection: This involves professional personnel periodically inspecting the exterior of the steel structure, using methods such as ultrasonic and radiographic testing to assess cracks and corrosion. Relying on human experience, the monitoring results are easily affected by human factors, resulting in low efficiency and an inability to achieve real-time monitoring.

[0004] Sensor monitoring: Stress sensors, vibration sensors, or corrosion sensors are installed at key locations in the steel structure, and the sensor data is used to analyze the health status of the steel structure. However, this method is usually based on data from a single type of sensor, which cannot comprehensively reflect the actual operating status of the steel structure, and the sensor data is susceptible to noise interference, affecting the accuracy of monitoring.

[0005] Finite element analysis (FEM) is used to simulate the stress state and crack propagation of steel structures. However, FEM requires offline calculations and cannot dynamically update the model using real-time monitoring data, thus making it unsuitable for adapting to the ever-changing marine environment.

[0006] Lagging early warning and maintenance mechanisms: Existing monitoring systems mostly use fixed thresholds for early warning or rely on manual experience to formulate maintenance plans. They cannot combine real-time monitoring data for scientific evaluation and optimization, which can lead to over-maintenance or delayed repairs and increase operation and maintenance costs. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for safety monitoring of steel structures in offshore substations, thereby resolving the problems mentioned in the background section.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] The artificial intelligence-based safety monitoring method for steel structures of offshore substations includes the following steps:

[0010] Step 1, Data Acquisition: Using multimodal sensors deployed on the steel structure of the offshore substation, data on stress, vibration, corrosion, cracks, and environmental conditions of the steel structure are collected, and the multimodal data are time-aligned using a time synchronization protocol; the acquisition frequency of the multimodal sensors is dynamically adjusted, using the first sampling frequency during normal operation and increasing to the second sampling frequency under extreme weather conditions;

[0011] The formula for calculating the first sampling frequency is: ,in, This indicates the sensor sampling frequency under normal operating conditions. This represents the average value of the stress signal collected by the sensor within a given time period. This indicates the maximum stress value that the sensor can detect. , The scaling factor and offset represent the sampling frequency;

[0012] The formula for calculating the second sampling frequency is: ,in, This indicates the sensor sampling frequency under extreme weather conditions. This indicates the wind speed as monitored in real time. This indicates the set wind speed threshold. , The scaling factor and offset represent the sampling frequency of extreme weather events;

[0013] Step 2, Data Preprocessing: The collected multimodal data is subjected to signal denoising and normalization, and abnormal data is filtered and marked by setting an anomaly threshold;

[0014] Step 3, Edge Computing: On the edge computing device, feature extraction and preliminary anomaly detection are performed on the processed multimodal data, and the filtered abnormal data is uploaded to the cloud analysis system in real time;

[0015] Step 4, Artificial Intelligence Analysis: The uploaded abnormal data and related multimodal data are fused and analyzed using a cloud-based analysis system to construct an anomaly detection module and a trend prediction module based on a deep learning model, predicting the crack propagation and corrosion development trends of the steel structure. The artificial intelligence analysis uses a long short-term memory network to detect anomalies in time-series data, and the constructed trend prediction module predicts crack propagation and corrosion rates based on a gradient-enhanced decision tree. The gradient-enhanced decision tree is an ensemble learning algorithm based on an additive model, forward step-by-step optimization, and decision trees.

[0016] Step 5: Digital Twin Model Construction: Based on the initial finite element model of the steel structure and the results of artificial intelligence analysis, a digital twin model is established and dynamically updated to simulate the health status of the steel structure under extreme load conditions.

[0017] Step 6, Intelligent Early Warning and Maintenance: Based on the digital twin model and artificial intelligence analysis results, calculate the health score, trigger tiered early warnings, and generate an optimized predictive maintenance plan; Health score calculation formula: ,in, Rate your health. As of the current level of damage, The threshold is the critical damage value; based on the health score, low, medium, or high-level warnings are triggered, and corresponding maintenance plans are generated.

[0018] A further improvement of this invention is that wavelet transform is used for signal denoising in step 2, and normalization is performed using the following method:

[0019] The wavelet transform: ,

[0020] in, The signal value at time t. The number of layers in the decomposition. The time step of the signal, Let K be the coefficient of the j-th layer at wavelet decomposition position k. Let be the basis function of the j-th layer and wavelet decomposition position k;

[0021] Normalization formula: ,in, The normalized signal value, with a range of (0, 1); This is the original value of the current signal. The minimum value of the signal. This represents the maximum value of the signal.

