Bridge detection robot structure defect identification method and system based on deep learning
By employing deep learning methods to divide bridges into structural units and fuse multimodal data, the problems of low efficiency and insufficient identification in existing bridge inspection technologies have been solved, enabling accurate identification and stable detection of bridge structural defects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-03-13
AI Technical Summary
Existing bridge inspection technologies suffer from low efficiency, high subjectivity, poor environmental adaptability, and difficulty in fully covering key parts. Furthermore, existing deep learning methods are insufficient in real-world engineering environments and cannot effectively capture minute cracks and internal defects.
The bridge is divided into multiple structural units, detection weight factors are calculated to generate detection trajectories, multiple physical modal data are collected for spatiotemporal registration, and feature tensors are generated by fusing morphological encoding and dynamic evolution encoding through deep learning to identify outliers in structural units.
It enables accurate and stable identification of bridge structural defects, possesses higher engineering application value and promotion potential, and can simultaneously capture the local geometric morphology and temporal variation patterns of the structural surface.
Smart Images
Figure CN121659128A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of bridge defect recognition technology, and more specifically, relates to a method and system for structural defect recognition of bridge inspection robots based on deep learning. Background Technology
[0002] Existing bridge structural defect detection and identification technologies can be mainly divided into three categories: first, manual inspection and traditional non-destructive testing methods; second, data acquisition and processing methods based on single sensors or single modalities; and third, image recognition or point cloud recognition methods based on deep learning that have been developed in recent years. Manual inspection remains the primary method for most bridge inspections, relying on inspectors to observe and measure the bridge surface using tools such as telescopes, crack gauges, and rebound hammers. While this method offers some intuitiveness in practical applications, it suffers from low efficiency, strong subjectivity, and poor environmental adaptability. Especially in high piers, long spans, or complex water conditions, it is difficult for personnel to reach all critical areas, easily creating blind spots and leading to missed defects or incomplete records. Traditional non-destructive testing methods such as ultrasonic, magnetic particle, and penetrant testing can identify some internal defects, but they typically require deploying a large number of sensors or engaging in contact operations, resulting in long inspection cycles, high costs, and significant difficulties in implementation within the bridge's operating environment. In recent years, with the development of robotic platforms and sensor technologies, automated inspection methods have emerged that use drones, tracked robots, or wheeled robots equipped with high-definition cameras, LiDAR, and other devices for data collection. These methods have addressed the low efficiency of manual inspections to some extent. However, these methods are often limited to single-modal data. For example, image-based crack identification relies on surface texture features and is extremely sensitive to environmental factors such as lighting, shadows, pollution, and coating peeling, making it prone to misjudgments or missed detections. While geometric modeling and defect detection based on 3D point clouds can reflect the morphological changes of bridge components, it struggles to capture early features of fine cracks and surface corrosion, and cannot reflect the evolution of internal voids and steel reinforcement corrosion, resulting in insufficient feature dimensions. In terms of deep learning applications, most existing methods directly input the collected images or point clouds into convolutional neural networks or transformer networks to train models for structural defect identification. These methods perform well on standard datasets, but have significant limitations in real-world engineering environments. Summary of the Invention
[0003] To address the above technical problems, this invention proposes a deep learning-based method for identifying structural defects in bridge inspection robots, comprising: The bridge is divided into multiple structural units, the detection weight factor of each structural unit is calculated, and the robot's detection trajectory is generated based on the detection weight factor. During the execution of the detection trajectory, the robot collects multiple physical modal data, performs spatiotemporal registration on each type of physical modal data, and generates the original detection matrix; The original detection matrix is input into the feature construction network, and morphological encoding and dynamic evolution encoding are extracted, concatenated and fused to generate a feature tensor; The outlier values of structural units are calculated based on the feature tensor, and structural units that exceed the preset outlier threshold are identified as structural units with structural defects.
[0004] Furthermore, the calculation of the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.
[0005] Furthermore, the detection weighting factor for all structural units Sort according to detection weight factors The robot generates detection trajectories in descending order of size.
[0006] Furthermore, the collection of multiple physical modal data and the spatiotemporal registration of each physical modal data include: synchronously collecting bridge surface and internal structure data using image sensors and ultrasonic probes mounted on the robot and employing a unified time reference, and projecting the bridge surface and internal structure data onto the bridge coordinate system to complete the spatiotemporal registration.
