A deep learning-based navigation facility fatigue life assessment system

Abstract

This invention discloses a deep learning-based fatigue life assessment system for general aviation facilities, comprising the following modules: a data acquisition and preprocessing module for acquiring multi-dimensional monitoring data of various components of the general aviation facility and performing preprocessing; a data dimensionality reduction module for applying the Isomap algorithm for dimensionality reduction; a facility status quantification module for performing quantification processing and generating a facility status quantification feature set; a crack propagation path map generation module for generating a crack propagation path map based on the Gromov-Wasserstein distance; a fatigue life assessment module for performing fatigue life assessment based on crack propagation characteristics and the structural characteristics of the facility; a risk assessment module for performing risk assessment on the fatigue life assessment result set; and a result visualization module for presenting the fatigue life assessment result set through a graphical interface. This invention combines multi-dimensional data with deep learning to accurately assess the fatigue life and crack propagation of general aviation facilities.

Description

Technical Field

[0001] This invention relates to the field of facility health monitoring and assessment technology, and in particular to a deep learning-based fatigue life assessment system for general aviation facilities. Background Technology

[0002] With the rapid development of modern infrastructure, especially in the transportation sector, the construction and maintenance of navigation facilities (such as bridges, tunnels, and docks) has become a crucial component of engineering management. Particularly during long-term use, these facilities are subjected to frequent external loads, climate change, and other environmental factors, making them prone to cracking and fatigue damage. This poses a significant threat to the safety, stability, and service life of these facilities. Therefore, accurately predicting the fatigue life of facilities, promptly identifying potential damage, and preventing structural failures and accidents have become major challenges in current facility maintenance and management.

[0003] Existing fatigue life assessment methods primarily rely on traditional mechanical models, empirical formulas, and periodic manual inspections. These methods typically calculate crack propagation and fatigue damage based on models such as finite element analysis (FEA) or stress-life curves (SN curves). However, these traditional methods have significant limitations. First, while finite element analysis can provide accurate local analysis, it usually requires substantial computational resources and time, making real-time analysis difficult in practical applications. Furthermore, traditional crack propagation prediction methods often rely on simplified assumptions and theories, failing to adequately consider the complex factors in actual operating environments, such as dynamic loads and changing environmental conditions. Second, traditional manual inspections depend on manual checks and periodic structural assessments, but these often rely on the experience and skill level of the inspectors and cannot achieve real-time, accurate monitoring throughout the facility's entire lifecycle.

[0004] To improve the accuracy and efficiency of facility fatigue life assessment, intelligent monitoring technologies based on machine learning and deep learning have been increasingly introduced into facility management in recent years. Deep learning technology can automatically learn more complex features from large amounts of sensor data, thereby compensating for the shortcomings of traditional methods in prediction accuracy and real-time performance. Existing deep learning methods often focus on single types of monitoring data (such as temperature and strain) and fail to fully consider the fusion of data from multiple dimensions, such as the influence of environmental impacts and historical usage conditions. This results in assessment models lacking a comprehensive analysis of facility fatigue damage under complex and variable environmental conditions.

[0005] Therefore, how to provide a fatigue life assessment system for general aviation facilities based on deep learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a deep learning-based fatigue life assessment system for general aviation facilities. This invention combines the Isomap algorithm with the Gromov-Wasserstein distance to accurately assess the fatigue life of general aviation facilities. Through multi-dimensional data integration, crack propagation path simulation, and fatigue damage assessment, it provides more accurate crack propagation prediction and fatigue life assessment. The system achieves automated analysis and real-time monitoring, supports risk assessment and visualization output, provides a scientific basis for facility maintenance, and improves management efficiency and facility safety.

[0007] A fatigue life assessment system for general aviation facilities based on deep learning according to an embodiment of the present invention includes the following modules:

[0008] The data acquisition and preprocessing module is used to collect multi-dimensional monitoring data of various components of the navigation facility in real time through sensors and to perform preprocessing.

[0009] The data dimensionality reduction module is used to apply the Isomap algorithm to reduce the dimensionality of preprocessed, unified-structure, multi-dimensional monitoring data.

[0010] The facility status quantization module is used to extract the facility status feature set from the low-dimensional manifold graph, perform quantization processing, and generate a facility status quantization feature set.

[0011] The crack propagation path map generation module is used to calculate the similarity between the state feature space and structural similarity space of component nodes based on the Gromov-Wasserstein distance, and generate a crack propagation path map. Specifically:

[0012] Based on the physical structure of the general aviation facility and the connection methods between its components, a facility component association diagram is established. Each node in the facility component association diagram corresponds to a general aviation facility component. The characteristics of the node include the component's geometry, material properties, and load-bearing capacity. Each edge of the facility component association diagram represents the physical connection relationship between the components.

[0013] Based on the facility status quantification feature set, each component node in the facility component association diagram is mapped to a metric space. The metric space includes the status feature space of the general aviation facility components and the structural similarity space of the general aviation facility components. The structural similarity space is a space defined by the physical connection relationship and mechanical properties between the general aviation facility components.

