Three-dimensional detection and diagnosis simulation verification method for tunnel apparent disease
The three-dimensional detection method for tunnel surface defects by combining laser and camera modules with machine learning models solves the cost and efficiency problems of existing tunnel defect diagnosis model verification, and achieves efficient and accurate tunnel defect diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2023-12-29
- Publication Date
- 2026-06-23
AI Technical Summary
Existing non-physical verification and real-vehicle verification methods for tunnel surface defect diagnosis algorithms are costly, time-consuming, and limited by equipment and data, making them difficult to implement effectively and unable to ensure data accuracy and efficiency.
Point cloud data of the tunnel model is acquired using a laser module and image data is acquired using a camera module. Combined with a machine learning model, three-dimensional detection and diagnosis are performed through a simulation test platform to locate suspected disease areas. The effectiveness of the machine learning model is verified by comparing it with preset real disease information.
By comprehensively utilizing point cloud data and image data in a simulated environment, the verification effect and efficiency of machine learning models are improved, the diagnostic accuracy is optimized, and the limitations of real vehicle verification and non-physical verification are avoided.
Smart Images
Figure CN117788447B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of tunnel inspection technology, and in particular to a three-dimensional detection and diagnostic simulation verification method, platform and storage medium for tunnel surface defects. Background Technology
[0002] Tunnels, as vital transportation infrastructure, bear the burden of transporting a large number of vehicles and passengers. However, various apparent defects may exist inside tunnels, such as deformation, misalignment, spalling, and cracks. If these apparent defects are not detected and identified in a timely manner, they may lead to serious adverse consequences, including accidents, traffic disruptions, increased maintenance costs, and a shortened tunnel lifespan.
[0003] Therefore, tunnel surface defect diagnosis technology is crucial. To improve related algorithm models for tunnel surface defect diagnosis, current methods often employ non-physical verification based on simulation technology or real-vehicle verification based on actual tunnel environments. However, non-physical verification methods are not conducted in a real physical environment, are limited by model assumptions, and cannot guarantee data accuracy; real-vehicle verification methods typically require significant investment of human and material resources, cannot simulate all possible tunnel conditions and defect types, and are difficult to replicate to obtain large amounts of test data.
[0004] In summary, non-physical verification methods and vehicle verification methods for tunnel surface defect diagnosis are difficult to implement or have limited effectiveness and efficiency due to limitations in cost, time, equipment, data, and human resources. Therefore, a new simulation verification method is urgently needed to avoid the above-mentioned limitations of non-physical verification methods and vehicle verification methods, thereby improving the accuracy of tunnel surface defect diagnosis. Summary of the Invention
[0005] The main purpose of this application is to provide a three-dimensional detection and diagnosis simulation verification method, platform and storage medium for tunnel surface defects, aiming to solve the problem that existing non-physical verification methods and real vehicle verification methods for tunnel surface defect detection and diagnosis are difficult to implement or limit their effectiveness and efficiency.
[0006] To achieve the above objectives, this application provides a three-dimensional detection and diagnostic simulation verification method for tunnel surface defects. The method is applied to a simulation test platform, which includes a detection trolley and a tunnel model. The detection trolley is equipped with a laser module and a camera module. The method includes:
[0007] The point cloud data of the tunnel model is acquired through the laser module, and a first appearance state detection result of the tunnel model is obtained based on the point cloud data analysis; and the image data of the tunnel model is acquired through the camera module, and a second appearance state detection result of the tunnel model is obtained based on the image data analysis.
[0008] Based on the first apparent state detection result and the second apparent state detection result, locate the suspected defect areas of the tunnel model;
[0009] The point cloud data and image data corresponding to the suspected disease area are input into a preset machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model. The disease diagnosis result includes the location, type, and three-dimensional size of the diagnosed disease.
[0010] The diagnostic results of the defects are compared with the preset real defect information of the tunnel model. Based on the comparison results, the machine learning model is tested to see if it meets the preset validity conditions. The preset real defect information includes the location, type, and three-dimensional dimensions of the real defects. The machine learning model is used to diagnose the apparent defects of the tunnel.
[0011] Optionally, the detection vehicle is equipped with a computer, and the computer is equipped with a data analysis module. The step of obtaining the first apparent state detection result of the tunnel model based on the point cloud data analysis includes:
[0012] The data analysis module performs surface grayscale difference analysis and stereo depth difference analysis on the point cloud data to obtain the first appearance state detection result of the tunnel model.