[0007] Furthermore, the feature construction network includes: a morphological encoding layer, a dynamic evolution layer, and a fusion layer; The morphological coding layer is used to encode features of single-time data in physical modal data and generate single-time feature vectors. The dynamic evolution layer is used to encode the temporal features of multi-time sequence data in physical modal data and generate temporal feature vectors; The fusion layer is used to concatenate and fuse single-time feature vectors and temporal feature vectors to generate feature tensors.
[0008] Furthermore, the outliers of structural elements calculated based on the characteristic tensor include: , in, Structural unit outliers, Structural unit Feature tensor, Structural unit The characteristic tensor average value, Structural unit The characteristic tensor standard deviation, For numerically stable terms, For boundary-sensitive weights, Structural unit The characteristic gradient magnitude.
[0009] Furthermore, classifying structural units that exceed a preset anomaly threshold as structural units with structural defects includes: if the structural unit outliers If the preset abnormal threshold is exceeded, the structural unit It is a structural unit with structural defects.
[0010] Furthermore, Structural unit The feature gradient magnitudes include: , in, Structural unit The x-coordinate of the center point Structural unit Spatial scale filter kernel, Structural unit The y-coordinate of the center point Structural unit The z-coordinate of the center point For time, Structural unit The time-scale filtering kernel.
[0011] This invention also proposes a bridge inspection robot structural defect identification system based on deep learning, comprising: The trajectory generation module is used to divide the bridge into multiple structural units, calculate the detection weight factor for each structural unit, and generate the robot's detection trajectory based on the detection weight factor. The data acquisition module is used to collect multiple physical modal data during the execution of the detection trajectory by the robot, perform spatiotemporal registration on each physical modal data, and generate the original detection matrix; The data processing module is used to input the original detection matrix into the feature construction network, extract morphological codes and dynamic evolution codes, and splice and fuse them to generate feature tensors; The detection module is used to calculate the outlier values of structural units based on the feature tensor, and to identify structural units that exceed a preset outlier threshold as structural units with structural defects.
[0012] Furthermore, the calculation of the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.
[0013] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art: The technical solution of this invention sets up a morphological encoding module, a dynamic evolution module, and a fusion layer, forming a multimodal, multi-scale, and cross-temporal deep feature construction mechanism. It can simultaneously capture the local geometric morphology, crack texture features, and their expansion patterns over time on the structural surface during the operation of the bridge inspection robot, and perform unified modeling and fusion of sensor signals, thereby achieving accurate, stable, and intelligent identification of bridge structural defects, and possessing higher engineering application value and promotion potential. Attached Figure Description
[0015] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a system structure diagram of Embodiment 2 of the present invention. Detailed Implementation
[0017] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.
[0018] The method provided by this invention can be implemented in a terminal environment that may include one or more of the following components: a processor, a storage medium, and a display screen. The storage medium stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.
[0019] A processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the terminal, and performs various functions and processes data by running or executing instructions, programs, code sets or instruction sets stored in the storage medium, and by calling data stored in the storage medium.
[0020] Storage media can include random access memory (RAM) or read-only memory (ROM). Storage media can be used to store instructions, programs, code, code sets, or instructions.
[0021] The display screen is used to show the user interface of each application.
[0022] In addition, those skilled in the art will understand that the structure of the terminal described above does not constitute a limitation on the terminal. The terminal may include more or fewer components, or combine certain components, or have different component arrangements. For example, the terminal may also include radio frequency circuits, input units, sensors, audio circuits, power supplies, and other components, which will not be described in detail here.
[0023] Example 1 like Figure 1 As shown in the figure, this embodiment proposes a method for identifying structural defects in a bridge inspection robot based on deep learning, including: Step 101: Divide the bridge into multiple structural units, calculate the detection weight factor for each structural unit, and generate the robot's detection trajectory based on the detection weight factor. Specifically, calculating the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.
[0024] Preferably, calculate the first Mechanical importance index of each structural unit The formula is: , in, For example, a normalization function or , will the Mechanical importance index of each structural unit Mapped to [0,1] As the weight of the bending moment, For the first The absolute value of the maximum bending moment of each structural unit This is a reference value for the maximum bending moment. As the weight of the shear force, For the first The absolute value of the maximum shear force of each structural unit. This is a reference value for the maximum shear force. As the weight of the displacement, For the first Displacement of each structural unit This is a reference value for the displacement. The weights for the functional importance grading values, For the first Functional importance grading values for each structural unit (e.g., main beam = 1, secondary beam = 0.5, railing = 0.1).