[0014] A similarity matrix is ​​constructed by calculating the Gromov-Wasserstein distance between the state feature space and the structural similarity space of each component node;

[0015] The initial location of the crack is set according to the state type, and the component where the crack originates is taken as the starting point for crack propagation.

[0016] Based on the values ​​in the similarity matrix, calculate the propagation probability of the crack from the source component to the adjacent component, and select the component with the highest propagation probability as the next target of the crack.

[0017] During crack propagation, record all components from the source component to the source component, and record the propagation path at each step. For each propagation step, record the time step and propagation path of crack propagation.

[0018] If the probability of a crack propagating at a certain component is lower than a set threshold or the current component reaches the preset maximum damage limit, the simulation propagation will be interrupted and the crack propagation will stop.

[0019] A crack propagation path map is generated based on the crack propagation path sequence. In the crack propagation path map, nodes represent various components of the navigation facility, edges represent the crack propagation path, and the weight of the edge is the crack propagation probability.

[0020] The fatigue life assessment module is used to obtain crack propagation characteristics based on crack propagation path maps, and combined with the structural characteristics of the navigation facility, to conduct fatigue life assessment using SN curves, generating a fatigue life assessment result set for each component of the navigation facility.

[0021] The risk assessment module is used to analyze the fatigue life assessment result set and conduct risk assessments of navigation facilities.

[0022] The results visualization module is used to present the fatigue life assessment results set through a graphical interface, showing the remaining fatigue life, crack propagation path, crack depth, and propagation rate of each component.

[0023] A method for assessing the fatigue life of general aviation facilities based on deep learning according to an embodiment of the present invention includes the following steps:

[0024] Step 1: Collect multi-dimensional monitoring data of navigation facilities through sensors, and preprocess the multi-dimensional monitoring data to generate multi-dimensional monitoring data with a unified structure;

[0025] Step 2: Apply the Isomap algorithm to reduce the dimensionality of the multi-dimensional monitoring data with a unified structure, and construct a low-dimensional manifold map of the navigation facilities;

[0026] Step 3: Based on the low-dimensional manifold map, extract the navigation facility status feature set, and perform quantization processing to generate a facility status quantization feature set;

[0027] Step 4: Calculate the similarity between the state feature space and the structural similarity space of each navigation facility component node in the metric space using the Gromov-Wasserstein distance, and simulate the crack propagation path inside the navigation facility to generate a crack propagation path map;

[0028] Step 5: Obtain crack propagation characteristics based on crack propagation path map, perform fatigue life assessment, and generate fatigue life assessment result set;

[0029] Step 6: Visualize the fatigue life assessment results set.

[0030] Optionally, the multi-dimensional monitoring data includes strain response, vibration frequency, temperature, and environmental data of various components of the navigation facility; the preprocessing steps include using Gaussian filtering to remove abnormal data caused by sensor noise and environmental interference, filling missing data that occurred during the acquisition process with mean interpolation, and performing Min-Max normalization to generate multi-dimensional monitoring data with a unified structure.

[0031] Optionally, the application of the Isomap algorithm to reduce the dimensionality of multi-dimensional monitoring data with a unified structure, and to construct a low-dimensional manifold map of navigation facilities, specifically:

[0032] Calculate the Euclidean distance of each pair of data points in high-dimensional space and construct an adjacency matrix. The Euclidean distance measures the direct spatial distance between two data points. The adjacency matrix records the Euclidean distance between each pair of data points.

[0033] The Isomap algorithm transforms the adjacency matrix into a manifold graph using a minimum spanning tree. In this manifold graph, nodes represent data points, and edges represent the Euclidean distances between data points.

[0034] The shortest path between each pair of nodes in the manifold graph is calculated using Dijkstra's algorithm, and the manifold graph is mapped from a high-dimensional space to a low-dimensional space using principal component analysis to generate a low-dimensional manifold graph.

[0035] Optionally, the step of extracting a navigation facility status feature set based on a low-dimensional manifold map and performing quantization processing to generate a facility status quantized feature set specifically involves:

[0036] Extract the state feature set of navigation facilities from the low-dimensional manifold graph. The state feature set includes the dynamic response characteristics, environmental impact characteristics and historical usage characteristics of the navigation facilities.

[0037] The dynamic response characteristics of the navigation facilities include vibration modes, strain change rates, and displacement changes; the environmental impact characteristics include temperature changes, humidity fluctuations, and air pressure changes; and the historical usage characteristics include the usage frequency and operating cycle of the navigation facilities.

[0038] For time series and discrete data in the state feature set, the time window division method is used to extract the statistical features within each time window and convert them into time numerical data; for classification features, the label encoding method is used to convert them into classification feature numerical data, and the classification features include navigation facility category and state type.

[0039] The label encoding step includes: for the category and status type of navigation facility, mapping each category to an integer in order of frequency of occurrence, and replacing the original classification data with the corresponding integer to obtain numerical data;

[0040] The transformed time-based numerical data and classification feature numerical data are combined to generate a quantitative feature set of facility status.