[0022] Mean based on historical data and standard deviation The data is filtered, and data that meets the following criteria is marked as anomalies:

[0023] ,

[0024] in, This is the original value of the current signal. The historical average value of the signal. The standard deviation of the signal. This is the threshold for anomaly detection.

[0025] A further improvement of the present invention is that the edge computing device in step 3 extracts features from the marked abnormal data and transmits them to the cloud in real time, while storing the normal data locally.

[0026] A further improvement of this invention is that, in step 4, the gradient-enhanced decision tree is based on the following basic formula of an ensemble learning algorithm that combines an additive model, forward stepwise optimization, and decision trees:

[0027] Additive model representation: ,

[0028] in, It is the output of the ensemble model. It is the total number of decision trees. This represents the prediction result for the m-th tree;

[0029] Objective function: ,

[0030] in, L is the regularization term, and L is the objective function of the gradient boosting decision tree. It is the true value of the i-th sample. It is the input feature of the i-th sample. It is the first Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, Represents the loss function;

[0031] Loss function expansion, gradient update part:

[0032] ,

[0033] in, It is a first-order gradient. It is a second-order gradient. It is the true value of the i-th sample. It is the input feature of the i-th sample. No. Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, It is a loss function;

[0034] Regularization term: ,

[0035] in, It is the number of leaf nodes. It is the leaf node weight. and It is the regularization coefficient. It is a regularization term;

[0036] Node splitting gain: ,

[0037] Where I is the sample set of the current node, and This represents the set of left and right node samples after the split. This represents the gain from node splitting. It is the sum of the first-order gradients of the current node sample. It is the sum of the second-order gradients of the current node sample. and It is the regularization coefficient.

[0038] A further improvement of this invention is that the multimodal data fusion in step 4 employs the following deep neural network formula:

[0039] ,

[0040] in, For multimodal data input, This is the weight matrix. For bias terms, This is the activation function.

[0041] A further improvement of this invention is that the digital twin model established in step 5 is based on the finite element analysis of the steel structure, updates the geometric parameters and physical state of the model in real time, and dynamically simulates the structural response under extreme load conditions; the stress distribution of key parts is calculated based on the finite element method.

[0042] ,

[0043] in, For stress in critical parts, For the action force, This represents the cross-sectional area of ​​the stressed part;

[0044] The dynamic stress update formula, combined with real-time data collected by sensors, updates the stress state of the steel structure under dynamic loads:

[0045] ,

[0046] in, For real-time dynamic stress, The initial static stress, For dynamic stress changes;

[0047] Crack propagation prediction formula, based on crack propagation theory, predicts the crack propagation rate of steel structures:

[0048] ,

[0049] in, The crack growth rate caused by each cyclic load; , These are material property constants. The range of stress intensity factor is given by the following formula:

[0050] ,

[0051] in, For crack geometry factor, The length of the crack. This refers to dynamic stress changes.

[0052] An AI-based safety monitoring system for offshore substation steel structures, based on the aforementioned AI-based safety monitoring method for offshore substation steel structures, includes:

[0053] The data acquisition module collects stress, vibration, corrosion, crack and environmental data of the steel structure through multimodal sensors deployed on the steel structure of the offshore substation, and performs time synchronization of the multimodal data through a time synchronization protocol.

[0054] The data preprocessing module performs signal denoising and normalization on the collected multimodal data, and filters and marks abnormal data by setting an anomaly threshold.

[0055] The edge computing module performs feature extraction and preliminary anomaly detection on the processed multimodal data on the edge computing device, and the filtered abnormal data is uploaded to the cloud analysis system in real time.

[0056] The artificial intelligence analysis module integrates and analyzes uploaded abnormal data and related multimodal data through a cloud-based analysis system, and constructs an anomaly detection module and a trend prediction module based on a deep learning model to predict the crack propagation and corrosion development trends of steel structures.

[0057] The digital twin model building module, combined with the results of artificial intelligence analysis, establishes and dynamically updates a digital twin model based on the initial finite element model of the steel structure, which is used to simulate the health status of the steel structure under extreme load conditions.