[0025] Preferably, calculate the first Component failure consequence coefficient of each structural unit include: , in, As the weight of the personnel risk coefficient, For the first Personnel risk coefficient of each structural unit ( = (Daily average traffic volume × Average number of occupants) + Daily average number of pedestrians). The weight of the traffic criticality coefficient, For the first Traffic criticality coefficient for each structural unit (expressway / arterial road = 1, secondary arterial road = 0.5, local road = 0.2; if the bridge is the only access route and there is no alternative route, then set to 1). The weight for redundancy deficiency, For the first Redundancy deficiency of the first structural unit (the redundancy deficiency of the first structural unit) A structural unit is virtually removed, and the remaining load-bearing capacity of the entire bridge is observed to see if it exceeds the safety factor. If the removal causes overall structural instability or a decrease in load-bearing capacity of more than 50%, then the bridge is considered unsuitable. =1, if the impact is limited (decline less than 10%), then take... =0.1).
[0026] Preferably, calculate the first Vulnerability index of each structural unit include: , in, As the weight of the environmental erosion index, For the first Environmental erosion index of each structural unit The weighting of historical defect frequencies, The weights for the sensor observation anomaly scores, For the first Historical defect frequency of structural unit (statistics of the first) The frequency of historical defects is obtained by dividing the number of times a structural unit has had defects in the past n years by the number of inspections. For the first Anomaly score for sensor observations of each structural unit (comparing the signal characteristics output by the sensor to the normal range; for example, if the ultrasonic echo attenuation rate exceeds a threshold of 30%). Recorded as 0.8; if it is within the normal fluctuation range, (Recorded as 0.1), , and Normalize, and .
[0027] Preferably, calculate the first Environmental erosion index of each structural unit include: , in, For the first Temperature difference of each structural unit For the first Humidity of each structural unit For the first The chloride ion concentration of the first structural unit, and the first Environmental erosion index of each structural unit Normalize to [0,1].
[0028] Specifically, the detection weighting factor for all structural units Sort according to detection weight factors The robot generates detection trajectories in descending order of size.
[0029] Step 102: During the execution of the detection trajectory, the robot collects multiple physical modal data, performs spatiotemporal registration on each type of physical modal data, and generates the original detection matrix; Specifically, collecting multiple physical modal data and performing spatiotemporal registration for each physical modal data includes: synchronously collecting bridge surface and internal structure data using image sensors and ultrasonic probes mounted on the robot and employing a unified time reference, and projecting the bridge surface and internal structure data onto the bridge coordinate system to complete the spatiotemporal registration.
[0030] Step 103: Input the original detection matrix into the feature construction network, extract the morphological encoding and dynamic evolution encoding, and concatenate and fuse them to generate a feature tensor; Specifically, the feature construction network includes: a morphological encoding layer, a dynamic evolution layer, and a fusion layer; The morphological encoding layer is used to encode features of single-time data in physical modal data and generate single-time feature vectors. The morphological encoding layer can be a convolutional neural network (CNN) or a graph convolutional network (GCN). The dynamic evolution layer is used to encode the temporal features of multi-time sequence data in physical modal data and generate temporal feature vectors. The dynamic evolution layer can be a temporal convolutional network (TCN) or a Transformer temporal modeling (Time-seriesTransformer). The fusion layer is used to concatenate and fuse single-time feature vectors and temporal feature vectors to generate feature tensors. The fusion layer can be a multimodal fusion network.
[0031] Step 104: Calculate the outlier values of the structural units based on the feature tensor, and identify structural units that exceed the preset outlier threshold as structural units with structural defects.
[0032] Specifically, outliers of structural units calculated based on feature tensors include: , in, Structural unit outliers, Structural unit Feature tensor, Structural unit The characteristic tensor average value, Structural unit The characteristic tensor standard deviation, For numerically stable terms, For boundary-sensitive weights, Structural unit The characteristic gradient magnitude.