[0041] Optionally, the step of calculating the similarity between the state feature space and the structural similarity space of each navigation facility component node in the metric space using the Gromov-Wasserstein distance, and simulating the crack propagation path within the navigation facility to generate a crack propagation path map, specifically involves:

[0042] Based on the physical structure of the general aviation facility and the connection methods between its components, a facility component association diagram is established. Each node in the facility component association diagram corresponds to a general aviation facility component. The characteristics of the node include the component's geometry, material properties, and load-bearing capacity. Each edge of the facility component association diagram represents the physical connection relationship between the components.

[0043] Based on the facility status quantification feature set, each component node in the facility component association diagram is mapped to a metric space. The metric space includes the status feature space of the general aviation facility components and the structural similarity space of the general aviation facility components. The structural similarity space is a space defined by the physical connection relationship and mechanical properties between the general aviation facility components.

[0044] A similarity matrix is ​​constructed by calculating the Gromov-Wasserstein distance between the state feature space and the structural similarity space of each component node;

[0045] The initial location of the crack is set according to the state type, and the component where the crack originates is taken as the starting point for crack propagation.

[0046] Based on the values ​​in the similarity matrix, calculate the propagation probability of the crack from the source component to the adjacent component, and select the component with the highest propagation probability as the next target of the crack.

[0047] During crack propagation, record all components from the source component to the source component, and record the propagation path at each step. For each propagation step, record the time step and propagation path of crack propagation.

[0048] If the probability of a crack propagating at a certain component is lower than a set threshold or the current component reaches the preset maximum damage limit, the simulation propagation will be interrupted and the crack propagation will stop.

[0049] A crack propagation path map is generated based on the crack propagation path sequence. In the crack propagation path map, nodes represent various components of the navigation facility, edges represent crack propagation paths, and the weight of the edges is the crack propagation probability.

[0050] Optionally, the step of obtaining crack propagation characteristics based on the crack propagation path map to perform fatigue life assessment and generate a fatigue life assessment result set specifically involves:

[0051] Crack propagation features and structural characteristics of navigation facilities are extracted from crack propagation path maps. The crack propagation features include crack propagation rate, crack propagation depth, and crack propagation path. The crack propagation rate is obtained by recording the crack propagation time and propagation path length. The structural characteristics of navigation facilities include material strength, fatigue limit, stiffness, and elastic modulus. The fatigue limit is the maximum stress that the material can withstand.

[0052] Based on crack propagation rate, crack propagation depth, crack propagation path, and structural characteristics of navigation facilities, the fatigue life of navigation facilities is evaluated using stress-life curves.

[0053] The stress-life curve assessment step includes: for each navigation facility component, calculating the fatigue damage of the crack to the navigation facility based on the crack propagation rate and the stress intensity factor range of the component, and estimating the fatigue life of each stage in the crack propagation process using the stress-life curve by calculating the crack depth and the corresponding stress concentration effect, and generating a fatigue life assessment result set.

[0054] The fatigue life assessment result set includes fatigue life prediction results, crack propagation path, depth and propagation rate, and overall fatigue damage for each component.

[0055] Optionally, the visualization output of the fatigue life assessment result set is specifically as follows:

[0056] The graphical interface displays the remaining fatigue life of each navigation facility component, with a bar chart representing the length of the fatigue life, and shows the path of the crack from its source to the component, as well as the extent of crack propagation in each component.

[0057] The depth of the crack in the component is indicated by the shade of color, and the speed bar indicates the rate of crack propagation.

[0058] Based on crack propagation and fatigue life assessments, risk assessments are conducted on navigation facilities. Components with risks exceeding a preset threshold are highlighted.

[0059] The beneficial effects of this invention are:

[0060] 1. The deep learning-based fatigue life assessment system for general aviation facilities provided by this invention can effectively improve the accuracy of crack propagation and fatigue life assessment. By integrating and processing multi-dimensional monitoring data (such as strain response, vibration frequency, temperature, environmental data, etc.), the system can comprehensively analyze the health status of the facility, avoiding the shortcomings of traditional methods that rely on only a single data source. Combining the Isomap algorithm for data dimensionality reduction effectively reduces data complexity while retaining key features, thereby improving the accuracy of fatigue life prediction.

[0061] 2. Using Gromov-Wasserstein distance for component similarity calculation enables accurate simulation of crack propagation paths within a facility. Traditional crack propagation simulations often rely on simple assumptions and local models, which are often inadequate for complex facility structures and real-world application environments. This invention, by calculating the similarity between the state feature space and structural similarity space of facility component nodes, can more accurately describe the dynamic process of crack propagation and update the crack propagation path map in real time, thus aiding in the assessment of facility fatigue life.

[0062] 3. This invention effectively integrates crack propagation characteristics with structural characteristic data of the facility by combining SN curves, further improving the accuracy of fatigue life assessment. Traditional fatigue life assessment methods mostly ignore dynamic factors such as crack propagation path, propagation rate, and depth. However, this invention, based on crack propagation characteristics, utilizes an accurate damage accumulation model to comprehensively predict the remaining life of each component of the facility, avoiding the problem of large estimation errors in traditional methods.

[0063] 4. The visualization output module of this invention can display key information such as fatigue life prediction results, crack propagation paths, and crack depths of various components in an intuitive graphical interface. Users can easily identify high-risk components and quickly obtain the specific crack propagation path, providing a scientific basis for maintenance decisions. This enables facility managers to track the crack propagation process and the health status of the facility in real time. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a schematic diagram of the structure of a deep learning-based fatigue life assessment system for general aviation facilities proposed in this invention.