[0013] The step of obtaining the second apparent state detection result of the tunnel model based on the image data analysis includes:
[0014] The data analysis module performs surface grayscale difference analysis and surface spectral value difference analysis on the image data to obtain the second appearance state detection result of the tunnel model.
[0015] Optionally, the laser module is equipped with an encoder and an inertial sensor, and the method further includes:
[0016] The encoder acquires mileage increment information and speed information while in motion; and the inertial sensor acquires speed increment information and angular velocity increment information while in motion.
[0017] The step of obtaining the first appearance state detection result of the tunnel model by performing surface grayscale difference analysis and stereo depth difference analysis based on the point cloud data through the data analysis module includes:
[0018] The data analysis module performs data fusion based on the point cloud data, the mileage increment information, the speed information, the speed increment information, and the angular velocity increment information to obtain a three-dimensional point cloud model of the tunnel model.
[0019] The first appearance state detection result of the tunnel model is obtained by performing surface grayscale difference analysis and stereo depth difference analysis on the three-dimensional point cloud model.
[0020] Optionally, the step of locating the suspected defect areas of the tunnel model based on the first apparent state detection result and the second apparent state detection result includes:
[0021] The data analysis module performs cross-validation on the first apparent state detection result and the second apparent state detection result to obtain the cross-validation result.
[0022] The data analysis module locates the suspected defect areas of the tunnel model based on the cross-validation results.
[0023] Optionally, the computer is equipped with a central control system, and the laser module is equipped with a laser scanner. The step of acquiring the point cloud data of the tunnel model through the laser module includes:
[0024] The central control system sends a point cloud acquisition command to the laser module, controls the laser scanner of the laser module to acquire the point cloud data of the tunnel model, and transmits the point cloud data of the tunnel model to the data analysis module.
[0025] The step of acquiring image data of the tunnel model through the camera module includes:
[0026] The central control system sends image acquisition commands to the camera module, controls the camera module to acquire image data of the tunnel model, and transmits the image data of the tunnel model to the data analysis module.
[0027] Optionally, the detection vehicle is equipped with a drive module, and the method further includes:
[0028] In response to the input moving speed and moving direction parameters, the central control system generates and sends a driving command to the driving module, so that the driving module moves the detection vehicle based on the driving command.
[0029] Optionally, the detection vehicle is equipped with an attitude adjustment module, the attitude adjustment module and the laser module are connected by a drive connection, and the method further includes:
[0030] In response to the input attitude adjustment parameters, the central control system generates and sends attitude adjustment commands to the attitude adjustment module, so that the attitude adjustment module adjusts the attitude of the laser module based on the attitude adjustment commands.
[0031] Optionally, the step of inputting the point cloud data and image data corresponding to the suspected disease area into a preset machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model includes:
[0032] Feature information of the suspected disease area is extracted from the point cloud data and image data corresponding to the suspected disease area;
[0033] The feature information of the suspected disease area is input into the machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model.
[0034] This application also proposes a simulation test platform, which performs the steps of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described above.
[0035] This application also proposes a computer-readable storage medium storing a simulated tunnel defect diagnosis and verification program. When the simulated tunnel defect diagnosis and verification program is executed by a processor, it implements the steps of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described above.