[0033] Specifically, structural units exceeding a preset anomaly threshold are considered structural units with structural defects, including: if the structural unit outliers If the preset abnormal threshold is exceeded, the structural unit It is a structural unit with structural defects.
[0034] Specifically, Structural unit The feature gradient magnitudes include: , in, Structural unit The x-coordinate of the center point Structural unit Spatial scale filter kernel, Structural unit The y-coordinate of the center point Structural unit The z-coordinate of the center point For time, Structural unit The time-scale filtering kernel.
[0035] Preferred structural unit Spatial Scale Filtering Kernel for: , in, For spatial scale.
[0036] Preferably, this embodiment provides a method for obtaining spatial scale. An example is shown below: Define the physical scale range of a certain defect as follows: (For example, crack width 0.2mm-5mm); the spatial resolution of the image sensor is... The pixel scale range corresponding to the defect is: , in, The smallest pixel scale corresponding to the defect. If the maximum pixel scale corresponding to the defect is set, then... .
[0037] Preferred structural unit Time-scale filter kernel for: , in, For time scale.
[0038] Preferably, this embodiment provides the acquisition time scale. An example is shown below: Assuming the robot's moving speed (m / s), sensor sampling frequency If (Hz), then the physical displacement corresponding to the adjacent frame is To detect the physical length Dynamic changes within the range, .
[0039] Example 2 like Figure 2 As shown in the figure, this embodiment proposes a bridge inspection robot structural defect identification system based on deep learning, including: The trajectory generation module is used to divide the bridge into multiple structural units, calculate the detection weight factor for each structural unit, and generate the robot's detection trajectory based on the detection weight factor. Specifically, calculating the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.
[0040] Preferably, calculate the first Mechanical importance index of each structural unit The formula is: , in, For example, a normalization function or , will the Mechanical importance index of each structural unit Mapped to [0,1] As the weight of the bending moment, For the first The absolute value of the maximum bending moment of each structural unit This is a reference value for the maximum bending moment. As the weight of the shear force, For the first The absolute value of the maximum shear force of each structural unit. This is a reference value for the maximum shear force. As the weight of the displacement, For the first Displacement of each structural unit This is a reference value for the displacement. The weights for the functional importance grading values, For the first Functional importance grading values for each structural unit (e.g., main beam = 1, secondary beam = 0.5, railing = 0.1).
[0041] Preferably, calculate the first Component failure consequence coefficient of each structural unit include: , in, As the weight of the personnel risk coefficient, For the first Personnel risk coefficient of each structural unit ( = (Daily average traffic volume × Average number of occupants) + Daily average number of pedestrians). The weight of the traffic criticality coefficient, For the first Traffic criticality coefficient for each structural unit (expressway / arterial road = 1, secondary arterial road = 0.5, local road = 0.2; if the bridge is the only access route and there is no alternative route, then set to 1). The weight for redundancy deficiency, For the first Redundancy deficiency of the first structural unit (the redundancy deficiency of the first structural unit) A structural unit is virtually removed, and the remaining load-bearing capacity of the entire bridge is observed to see if it exceeds the safety factor. If the removal causes overall structural instability or a decrease in load-bearing capacity of more than 50%, then the bridge is considered unsuitable. =1, if the impact is limited (decline less than 10%), then take... =0.1).
[0042] Preferably, calculate the first Vulnerability index of each structural unit include: , in, As the weight of the environmental erosion index, For the first Environmental erosion index of each structural unit The weighting of historical defect frequencies, The weights for the sensor observation anomaly scores, For the first Historical defect frequency of structural unit (statistics of the first) The frequency of historical defects is obtained by dividing the number of times a structural unit has had defects in the past n years by the number of inspections. For the first Anomaly score for sensor observations of each structural unit (comparing the signal characteristics output by the sensor to the normal range; for example, if the ultrasonic echo attenuation rate exceeds a threshold of 30%). Recorded as 0.8; if it is within the normal fluctuation range, (Recorded as 0.1), , and Normalize, and .
[0043] Preferably, calculate the first Environmental erosion index of each structural unit include: , in, For the first Temperature difference of each structural unit For the first Humidity of each structural unit For the first The chloride ion concentration of the first structural unit, and the first Environmental erosion index of each structural unit Normalize to [0,1].