[0066] Figure 2This is an overall flowchart of a deep learning-based fatigue life assessment method for general aviation facilities proposed in this invention. Detailed Implementation

[0067] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0068] refer to Figure 1 A deep learning-based fatigue life assessment system for general aviation facilities includes the following modules:

[0069] The data acquisition and preprocessing module is used to collect multi-dimensional monitoring data of various components of the navigation facility in real time through sensors and to perform preprocessing.

[0070] The data dimensionality reduction module is used to apply the Isomap algorithm to reduce the dimensionality of preprocessed, unified-structure, multi-dimensional monitoring data.

[0071] The facility status quantization module is used to extract the facility status feature set from the low-dimensional manifold graph, perform quantization processing, and generate a facility status quantization feature set.

[0072] The crack propagation path map generation module is used to calculate the similarity between the state feature space and structural similarity space of component nodes based on the Gromov-Wasserstein distance, and generate a crack propagation path map. Specifically:

[0073] Based on the physical structure of the general aviation facility and the connection methods between its components, a facility component association diagram is established. Each node in the facility component association diagram corresponds to a general aviation facility component. The characteristics of the node include the component's geometry, material properties, and load-bearing capacity. Each edge of the facility component association diagram represents the physical connection relationship between the components.

[0074] Based on the facility status quantification feature set, each component node in the facility component association diagram is mapped to a metric space. The metric space includes the status feature space of the general aviation facility components and the structural similarity space of the general aviation facility components. The structural similarity space is a space defined by the physical connection relationship and mechanical properties between the general aviation facility components.

[0075] A similarity matrix is ​​constructed by calculating the Gromov-Wasserstein distance between the state feature space and the structural similarity space of each component node;

[0076] The initial location of the crack is set according to the state type, and the component where the crack originates is taken as the starting point for crack propagation.

[0077] Based on the values ​​in the similarity matrix, calculate the probability of crack propagation from the source component to adjacent components:

[0078] ;

[0079] in, Let be the probability of a crack propagating from navigation facility component i to navigation facility component j. is a positive constant used to control the degree to which similarity affects the propagation probability. Let the Gromov-Wasserstein distance be the distance between components i and j. Let i be the set of neighborhood nodes of navigation facility component i, which includes all components directly connected to component i;

[0080] Select the component with the highest probability of propagation as the next target for the crack;

[0081] During crack propagation, record all components from the source component to the source component, and record the propagation path at each step. For each propagation step, record the time step and propagation path of crack propagation.

[0082] If the probability of a crack propagating at a certain component is lower than a set threshold or the current component reaches the preset maximum damage limit, the simulation propagation will be interrupted and the crack propagation will stop.

[0083] A crack propagation path map is generated based on the crack propagation path sequence. In the crack propagation path map, nodes represent various components of the navigation facility, edges represent the crack propagation path, and the weight of the edge is the crack propagation probability.

[0084] The fatigue life assessment module is used to obtain crack propagation characteristics based on crack propagation path maps, and combined with the structural characteristics of the facility, to conduct fatigue life assessment using SN curves, generating a fatigue life assessment result set for each component of the facility.

[0085] The risk assessment module is used to analyze the fatigue life assessment result set, assess the fatigue damage and remaining life of each component in the facility, identify components with more severe crack propagation, conduct risk assessment of the facility, and help determine the components that need priority maintenance or replacement.

[0086] The results visualization module is used to present the fatigue life assessment results set through a graphical interface, showing the remaining fatigue life, crack propagation path, crack depth, and propagation rate of each component.

[0087] This invention utilizes data acquisition and preprocessing modules to enable the system to collect and clean multi-dimensional data in real time, providing high-quality input for subsequent analysis. The application of the Isomap algorithm effectively reduces the dimensionality of complex data, extracts key features of facility status, and generates a quantitative feature set of facility status through quantification processing, further improving the accuracy of the assessment. In the crack propagation path map generation module, by combining the Gromov-Wasserstein distance, the system can accurately simulate the crack propagation path within the facility, providing a true reflection of facility damage. By combining SN curves for fatigue life assessment, the system can provide accurate life predictions for each component. Simultaneously, the risk assessment and result visualization modules provide users with clear maintenance decision support, ensuring the safe operation and efficient management of the facility.

[0088] refer to Figure 2 A method for assessing the fatigue life of general aviation facilities based on deep learning includes the following steps:

[0089] Step 1: Collect multi-dimensional monitoring data of navigation facilities through sensors, and preprocess the multi-dimensional monitoring data to generate multi-dimensional monitoring data with a unified structure;

[0090] Step 2: Apply the Isomap algorithm to reduce the dimensionality of the multi-dimensional monitoring data with a unified structure, and construct a low-dimensional manifold map of the navigation facilities;

[0091] Step 3: Based on the low-dimensional manifold map, extract the navigation facility status feature set, and perform quantization processing to generate a facility status quantization feature set;

[0092] Step 4: Calculate the similarity between the state feature space and the structural similarity space of each navigation facility component node in the metric space using the Gromov-Wasserstein distance, and simulate the crack propagation path inside the navigation facility to generate a crack propagation path map;

[0093] Step 5: Obtain crack propagation characteristics based on crack propagation path map, perform fatigue life assessment, and generate fatigue life assessment result set;

[0094] Step 6: Visualize the fatigue life assessment results set.