[0058] The intelligent early warning and maintenance module calculates a health score based on a digital twin model and artificial intelligence analysis results, triggers tiered early warnings, and generates optimized predictive maintenance plans.

[0059] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0060] This invention constructs a dynamically updatable digital twin model based on finite element analysis and real-time sensor data, breaking through the limitations of traditional static monitoring. It can reflect the stress distribution and damage state of steel structures in real time, and provides a scientific basis for simulating structural response under extreme loads, significantly improving the accuracy and predictability of steel structure safety assessment in various environments.

[0061] This invention achieves intelligent fusion analysis of multimodal data through deep learning algorithms, which can accurately detect anomalies and predict crack propagation and corrosion rates. Furthermore, the artificial intelligence model can extract implicit patterns in the data, demonstrating significant advantages in multimodal data fusion and dynamic trend prediction, thus enabling steel structure safety monitoring to enter an intelligent stage.

[0062] This invention designs a graded early warning mechanism based on health scores, which comprehensively considers multi-dimensional information such as stress distribution, crack propagation, and corrosion degree. The quantitative analysis of health scores can accurately trigger low-level, medium-level, or high-level early warnings, significantly reducing the delay in identifying safety hazards. It has advantages in the variable marine environment, can buy management personnel reaction time, and effectively reduce the risk of accidents.

[0063] This invention combines artificial intelligence models to predict crack propagation and corrosion trends, enabling trend-based maintenance optimization. Predictive maintenance can plan maintenance time and scope in advance, avoiding resource waste and delayed repairs, improving the targeting and accuracy of maintenance, significantly reducing overall maintenance costs, and extending the service life of steel structures. Attached Figure Description

[0064] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0065] Figure 1 This is a flowchart of the artificial intelligence-based safety monitoring method for steel structures of offshore substations according to the present invention.

[0066] Figure 2 This is a structural block diagram of the steel structure safety monitoring system for offshore substations based on artificial intelligence, as described in this invention. Detailed Implementation

[0067] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0068] Example 1

[0069] Please see the appendix Figure 1The present invention provides an artificial intelligence-based method for safety monitoring of steel structures in offshore substations, comprising the following steps:

[0070] Step 1, Data Acquisition: By deploying multimodal sensors on the steel structure of the offshore substation, data on stress, vibration, corrosion, cracks, and the environment of the steel structure are collected. The multimodal data is time-aligned using a time synchronization protocol. The multimodal sensors include stress sensors, vibration sensors, corrosion sensors, ultrasonic sensors, and environmental sensors, and the sensors are deployed at welds, nodes, and load-bearing components of the steel structure.

[0071] Step 2, Data Preprocessing: Perform signal denoising and normalization on the multimodal data collected in Step 1, and filter and mark abnormal data by setting an anomaly threshold;

[0072] Step 3, Edge Computing: On the edge computing device, feature extraction and preliminary anomaly detection are performed on the multimodal data processed in Step 2, and the filtered abnormal data is uploaded to the cloud analysis system in real time;

[0073] Step 4, Artificial Intelligence Analysis: The abnormal data and related multimodal data uploaded in Step 3 are fused and analyzed through a cloud-based analysis system to construct an anomaly detection module and a trend prediction module based on a deep learning model, predicting the crack propagation and corrosion development trends of steel structures.

[0074] Step 5: Digital Twin Model Construction: Based on the results of the artificial intelligence analysis in Step 4, a digital twin model is established and dynamically updated to simulate the health status of the steel structure under extreme load conditions, using the initial finite element model of the steel structure.

[0075] Step 6, Intelligent Early Warning and Maintenance: Based on the digital twin model and artificial intelligence analysis results from Step 5, calculate the health score, trigger graded early warnings, and generate an optimized predictive maintenance plan.

[0076] Step 1: By deploying multimodal sensors at key locations on the steel structure of the offshore substation, stress, vibration, corrosion, crack, and environmental data can be collected, covering multi-dimensional information on the health status of the steel structure. The application of a time synchronization protocol ensures the time consistency of data from different sensors, laying the foundation for subsequent data fusion analysis. Comprehensive acquisition of multimodal data effectively compensates for the shortcomings of traditional single-data monitoring methods, significantly improving the comprehensiveness and accuracy of monitoring.