[0044] Specifically, the detection weighting factor for all structural units Sort according to detection weight factors The robot generates detection trajectories in descending order of size.
[0045] The data acquisition module is used to collect multiple physical modal data during the execution of the detection trajectory by the robot, perform spatiotemporal registration on each physical modal data, and generate the original detection matrix; Specifically, collecting multiple physical modal data and performing spatiotemporal registration for each physical modal data includes: synchronously collecting bridge surface and internal structure data using image sensors and ultrasonic probes mounted on the robot and employing a unified time reference, and projecting the bridge surface and internal structure data onto the bridge coordinate system to complete the spatiotemporal registration.
[0046] The data processing module is used to input the original detection matrix into the feature construction network, extract morphological codes and dynamic evolution codes, and splice and fuse them to generate feature tensors; Specifically, the feature construction network includes: a morphological encoding layer, a dynamic evolution layer, and a fusion layer; The morphological encoding layer is used to encode features of single-time data in physical modal data and generate single-time feature vectors. The morphological encoding layer can be a convolutional neural network (CNN) or a graph convolutional network (GCN). The dynamic evolution layer is used to encode the temporal features of multi-time sequence data in physical modal data and generate temporal feature vectors. The dynamic evolution layer can be a temporal convolutional network (TCN) or a Transformer temporal modeling (Time-seriesTransformer). The fusion layer is used to concatenate and fuse single-time feature vectors and temporal feature vectors to generate feature tensors. The fusion layer can be a multimodal fusion network.
[0047] The detection module is used to calculate the outlier values of structural units based on the feature tensor, and to identify structural units that exceed a preset outlier threshold as structural units with structural defects.
[0048] Specifically, outliers of structural units calculated based on feature tensors include: , in, Structural unit outliers, Structural unit Feature tensor, Structural unit The characteristic tensor average value, Structural unit The characteristic tensor standard deviation, For numerically stable terms, For boundary-sensitive weights, Structural unit The characteristic gradient magnitude.
[0049] Specifically, structural units exceeding a preset anomaly threshold are considered structural units with structural defects, including: if the structural unit outliers If the preset abnormal threshold is exceeded, the structural unit It is a structural unit with structural defects.
[0050] Specifically, Structural unit The feature gradient magnitudes include: , in, Structural unit The x-coordinate of the center point Structural unit Spatial scale filter kernel, Structural unit The y-coordinate of the center point Structural unit The z-coordinate of the center point For time, Structural unit The time-scale filtering kernel.
[0051] Preferred structural unit Spatial Scale Filtering Kernel for: , in, For spatial scale.
[0052] Preferably, this embodiment provides a method for obtaining spatial scale. An example is shown below: Define the physical scale range of a certain defect as follows: (For example, crack width 0.2mm-5mm); the spatial resolution of the image sensor is... The pixel scale range corresponding to the defect is: , in, The smallest pixel scale corresponding to the defect. If the maximum pixel scale corresponding to the defect is set, then... .
[0053] Preferred structural unit Time-scale filter kernel for: , in, For time scale.
[0054] Preferably, this embodiment provides the acquisition time scale. An example is shown below: Assuming the robot's moving speed (m / s), sensor sampling frequency If (Hz), then the physical displacement corresponding to the adjacent frame is To detect the physical length Dynamic changes within the range, .
[0055] Example 3 This invention also proposes a storage medium storing multiple instructions for implementing the aforementioned deep learning-based bridge inspection robot structural defect identification method.
[0056] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.
[0057] Optionally, in this embodiment, the storage medium is configured to store program code for performing the method steps of Embodiment 1.
[0058] Example 4 This invention also proposes an electronic device, including a processor and a storage medium connected to the processor. The storage medium stores multiple instructions, which can be loaded and executed by the processor to enable the processor to execute the aforementioned deep learning-based bridge inspection robot structural defect identification method.
[0059] Specifically, the electronic device in this embodiment can be a computer terminal, which may include one or more processors and a storage medium.
[0060] The storage medium can be used to store software programs and modules, such as the deep learning-based bridge inspection robot structural defect identification method in this embodiment of the invention. The corresponding program instructions / modules are executed by the processor through running the software programs and modules stored in the storage medium, thereby performing various functional applications and data processing, thus realizing the aforementioned deep learning-based bridge inspection robot structural defect identification method. The storage medium may include high-speed random access storage media, and may also include non-volatile storage media, such as one or more magnetic storage systems, flash memory, or other non-volatile solid-state storage media. In some instances, the storage medium may further include storage media remotely configured relative to the processor, which can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0061] The processor can execute the method steps of Embodiment 1 by calling the information and application stored in the storage medium through the transmission system.