[0095] This invention provides an efficient and accurate method for assessing facility fatigue life by integrating multi-dimensional monitoring data, manifold learning algorithms, and crack propagation path simulation. Compared with traditional methods, the system can collect and preprocess facility status data in real time, generating data with a uniform structure, avoiding the time delay and subjective errors of manual inspection. By applying the Isomap algorithm to reduce the dimensionality of the data, key facility status features are successfully extracted. Similarity calculations based on Gromov-Wasserstein distance and crack propagation path simulation effectively improve the accuracy of fatigue life prediction. Furthermore, the system provides intuitive visualization of fatigue life predictions, crack propagation paths, and related risks for each component, offering real-time and accurate data support for facility management and helping maintenance personnel make informed decisions.

[0096] In this embodiment, the multi-dimensional monitoring data includes strain response, vibration frequency, temperature and environmental data of various components of the navigation facility; the preprocessing steps include using Gaussian filtering to remove abnormal data caused by sensor noise and environmental interference, filling missing data that occurred during the acquisition process by mean interpolation, and performing Min-Max normalization to generate multi-dimensional monitoring data with a unified structure.

[0097] This step overcomes the problems of incomplete data and severe noise interference in traditional methods by acquiring and preprocessing multi-dimensional monitoring data of navigation facilities in real time through sensors. Data acquisition employs Gaussian filtering to remove noise, mean imputation to fill missing data, and Min-Max normalization to ensure data consistency and accuracy. The uniformly structured data provides a reliable foundation for subsequent analysis and modeling, especially for complex multi-dimensional data, reducing the impact of data noise on the analysis results. Simultaneously, this preprocessing step enables the system to efficiently handle various data types from different sensors, improving the model's generalization ability and enhancing the accuracy and reliability of the evaluation results.

[0098] In this embodiment, the application of the Isomap algorithm to reduce the dimensionality of multi-dimensional monitoring data with a unified structure and construct a low-dimensional manifold map of navigation facilities specifically involves:

[0099] Calculate the Euclidean distance of each pair of data points in high-dimensional space and construct an adjacency matrix. The Euclidean distance measures the direct spatial distance between two data points. The adjacency matrix records the Euclidean distance between each pair of data points.

[0100] The Isomap algorithm transforms the adjacency matrix into a manifold graph using a minimum spanning tree. In this manifold graph, nodes represent data points, and edges represent the Euclidean distances between data points.

[0101] The shortest path between each pair of nodes in the manifold graph is calculated using Dijkstra's algorithm, and the manifold graph is mapped from a high-dimensional space to a low-dimensional space using principal component analysis to generate a low-dimensional manifold graph.

[0102] This step applies the Isomap algorithm for data dimensionality reduction, enabling complex multi-dimensional monitoring data to be effectively represented in a lower-dimensional space, thus avoiding the computational difficulties and overfitting problems that may occur with high-dimensional data. By calculating the Euclidean distance of each pair of data points and constructing an adjacency matrix, the Isomap algorithm preserves the relative structural relationships between data points, and then transforms them into a manifold graph through a minimum spanning tree. The shortest path is calculated using Dijkstra's algorithm, and principal component analysis is used to map the manifold graph from high-dimensional space to low-dimensional space, greatly improving the data's processability. The low-dimensional manifold graph helps extract important features of facility health status, providing an accurate and compact feature representation for subsequent fatigue life assessment.

[0103] In this embodiment, the step of extracting the navigation facility status feature set based on the low-dimensional manifold map and performing quantization processing to generate a facility status quantized feature set specifically involves:

[0104] Extract the state feature set of navigation facilities from the low-dimensional manifold graph. The state feature set includes the dynamic response characteristics, environmental impact characteristics and historical usage characteristics of the navigation facilities.

[0105] The dynamic response characteristics of the navigation facilities include vibration modes, strain change rates, and displacement changes; the environmental impact characteristics include temperature changes, humidity fluctuations, and air pressure changes; and the historical usage characteristics include the usage frequency and operating cycle of the navigation facilities.

[0106] For time series and discrete data in the status feature set, the time window division method is used to extract the statistical features within each time window and convert them into time numerical data; for classification features, the label encoding method is used to convert them into classification feature numerical data. The classification features include navigation facility category and status type. The navigation facility category includes bridges, tunnels, docks, etc., and the status type includes good, minor damage, serious damage, and under repair.

[0107] The label encoding step includes: for the category and status type of navigation facility, mapping each category to an integer in order of frequency of occurrence, and replacing the original classification data with the corresponding integer to obtain numerical data;

[0108] The transformed time-based numerical data and classification feature numerical data are combined to generate a quantitative feature set of facility status.