[0077] Step 2 involves denoising and normalizing the collected multimodal data, and filtering out abnormal data by setting an anomaly threshold to identify potential problems. Step 2 uses denoising technology to eliminate the impact of environmental interference on sensor data, normalization to address differences between data of different dimensions, and anomaly screening to detect structural problems early. Data preprocessing improves data quality, providing high-quality input for subsequent feature extraction and artificial intelligence analysis.

[0078] Step 3: The edge computing device performs feature extraction and preliminary anomaly detection on the preprocessed multimodal data, and uploads the selected anomaly data to the cloud analysis system in real time. This step extracts key information through feature extraction and utilizes the edge computing device to achieve preliminary local anomaly detection, reducing the computational load and data transmission pressure on the cloud system. Real-time anomaly uploading enables the system to respond quickly, improving the efficiency and sensitivity of steel structure safety monitoring.

[0079] Step 4: In the cloud-based analysis system, a deep learning model is used to fuse and analyze the abnormal data and multimodal data uploaded from edge computing, achieving anomaly detection and trend prediction. The artificial intelligence model can extract implicit patterns from multimodal data, accurately detect abnormal conditions, and predict the crack propagation and corrosion development trends of steel structures. Compared to traditional monitoring methods, artificial intelligence analysis significantly improves the sensitivity of anomaly detection and the accuracy of trend prediction, providing strong technical support for the dynamic assessment of structural health status.

[0080] Step 5: Combining the AI ​​analysis results and the initial finite element model, establish and dynamically update a digital twin model to simulate the health status of steel structures under extreme load conditions. The digital twin model integrates real-time sensor data to visually display the stress distribution, crack propagation, and corrosion level of the steel structure, while simulating the structural response under extreme wind loads or wave impacts. This model overcomes the limitations of traditional static monitoring, providing a more accurate and scientific basis for structural health assessment in the environment.

[0081] Step 6: Based on the digital twin model and artificial intelligence analysis results, calculate the health score, trigger tiered early warnings, and generate an optimized predictive maintenance plan. Through the health score, the system can quantify the health status of the steel structure into intuitive indicators and trigger low, medium, or high-level early warnings, significantly reducing the delay in identifying safety hazards. Predictive maintenance, by planning maintenance time and scope in advance, optimizes resource allocation, avoids unnecessary maintenance costs and delayed repairs, extends the service life of the steel structure, and improves operation and maintenance efficiency.

[0082] In summary, each step of this invention is a key link in the overall safety monitoring process. From the comprehensive collection of multimodal data to the accurate prediction of artificial intelligence analysis, to the dynamic simulation and intelligent early warning of the digital twin model, each step aims to solve the shortcomings of traditional monitoring methods and improve the intelligence level and economy of steel structure life cycle management.

[0083] The sampling frequency of the multimodal sensor is dynamically adjusted. During normal operation, the first sampling frequency is used, and under extreme weather conditions, it is increased to the second sampling frequency.

[0084] The formula for calculating the first sampling frequency is as follows:

[0085] ,

[0086] in, This indicates the sensor sampling frequency under normal operating conditions. This represents the average value of the stress signal collected by the sensor within a given time period. This indicates the maximum stress value that the sensor can detect. , The scaling factor and offset represent the sampling frequency;

[0087] The formula for calculating the second sampling frequency is as follows:

[0088] ,

[0089] in, This indicates the sensor sampling frequency under extreme weather conditions. This indicates the wind speed as monitored in real time. This indicates the set wind speed threshold. , The scaling factor and offset represent the sampling frequency of extreme weather events.

[0090] The sampling frequency of the multimodal sensor is dynamically adjusted to optimize the utilization efficiency and response speed of monitoring resources according to changes in the operating environment. Under normal operating conditions, the first sampling frequency is used, and the sampling frequency is dynamically adjusted by the ratio of the average value of the sensor stress signal to the maximum detection value to ensure the accuracy of monitoring while avoiding resource waste caused by excessively high sampling frequencies.

[0091] Under extreme weather conditions, a second sampling frequency is calculated based on real-time wind speed and wind speed thresholds to ensure the capture of detailed data in high-risk environments and rapid identification of potential problems. This mechanism effectively balances monitoring accuracy and system resource utilization, improving adaptability and monitoring efficiency under various operating conditions.