[0062] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0063] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.
[0064] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0065] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0066] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, read-only storage media (ROM), random access storage media (RAM), portable hard drives, magnetic disks, optical disks, and other media capable of storing program code.
[0067] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A method for identifying structural defects in a bridge inspection robot based on deep learning, characterized in that, include: The bridge is divided into multiple structural units, the detection weight factor of each structural unit is calculated, and the robot's detection trajectory is generated based on the detection weight factor. During the execution of the detection trajectory, the robot collects multiple physical modal data, performs spatiotemporal registration on each type of physical modal data, and generates the original detection matrix; The original detection matrix is input into the feature construction network, and morphological encoding and dynamic evolution encoding are extracted, concatenated and fused to generate a feature tensor; The outlier values of structural units are calculated based on the feature tensor, and structural units that exceed the preset outlier threshold are identified as structural units with structural defects.
2. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 1, characterized in that, The calculation of the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.
3. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 2, characterized in that, Detection weighting factors for all structural units Sort according to detection weight factors The robot generates detection trajectories in descending order of size.
4. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 1, characterized in that, The process of collecting multiple physical modal data and performing spatiotemporal registration for each type of physical modal data includes: synchronously collecting bridge surface and internal structure data using image sensors and ultrasonic probes mounted on the robot and employing a unified time reference, and then projecting the bridge surface and internal structure data onto the bridge coordinate system to complete the spatiotemporal registration.
5. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 1, characterized in that, The feature-constructed network includes: a morphological encoding layer, a dynamic evolution layer, and a fusion layer; The morphological coding layer is used to encode features of single-time data in physical modal data and generate single-time feature vectors. The dynamic evolution layer is used to encode the temporal features of multi-time sequence data in physical modal data and generate temporal feature vectors; The fusion layer is used to concatenate and fuse single-time feature vectors and temporal feature vectors to generate feature tensors.
6. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 1, characterized in that, The outliers of structural elements calculated from the characteristic tensor include: , in, Structural unit outliers, Structural unit Feature tensor Structural unit The characteristic tensor average value, Structural unit The characteristic tensor standard deviation, For numerically stable terms, For boundary-sensitive weights, Structural unit The characteristic gradient magnitude.
7. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 6, characterized in that, Structural units exceeding a preset anomaly threshold are considered structural units with structural defects, including: if structural units... outliers If the preset abnormal threshold is exceeded, the structural unit It is a structural unit with structural defects.
8. The method for identifying structural defects in a bridge inspection robot based on deep learning as described in claim 6, characterized in that, Structural unit The feature gradient magnitudes include: , in, Structural unit The x-coordinate of the center point Structural unit Spatial scale filter kernel, Structural unit The y-coordinate of the center point Structural unit The z-coordinate of the center point For time, Structural unit The time-scale filtering kernel.
9. A structural defect identification system for bridge inspection robots based on deep learning, characterized in that, include: The trajectory generation module is used to divide the bridge into multiple structural units, calculate the detection weight factor for each structural unit, and generate the robot's detection trajectory based on the detection weight factor. The data acquisition module is used to collect multiple physical modal data during the execution of the detection trajectory by the robot, perform spatiotemporal registration on each physical modal data, and generate the original detection matrix; The data processing module is used to input the original detection matrix into the feature construction network, extract morphological codes and dynamic evolution codes, and splice and fuse them to generate feature tensors; The detection module is used to calculate the outlier values of structural units based on the feature tensor, and to identify structural units that exceed a preset outlier threshold as structural units with structural defects.
10. The bridge inspection robot structural defect identification system based on deep learning as described in claim 9, characterized in that, The calculation of the detection weight factor for each structural unit includes: , in, For the first The detection weight factor for each structural unit. The weights of the mechanical importance index, For the first Mechanical importance index of each structural unit The weights of the component failure consequence coefficients, For the first The component failure consequence coefficient of each structural unit The weights of the fragility index, For the first The vulnerability index of each structural unit.