[0109] This step extracts facility status feature sets based on low-dimensional manifold maps and quantifies them, enabling the system to comprehensively analyze the dynamic response characteristics, environmental impacts, and historical usage status characteristics of navigation facilities. This method further optimizes the data representation capability by dividing time-series and discrete data into time windows and extracting statistical features within each window. The label encoding method transforms categorical features such as facility category and status type into numerical data, effectively providing standardized input for subsequent deep learning models. Furthermore, the system combines the quantified facility status feature set to comprehensively assess facility health status, providing accurate data support for crack propagation simulation and fatigue life prediction, thus improving prediction accuracy and reliability.

[0110] In this embodiment, the step of obtaining crack propagation characteristics based on the crack propagation path map, performing fatigue life assessment, and generating a fatigue life assessment result set specifically involves:

[0111] Crack propagation features and structural characteristics of navigation facilities are extracted from crack propagation path maps. The crack propagation features include crack propagation rate, crack propagation depth, and crack propagation path. The crack propagation rate is obtained by recording the crack propagation time and propagation path length. The structural characteristics of navigation facilities include material strength, fatigue limit, stiffness, and elastic modulus. The fatigue limit is the maximum stress that the material can withstand.

[0112] Based on crack propagation rate, crack propagation depth, crack propagation path, and structural characteristics of navigation facilities, the fatigue life of navigation facilities is evaluated using stress-life curves.

[0113] The stress-life curve assessment step includes: for each navigation facility component, calculating the fatigue damage of the crack to the navigation facility based on the crack propagation rate and the stress intensity factor range of the component, and estimating the fatigue life of each stage in the crack propagation process using the stress-life curve by calculating the crack depth and the corresponding stress concentration effect, and generating a fatigue life assessment result set.

[0114] The fatigue life assessment result set includes fatigue life prediction results, crack propagation path, depth and propagation rate, and overall fatigue damage for each component.

[0115] This step obtains crack propagation characteristics and performs fatigue life assessment through crack propagation path mapping. This invention combines crack propagation rate, depth, and propagation path with SN curves for fatigue damage assessment, achieving a comprehensive analysis of crack propagation and facility fatigue life. Compared to traditional methods, accurate calculation of crack propagation characteristics (such as crack propagation rate and depth) more intuitively reflects the fatigue condition of the facility, improving the accuracy of life assessment. By combining structural characteristics of the facility (such as material strength and fatigue limit) for comprehensive calculation, the system can provide more accurate fatigue life prediction based on crack characteristics. Simultaneously, the fatigue life assessment result set output by the system provides data support for facility managers, helping them develop more scientific maintenance and upgrade plans.

[0116] In this embodiment, the visualization output of the fatigue life assessment result set specifically includes:

[0117] The graphical interface displays the remaining fatigue life of each navigation facility component, with a bar chart representing the length of the fatigue life, and shows the path of the crack from its source to the component, as well as the extent of crack propagation in each component.

[0118] The depth of the crack in the component is indicated by the shade of color, and the speed bar indicates the rate of crack propagation.

[0119] Based on crack propagation and fatigue life assessments, risk assessments are conducted on navigation facilities. Components with risks exceeding a preset threshold are highlighted.

[0120] This step, in visualizing the fatigue life assessment results, provides users with clear and intuitive fatigue life predictions, crack propagation paths, and other key assessment data through a graphical interface. Users can view the facility's fatigue status in real time and track the crack propagation process. Through color coding, speed bars, and other methods, the crack propagation depth, rate, and risk level are readily apparent. The system can also highlight components with more severe crack propagation, helping users optimize maintenance and replacement strategies, significantly improving the accuracy and efficiency of facility management, and ensuring the facility's safety and long-term stable operation.

[0121] Example 1:

[0122] To verify the feasibility of this invention in practice, it was applied to a fatigue life assessment task for facilities at a large port terminal. This task involved assessing crack propagation and fatigue damage in several key facility components (such as hoisting equipment, rails, and support columns). The goal was to monitor and evaluate the health status of each component in real time, particularly the impact of crack propagation on the facility's fatigue life, thereby providing data support for equipment maintenance and management.

[0123] Traditional facility inspection methods rely on periodic manual inspections and simple stress-life curve models. These methods have limitations, including the inability to detect cracks in real time and difficulty in accurately assessing crack propagation paths and fatigue damage accumulation. Especially during crack propagation, the inability to comprehensively assess the fatigue life of components in a timely manner leads to delays in maintenance and may cause missed opportunities for optimal maintenance. Traditional models often fail to provide sufficiently accurate assessments when dealing with complex dynamic loads, environmental influences, and historical usage data, resulting in high rates of missed and false detections.

[0124] In this embodiment, the system first collects multi-dimensional monitoring data of the dock facilities through sensors, covering information such as strain response, vibration frequency, temperature change, and humidity fluctuation of hoisting equipment, rails, and support columns. The data collection period is 90 consecutive days, the sampling frequency of each device is 1Hz, and the total data volume is approximately 4TB, including data from various sensors, including structured time series data, image data, and a small amount of operator records.