[0092] Step 2 uses wavelet transform for signal denoising, and normalization is performed using the following formula:

[0093] Wavelet transform: ,

[0094] in, The signal value at time t. The number of layers in the decomposition. The time step of the signal, Let K be the coefficient of the j-th layer at wavelet decomposition position k. Let be the basis function of the j-th layer and wavelet decomposition position k;

[0095] Normalization formula: ,

[0096] in, The normalized signal value, with a range of (0, 1);

[0097] This is the original value of the current signal. The minimum value of the signal. This represents the maximum value of the signal.

[0098] Mean based on historical data and standard deviation The data is filtered, and data that meets the following criteria is marked as anomalies:

[0099] ,

[0100] in, This is the original value of the current signal. The historical average value of the signal. The standard deviation of the signal. The threshold for anomaly detection;

[0101] In step 3, the edge computing device extracts features from the abnormal data marked in step 2 and transmits it to the cloud in real time, while storing the normal data locally.

[0102] By performing signal denoising using wavelet transform in step 2, environmental noise in the sensor-acquired signals can be effectively removed while retaining key feature information, significantly improving data quality.

[0103] Normalization process unifies data of different dimensions into the same range (0, 1), which facilitates subsequent analysis and model fusion;

[0104] Anomaly detection mechanisms based on historical mean and standard deviation can quickly identify data points that significantly deviate from the normal range, providing accurate evidence for anomaly identification.

[0105] Combined with edge computing in step 3, abnormal data is extracted and transmitted to the cloud analysis system in real time to achieve rapid response, while normal data is stored locally to reduce the computing and transmission burden on the cloud, further optimizing the system's resource utilization and operating efficiency, improving the real-time performance and accuracy of data processing, enhancing the sensitivity to monitoring sudden anomalies, and providing high-quality data support for subsequent analysis and decision-making.

[0106] The artificial intelligence analysis in step 4 uses a long short-term memory network to detect anomalies in the time series data, and the constructed trend prediction module predicts crack propagation and corrosion rate based on a gradient-enhanced decision tree.

[0107] Gradient-boosting decision trees are an ensemble learning algorithm based on additive models, forward stepwise optimization, and decision trees. Their basic formula is as follows:

[0108] Additive model representation: ,

[0109] in, It is the output of the ensemble model. It is the total number of decision trees. This represents the prediction result for the m-th tree;

[0110] Objective function: ,

[0111] in, L is the regularization term, and L is the objective function of the gradient boosting decision tree. It is the true value of the i-th sample. It is the input feature of the i-th sample. No. Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, Represents the loss function;

[0112] Loss function expansion (gradient update part):

[0113] ,

[0114] in, It is a first-order gradient. It is a second-order gradient. It is the true value of the i-th sample. It is the input feature of the i-th sample. No. Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, It is a loss function;

[0115] Regularization term: ,

[0116] in, It is the number of leaf nodes. It is the leaf node weight. and It is the regularization coefficient. It is a regularization term;

[0117] Node splitting gain: ,

[0118] Where I is the sample set of the current node, and This represents the set of left and right node samples after the split. This represents the gain from node splitting. It is the sum of the first-order gradients of the current node sample. It is the sum of the second-order gradients of the current node sample. and It is the regularization coefficient;

[0119] In step 4, the multimodal data fusion uses the following deep neural network formula:

[0120] ,

[0121] in, For multimodal data input, This is the weight matrix. For bias terms, This is the activation function.

[0122] In step 4, artificial intelligence analysis is used to perform in-depth processing of multimodal data, enabling intelligent anomaly detection and trend prediction. Long Short-Term Memory (LSTM) networks are used to detect anomalies in time-series data, capturing the long-term dependencies and dynamic changes in the operating status of steel structures, effectively improving the accuracy and sensitivity of anomaly detection.

[0123] The trend prediction module based on gradient-enhanced decision trees accurately predicts crack propagation and corrosion rates through additive models, forward step-by-step optimization, and regularization terms, demonstrating superior generalization performance and modeling capabilities for nonlinear data.

[0124] Deep fusion analysis of multimodal data employs a neural network model, which can extract potential correlations between different data sources, improving the comprehensiveness and efficiency of the overall analysis.