[0036] The method, platform, and storage medium for three-dimensional detection and diagnosis simulation verification of tunnel surface defects proposed in this application mainly involve: acquiring point cloud data of the tunnel model through the laser module; analyzing the point cloud data to obtain a first appearance state detection result of the tunnel model; acquiring image data of the tunnel model through the camera module; analyzing the image data to obtain a second appearance state detection result of the tunnel model; locating suspected defect areas of the tunnel model based on the first and second appearance state detection results; inputting the point cloud data and image data corresponding to the suspected defect areas into a preset machine learning model to obtain a defect diagnosis result corresponding to the suspected defect areas output by the machine learning model, wherein the defect diagnosis result includes the location, type, and three-dimensional dimensions of the diagnosed defect; comparing the defect diagnosis result with preset real defect information of the tunnel model; and verifying whether the machine learning model meets preset validity conditions based on the comparison results, wherein the preset real defect information includes the location, type, and three-dimensional dimensions of the real defect, and the machine learning model is used to diagnose tunnel surface defects. Based on the proposed solution, a comprehensive inspection of the appearance of a tunnel model is achieved through a simulation test platform. Point cloud data and image data are comprehensively utilized in the simulation environment to verify the accuracy of the machine learning model used to diagnose tunnel appearance defects. This avoids the limitations of real vehicle verification and non-physical verification, effectively improves the verification effect and efficiency of the machine learning model, promotes the performance optimization of the machine learning model, and improves the diagnostic accuracy. Attached Figure Description
[0037] Figure 1 This is a schematic flowchart of the first exemplary embodiment of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects of this application;
[0038] Figure 2 This is a first schematic diagram of the simulation test platform involved in the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects in this application;
[0039] Figure 3 This is a second schematic diagram of the simulation test platform involved in the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects in this application;
[0040] Figure 4 This is a schematic diagram of the comprehensive process for the three-dimensional detection and diagnostic simulation verification method for tunnel surface defects in this application.
[0041] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0042] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.
[0043] The main solution of this application embodiment is as follows: Point cloud data of the tunnel model is acquired through the laser module; a first appearance state detection result of the tunnel model is obtained based on the point cloud data analysis; image data of the tunnel model is acquired through the camera module; a second appearance state detection result of the tunnel model is obtained based on the image data analysis; suspected defect areas of the tunnel model are located according to the first and second appearance state detection results; the point cloud data and image data corresponding to the suspected defect areas are input into a preset machine learning model to obtain a defect diagnosis result corresponding to the suspected defect areas output by the machine learning model, wherein the defect diagnosis result includes the location, type, and three-dimensional dimensions of the diagnosed defect; the defect diagnosis result is compared with preset real defect information of the tunnel model, and the machine learning model is used to diagnose tunnel appearance defects based on the comparison result. Based on the proposed solution, a comprehensive detection of the appearance of a tunnel model is achieved through a simulation test platform. Point cloud data and image data are comprehensively utilized in the simulation environment to verify the accuracy of the machine learning model used to diagnose tunnel appearance defects. This avoids the limitations of real vehicle verification and non-physical verification, and effectively improves the verification effect and efficiency of the machine learning model.
[0044] Reference Figure 1 The first embodiment of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects of this application provides a flowchart. The method is applied to a simulation test platform, which includes a detection trolley and a tunnel model. The detection trolley is equipped with a laser module and a camera module. The method includes:
[0045] Step S10: Obtain point cloud data of the tunnel model through the laser module, and analyze the point cloud data to obtain a first appearance state detection result of the tunnel model; and obtain image data of the tunnel model through the camera module, and analyze the image data to obtain a second appearance state detection result of the tunnel model.
[0046] Specifically, my country leads the world in the scale, difficulty, and number of tunnels and underground engineering projects. With a large number of tunnels put into operation, ensuring the safety of residents' travel has become a top priority for the transportation industry. Tunnels are subjected to vibrations and alternating loads from passing vehicles for extended periods, and may be affected by nearby construction and traversing adverse geological conditions such as soft soil rich in water, mining subsidence areas, and liquefiable soil. Under the influence of various factors, tunnels are prone to defects such as deformation, misalignment, spalling, and cracks. If these defects are not identified and maintained in a timely manner, they will significantly shorten the service life of the tunnels and even cause serious traffic accidents, threatening the lives and property of residents.
[0047] Therefore, tunnel surface defect diagnosis technology is crucial. To improve related algorithm models for tunnel surface defect detection and diagnosis, current methods often employ non-physical verification based on simulation technology or real-vehicle verification based on actual tunnel environments. However, non-physical verification methods, not conducted in a real physical environment, are limited by model assumptions and cannot guarantee data accuracy; real-vehicle verification methods typically require significant investment of human and material resources, cannot simulate all possible tunnel conditions and defect types, and are difficult to replicate to obtain large amounts of test data. In summary, non-physical and real-vehicle verification methods for tunnel surface defect diagnosis are difficult to implement or limit their effectiveness and efficiency due to limitations in cost, time, equipment, data, and human resources. Therefore, a new simulation verification method is urgently needed to avoid the aforementioned limitations of non-physical and real-vehicle verification methods, thereby improving the accuracy of tunnel surface defect diagnosis.