[0125] In the data preprocessing stage, the system performed Gaussian filtering to denoise the collected data, used mean interpolation to fill in missing values, and standardized various data types through Min-Max normalization to ensure data quality and consistency. In the feature extraction stage, the system applied the Isomap algorithm to reduce the dimensionality of the multi-dimensional monitoring data, constructing a low-dimensional manifold map of the terminal facilities. By calculating the Euclidean distance between each pair of data points, the dimensionality was successfully reduced to 3D space. Subsequently, the system extracted the facility's state feature set based on the low-dimensional manifold map. These feature sets include the facility's dynamic response characteristics (such as vibration modes, strain change rates, and displacement changes), environmental impact characteristics (such as temperature changes and humidity fluctuations), and historical usage characteristics (such as usage frequency and operating cycles). The system used a label encoding method to convert categorical features (such as facility category and state type) into numerical data, generating a quantitative feature set of the facility's state, providing accurate data input for subsequent crack propagation path simulation and fatigue life assessment.

[0126] In the crack propagation path map generation module, the system calculates the similarity between the state feature space and structural similarity space of each component node in the facility using the Gromov-Wasserstein distance, thereby simulating the crack propagation path within the facility. The crack propagates from a source component of the facility. The system calculates the propagation probability of the crack from the source component to adjacent components based on the similarity matrix and selects the component with the highest propagation probability as the next target of the crack. This ultimately generates a crack propagation path map, providing an accurate crack propagation process for fatigue life assessment. Fatigue life assessment is then performed using SN curves and a fatigue damage accumulation model to evaluate the crack propagation characteristics.

[0127] Table 1. Comparison of key indicators between traditional methods and the present invention in facility fatigue life assessment.

[0128]

[0129] As shown in Table 1, the fatigue life of the hoisting equipment is 4 years, the fatigue life of the rails is 8 years, and the fatigue life of the support columns is 6 years. Compared with the three common traditional methods, the present invention significantly improves the accuracy of fatigue life assessment. Taking hoisting equipment A as an example, the traditional stress-life curve method assesses its fatigue life as 2 years, while the present invention assesses it as 4 years, an improvement of 2 years, and significantly reduces the false alarm and missed alarm rates. In terms of risk assessment and early warning, the present invention can provide early warnings within a longer time window, helping port managers to carry out timely facility maintenance and avoid the risk of premature equipment failure.

[0130] Finally, in the results visualization module, the system presents the fatigue life assessment results in a graphical interface, using color coding to display the remaining life and crack propagation path of each component. For example, the remaining life of hoisting equipment A is displayed in red, indicating that its fatigue life is nearing its end; while the remaining life of rail B is displayed in green, indicating that it still has a long service life. Through dynamic graphics and path simulation, port managers can intuitively understand the crack propagation process and the health status of various components of the facility.

[0131] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fatigue life assessment system for general aviation facilities based on deep learning, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect multi-dimensional monitoring data of various components of the navigation facility in real time through sensors and to perform preprocessing. The data dimensionality reduction module is used to apply the Isomap algorithm to reduce the dimensionality of preprocessed, unified-structure, multi-dimensional monitoring data. The facility status quantization module is used to extract the facility status feature set from the low-dimensional manifold graph, perform quantization processing, and generate a facility status quantization feature set. The crack propagation path map generation module is used to calculate the similarity between the state feature space and structural similarity space of component nodes based on the Gromov-Wasserstein distance, and generate a crack propagation path map. Specifically: Based on the physical structure of the general aviation facility and the connection methods between its components, a facility component association diagram is established. Each node in the facility component association diagram corresponds to a general aviation facility component. The characteristics of the node include the component's geometry, material properties, and load-bearing capacity. Each edge of the facility component association diagram represents the physical connection relationship between the components. Based on the facility status quantification feature set, each component node in the facility component association diagram is mapped to a metric space. The metric space includes the status feature space of the general aviation facility components and the structural similarity space of the general aviation facility components. The structural similarity space is a space defined by the physical connection relationship and mechanical properties between the general aviation facility components. A similarity matrix is ​​constructed by calculating the Gromov-Wasserstein distance between the state feature space and the structural similarity space of each component node; The initial location of the crack is set according to the state type, and the component where the crack originates is taken as the starting point for crack propagation. Based on the values ​​in the similarity matrix, calculate the propagation probability of the crack from the source component to the adjacent component, and select the component with the highest propagation probability as the next target of the crack. During crack propagation, record all components from the source component to the source component, and record the propagation path at each step. For each propagation step, record the time step and propagation path of crack propagation. If the probability of a crack propagating at a certain component is lower than a set threshold or the current component reaches the preset maximum damage limit, the simulation propagation will be interrupted and the crack propagation will stop. A crack propagation path map is generated based on the crack propagation path sequence. In the crack propagation path map, nodes represent various components of the navigation facility, edges represent the crack propagation path, and the weight of the edge is the crack propagation probability. The fatigue life assessment module is used to obtain crack propagation characteristics based on crack propagation path maps, and combined with the structural characteristics of the navigation facility, to conduct fatigue life assessment using SN curves, generating a fatigue life assessment result set for each component of the navigation facility. The risk assessment module is used to analyze the fatigue life assessment result set and conduct risk assessments of navigation facilities. The results visualization module is used to present the fatigue life assessment results set through a graphical interface, showing the remaining fatigue life, crack propagation path, crack depth, and propagation rate of each component.

2. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 1, characterized in that, The modules are connected in the following way: Step 1: Collect multi-dimensional monitoring data of navigation facilities through sensors, and preprocess the multi-dimensional monitoring data to generate multi-dimensional monitoring data with a unified structure; Step 2: Apply the Isomap algorithm to reduce the dimensionality of the multi-dimensional monitoring data with a unified structure, and construct a low-dimensional manifold map of the navigation facilities; Step 3: Based on the low-dimensional manifold map, extract the navigation facility status feature set, and perform quantization processing to generate a facility status quantization feature set; Step 4: Calculate the similarity between the state feature space and the structural similarity space of each navigation facility component node in the metric space using the Gromov-Wasserstein distance, and simulate the crack propagation path inside the navigation facility to generate a crack propagation path map; Step 5: Obtain crack propagation characteristics based on crack propagation path map, perform fatigue life assessment, and generate fatigue life assessment result set; Step 6: Visualize the fatigue life assessment results set.

3. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 2, characterized in that, The multi-dimensional monitoring data includes strain response, vibration frequency, temperature, and environmental data of various components of the navigation facility; the preprocessing steps include using Gaussian filtering to remove abnormal data caused by sensor noise and environmental interference, filling missing data that occurred during the acquisition process with mean interpolation, and performing Min-Max normalization to generate multi-dimensional monitoring data with a unified structure.

4. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 2, characterized in that, The application of the Isomap algorithm to reduce the dimensionality of multi-dimensional monitoring data with a unified structure, and to construct a low-dimensional manifold map of navigation facilities, specifically: Calculate the Euclidean distance of each pair of data points in high-dimensional space and construct an adjacency matrix. The Euclidean distance measures the direct spatial distance between two data points. The adjacency matrix records the Euclidean distance between each pair of data points. The Isomap algorithm transforms the adjacency matrix into a manifold graph using a minimum spanning tree. In this manifold graph, nodes represent data points, and edges represent the Euclidean distances between data points. The shortest path between each pair of nodes in the manifold graph is calculated using Dijkstra's algorithm, and the manifold graph is mapped from a high-dimensional space to a low-dimensional space using principal component analysis to generate a low-dimensional manifold graph.

5. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 2, characterized in that, The process of extracting a set of navigation facility status features based on a low-dimensional manifold map and then quantifying it to generate a quantified feature set of facility status features is as follows: Extract the state feature set of navigation facilities from the low-dimensional manifold graph. The state feature set includes the dynamic response characteristics, environmental impact characteristics and historical usage characteristics of the navigation facilities. The dynamic response characteristics of the navigation facilities include vibration modes, strain change rates, and displacement changes; the environmental impact characteristics include temperature changes, humidity fluctuations, and air pressure changes; and the historical usage characteristics include the usage frequency and operating cycle of the navigation facilities. For time series and discrete data in the state feature set, the time window division method is used to extract the statistical features within each time window and convert them into time numerical data; for classification features, the label encoding method is used to convert them into classification feature numerical data, and the classification features include navigation facility category and state type. The label encoding step includes: for the category and status type of navigation facility, mapping each category to an integer in order of frequency of occurrence, and replacing the original classification data with the corresponding integer to obtain numerical data; The transformed time-based numerical data and classification feature numerical data are combined to generate a quantitative feature set of facility status.

6. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 2, characterized in that, The process of obtaining crack propagation characteristics based on crack propagation path maps and performing fatigue life assessment to generate a fatigue life assessment result set is as follows: Crack propagation features and structural characteristics of navigation facilities are extracted from crack propagation path maps. The crack propagation features include crack propagation rate, crack propagation depth, and crack propagation path. The crack propagation rate is obtained by recording the crack propagation time and propagation path length. The structural characteristics of navigation facilities include material strength, fatigue limit, stiffness, and elastic modulus. The fatigue limit is the maximum stress that the material can withstand. Based on crack propagation rate, crack propagation depth, crack propagation path, and structural characteristics of navigation facilities, the fatigue life of navigation facilities is evaluated using stress-life curves. The stress-life curve assessment step includes: for each navigation facility component, calculating the fatigue damage of the crack to the navigation facility based on the crack propagation rate and the stress intensity factor range of the component, and estimating the fatigue life of each stage in the crack propagation process using the stress-life curve by calculating the crack depth and the corresponding stress concentration effect, and generating a fatigue life assessment result set. The fatigue life assessment result set includes fatigue life prediction results, crack propagation path, depth and propagation rate, and overall fatigue damage for each component.

7. The fatigue life assessment system for general aviation facilities based on deep learning according to claim 2, characterized in that, The visualization output of the fatigue life assessment result set is specifically as follows: The graphical interface displays the remaining fatigue life of each navigation facility component, with a bar chart representing the length of the fatigue life, and shows the path of the crack from its source to the component, as well as the extent of crack propagation in each component. The depth of the crack in the component is indicated by the shade of color, and the speed bar indicates the rate of crack propagation. Based on crack propagation and fatigue life assessments, risk assessments are conducted on navigation facilities. Components with risks exceeding a preset threshold are highlighted.

Images