[0125] This step, by integrating the advantages of deep learning and gradient-enhanced decision trees, provides strong technical support for the accurate assessment of the health status of steel structures and the prediction of future trends, laying a solid foundation for subsequent early warning and maintenance decisions.

[0126] The digital twin model established in step 5 is based on the finite element analysis of the steel structure, updates the geometric parameters and physical state of the model in real time, and dynamically simulates the structural response under extreme load conditions.

[0127] Based on the finite element method, the stress distribution in key components is calculated:

[0128] ,

[0129] in, For stress in critical parts, For the action force, This represents the cross-sectional area of ​​the stressed part;

[0130] The dynamic stress update formula, combined with real-time data collected by sensors, updates the stress state of the steel structure under dynamic loads:

[0131] ,

[0132] in, For real-time dynamic stress, The initial static stress, For dynamic stress changes;

[0133] Crack propagation prediction formula, based on crack propagation theory, predicts the crack propagation rate of steel structures:

[0134] ,

[0135] in, The crack growth rate caused by each cyclic load; , These are material property constants. The range of stress intensity factor is given by the following formula:

[0136] ,

[0137] in, For crack geometry factor, The length of the crack. For dynamic stress changes, This refers to dynamic stress changes.

[0138] The digital twin model established in step 5, based on finite element analysis and real-time sensor data, enables dynamic simulation and prediction of the health status of the steel structure. By calculating the stress distribution of key components using the finite element method and combining it with sensor data to update geometric parameters and physical state in real time, it can dynamically reflect the stress changes of the steel structure under different load conditions.

[0139] The crack propagation prediction formula combines crack propagation theory and stress intensity factor to accurately simulate the crack propagation rate, providing a forward-looking assessment of structural damage development. Especially under extreme load conditions, the digital twin model can dynamically simulate the stress response and crack propagation trend of steel structures in real time, providing managers with a scientific basis and improving the accuracy of safety assessments.

[0140] This model overcomes the limitations of traditional static analysis and significantly enhances the ability to assess the health and predict the risks of steel structures in marine environments.

[0141] The formula for calculating the health score in step 6 is:

[0142] ,

[0143] in, Rate your health. As of the current level of damage, This is the critical damage value;

[0144] Based on the health score, low, medium, or high alerts are triggered, and corresponding maintenance plans are generated.

[0145] The health score calculation formula in step 6 quantifies the health status of the steel structure, providing a scientific basis for early warning and maintenance.

[0146] The health score H comprehensively considers the current degree of injury (D) and the critical injury value. It can transform the structural health status into intuitive numerical indicators, triggering low-level, medium-level, or high-level early warnings.

[0147] Low-level warnings indicate minor damage risks, medium-level warnings indicate potential hazards that require close attention, and high-level warnings alert to structural failure risks, giving managers more time to react.

[0148] By using a health score-driven tiered early warning and maintenance plan, targeted maintenance can be achieved, avoiding unnecessary waste of resources, reducing operation and maintenance costs, and significantly improving the safety and reliability of steel structure operation.

[0149] Example 2

[0150] like Figure 2 As shown, the artificial intelligence-based steel structure safety monitoring system for offshore substations provided by this invention includes:

[0151] The data acquisition module collects stress, vibration, corrosion, crack and environmental data of the steel structure through multimodal sensors deployed on the steel structure of the offshore substation, and performs time synchronization of the multimodal data through a time synchronization protocol.

[0152] The data preprocessing module performs signal denoising and normalization on the collected multimodal data, and filters and marks abnormal data by setting an anomaly threshold.

[0153] The edge computing module performs feature extraction and preliminary anomaly detection on the processed multimodal data on the edge computing device, and the filtered abnormal data is uploaded to the cloud analysis system in real time.

[0154] The artificial intelligence analysis module integrates and analyzes uploaded abnormal data and related multimodal data through a cloud-based analysis system, and constructs an anomaly detection module and a trend prediction module based on a deep learning model to predict the crack propagation and corrosion development trends of steel structures.

[0155] The digital twin model building module, combined with the results of artificial intelligence analysis, establishes and dynamically updates a digital twin model based on the initial finite element model of the steel structure, which is used to simulate the health status of the steel structure under extreme load conditions.

[0156] The intelligent early warning and maintenance module calculates a health score based on a digital twin model and artificial intelligence analysis results, triggers tiered early warnings, and generates optimized predictive maintenance plans.

[0157] In this embodiment, the sampling frequency of the multimodal sensor in the data acquisition module is dynamically adjusted. During normal operation, a first sampling frequency is used, and under extreme weather conditions, it is increased to a second sampling frequency.

[0158] The formula for calculating the first sampling frequency is as follows:

[0159] ,

[0160] in, This indicates the sensor sampling frequency under normal operating conditions. This represents the average value of the stress signal collected by the sensor within a given time period. This indicates the maximum stress value that the sensor can detect. , The scaling factor and offset represent the sampling frequency;

[0161] The formula for calculating the second sampling frequency is as follows:

[0162] ,

[0163] in, This indicates the sensor sampling frequency under extreme weather conditions. This indicates the wind speed as monitored in real time. This indicates the set wind speed threshold. , The scaling factor and offset represent the sampling frequency of extreme weather events.

[0164] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0165] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A safety monitoring method for steel structures of offshore substations based on artificial intelligence, characterized in that, Includes the following steps: Step 1, Data Acquisition: Using multimodal sensors deployed on the steel structure of the offshore substation, data on stress, vibration, corrosion, cracks, and environmental conditions of the steel structure are collected, and the multimodal data are time-aligned using a time synchronization protocol; the acquisition frequency of the multimodal sensors is dynamically adjusted, using the first sampling frequency during normal operation and increasing to the second sampling frequency under extreme weather conditions; The formula for calculating the first sampling frequency is: ,in, This indicates the sensor sampling frequency under normal operating conditions. This represents the average value of the stress signal collected by the sensor within a given time period. This indicates the maximum stress value that the sensor can detect. , The scaling factor and offset represent the sampling frequency; The formula for calculating the second sampling frequency is: ,in, This indicates the sensor sampling frequency under extreme weather conditions. This indicates the wind speed as monitored in real time. This indicates the set wind speed threshold. , The scaling factor and offset represent the sampling frequency of extreme weather events; Step 2, Data Preprocessing: The collected multimodal data is subjected to signal denoising and normalization, and abnormal data is filtered and marked by setting an anomaly threshold; Step 3, Edge Computing: On the edge computing device, feature extraction and preliminary anomaly detection are performed on the processed multimodal data, and the filtered abnormal data is uploaded to the cloud analysis system in real time; Step 4, Artificial Intelligence Analysis: The uploaded abnormal data and related multimodal data are fused and analyzed using a cloud-based analysis system to construct an anomaly detection module and a trend prediction module based on a deep learning model, predicting the crack propagation and corrosion development trends of the steel structure. The artificial intelligence analysis uses a long short-term memory network to detect anomalies in time-series data, and the constructed trend prediction module predicts crack propagation and corrosion rates based on a gradient-enhanced decision tree. The gradient-enhanced decision tree is an ensemble learning algorithm based on an additive model, forward step-by-step optimization, and decision trees. Step 5: Digital Twin Model Construction: Based on the initial finite element model of the steel structure and the results of artificial intelligence analysis, a digital twin model is established and dynamically updated to simulate the health status of the steel structure under extreme load conditions. Step 6, Intelligent Early Warning and Maintenance: Based on the digital twin model and artificial intelligence analysis results, calculate the health score, trigger tiered early warnings, and generate an optimized predictive maintenance plan; Health score calculation formula: ,in, Rate your health. As of the current level of damage, The threshold is the critical damage value; based on the health score, low, medium, or high-level warnings are triggered, and corresponding maintenance plans are generated.

2. The method for safety monitoring of steel structures of offshore substations based on artificial intelligence according to claim 1, characterized in that, In step 2, wavelet transform is used for signal denoising, and normalization is performed using the following method: The wavelet transform: , in, The signal value at time t. The number of layers in the decomposition. The time step of the signal, Let K be the coefficient of the j-th layer at wavelet decomposition position k. Let be the basis function of the j-th layer and wavelet decomposition position k; Normalization formula: ,in, The normalized signal value, with a range of (0, 1); This is the original value of the current signal. The minimum value of the signal. This represents the maximum value of the signal. Mean based on historical data and standard deviation The data is filtered, and data that meets the following criteria is marked as anomalies: ， in, This is the original value of the current signal. The historical average value of the signal. The standard deviation of the signal. This is the threshold for anomaly detection.

3. The method for safety monitoring of steel structures of offshore substations based on artificial intelligence according to claim 2, characterized in that, In step 3, the edge computing device extracts features from the marked abnormal data and transmits it to the cloud in real time, while storing the normal data locally.

4. The method for safety monitoring of steel structures of offshore substations based on artificial intelligence according to claim 1, characterized in that, In step 4, the gradient-enhanced decision tree is based on the following basic formula of an ensemble learning algorithm that combines an additive model, forward stepwise optimization, and decision trees: Additive model representation: , in, It is the output of the ensemble model. It is the total number of decision trees. This represents the prediction result for the m-th tree; Objective function: , in, L is the regularization term, and L is the objective function of the gradient boosting decision tree. It is the true value of the i-th sample. It is the input feature of the i-th sample. It is the first Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, Represents the loss function; Loss function expansion, gradient update part: ， in, It is a first-order gradient. It is a second-order gradient. It is the true value of the i-th sample. It is the input feature of the i-th sample. It is the first Wheel model for samples The predicted value, It is the m-th decision tree for the sample The incremental forecast value, It is a loss function; Regularization term: , in, It is the number of leaf nodes. It is the leaf node weight. and It is the regularization coefficient. It is a regularization term; Node splitting gain: , Where I is the sample set of the current node, and This represents the set of left and right node samples after the split. This represents the gain from node splitting. It is the sum of the first-order gradients of the current node sample. It is the sum of the second-order gradients of the current node sample. and It is the regularization coefficient.

5. The method for safety monitoring of steel structures of offshore substations based on artificial intelligence according to claim 4, characterized in that, The multimodal data fusion in step 4 uses the following deep neural network formula: ， in, For multimodal data input, This is the weight matrix. For bias terms, This is the activation function.

6. The method for safety monitoring of steel structures of offshore substations based on artificial intelligence according to claim 1, characterized in that, The digital twin model established in step 5 is based on the finite element analysis of the steel structure, updates the geometric parameters and physical state of the model in real time, and dynamically simulates the structural response under extreme load conditions. Based on the finite element method, the stress distribution in key components is calculated: ， in, For stress in critical parts, For the action force, This represents the cross-sectional area of ​​the stressed part; The dynamic stress update formula, combined with real-time data collected by sensors, updates the stress state of the steel structure under dynamic loads: ， in, For real-time dynamic stress, The initial static stress, For dynamic stress changes; Crack propagation prediction formula, based on crack propagation theory, predicts the crack propagation rate of steel structures: ， in, The crack growth rate caused by each cyclic load; , These are material property constants. The range of stress intensity factor is given by the following formula: ， in, For crack geometry factor, The length of the crack. This refers to dynamic stress changes.

7. An AI-based safety monitoring system for the steel structure of an offshore substation, characterized in that, The system is based on the artificial intelligence-based safety monitoring method for steel structures of offshore substations as described in claim 1, and includes: The data acquisition module collects stress, vibration, corrosion, crack and environmental data of the steel structure through multimodal sensors deployed on the steel structure of the offshore substation, and performs time synchronization of the multimodal data through a time synchronization protocol. The data preprocessing module performs signal denoising and normalization on the collected multimodal data, and filters and marks abnormal data by setting an anomaly threshold. The edge computing module performs feature extraction and preliminary anomaly detection on the processed multimodal data on the edge computing device, and the filtered abnormal data is uploaded to the cloud analysis system in real time. The artificial intelligence analysis module integrates and analyzes uploaded abnormal data and related multimodal data through a cloud-based analysis system, and constructs an anomaly detection module and a trend prediction module based on a deep learning model to predict the crack propagation and corrosion development trends of steel structures. The digital twin model building module, combined with the results of artificial intelligence analysis, establishes and dynamically updates a digital twin model based on the initial finite element model of the steel structure, which is used to simulate the health status of the steel structure under extreme load conditions. The intelligent early warning and maintenance module calculates a health score based on a digital twin model and artificial intelligence analysis results, triggers tiered early warnings, and generates optimized predictive maintenance plans.

Images