[0048] To address the limitations of existing non-physical verification and vehicle-based verification methods for diagnosing tunnel surface defects, which are difficult to implement or restrict the effectiveness and efficiency of these methods, this embodiment proposes a three-dimensional detection and diagnostic simulation verification method for tunnel surface defects that combines laser and camera technologies.
[0049] First, the three-dimensional detection and diagnostic simulation verification method for tunnel surface defects involved in this embodiment is applied to a simulation test platform. For example... Figure 2 , Figure 3 As shown, Figure 2 , Figure 3 These are the first and second schematic diagrams of the simulation test platform involved in the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects of this application. The simulation test platform includes a detection trolley and a tunnel model.
[0050] As a physical model, the tunnel model consists of tunnel segments, with the arc surfaces of the segments facing the laser and camera modules of the inspection vehicle. By manually adjusting the tunnel segments, the tunnel model can simulate defects such as deformation, misalignment, chipping, and cracks.
[0051] The inspection vehicle is a multi-functional device, primarily equipped with a laser module and a camera module. The laser module scans the tunnel model with a laser beam to acquire point cloud data of the tunnel model. The camera module acquires image data of the tunnel model through a vision sensor. These two modules work together to provide a comprehensive data source for the simulation test platform.
[0052] On the one hand, the simulation test platform analyzes point cloud data to obtain the first appearance state detection result of the tunnel model. The first appearance state detection result includes the appearance three-dimensional structural information of the tunnel model and the first type of suspected defect area determined based on the appearance three-dimensional structural information. On the other hand, the simulation test platform analyzes image data to obtain the second appearance state detection result of the tunnel model. The second appearance state detection result includes the appearance visual feature information of the tunnel model and the second type of suspected defect area determined based on the appearance visual feature information.
[0053] Step S20: Based on the first appearance state detection result and the second appearance state detection result, locate the suspected defect area of the tunnel model.
[0054] Specifically, if a camera module is used alone, the accuracy of its vision sensor is easily affected by the external environment and image quality; if a laser module is used alone, the accuracy of its laser scanner is generally lower than that of the vision sensor in the area, and it will be reduced because the device's posture and position are constantly changing, and the origin and coordinate axes of the coordinate system are constantly changing, resulting in the scanning points not being in the same coordinate system.
[0055] To address the aforementioned issues, this embodiment fuses the detection results from the laser module and the camera module to improve the accuracy of locating suspected defect areas. More specifically, both the first and second types of suspected defect areas are considered suspicious. Cross-validating the first and second apparent state detection results yields more accurate cross-validation results. Using these cross-validation results, the suspected defect areas of the tunnel model can be accurately located (equivalent to a highly reliable and precisely located area).
[0056] Step S30: Input the point cloud data and image data corresponding to the suspected disease area into a preset machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model. The disease diagnosis result includes the location, type, and three-dimensional size of the diagnosed disease.
[0057] Specifically, the purpose of this embodiment is to verify whether the machine learning model meets the preset validity conditions. The machine learning model is a pre-set model that supports the execution of algorithms related to tunnel surface defect diagnosis, and its function is to diagnose tunnel surface defects. The machine learning model requires continuous improvement, and the simulation test platform provides a convenient and reliable verification basis for improving the machine learning model.
[0058] After identifying suspected disease areas, the corresponding point cloud data and image data can be input into a machine learning model. The model will then output a disease diagnosis result for the suspected disease area, including its location, type, and three-dimensional dimensions. These dimensions can include length, width, and depth. Furthermore, the diagnosis result may also include the shape and extent of the disease. The diagnosed disease may overlap with the suspected disease area in area, or it may only occupy a portion of the suspected disease area. For example, the suspected disease area might be a rectangle, while the diagnosed disease might be an irregularly shaped crack within that area.
[0059] Because machine learning models learn the complex relationships between input data (point clouds and images) and corresponding output labels (disease diagnosis results) during the training phase, this learning ability enables them to make reasonable predictions when faced with new data. In this case, the machine learning model learns the association between point cloud data, image data, and disease features, thus enabling accurate disease diagnosis. Furthermore, machine learning models possess a certain degree of generalization ability, meaning they can extend to unseen data. By being exposed to diverse point cloud and image data during training, machine learning models learn general data patterns, rather than being limited to specific training data. Therefore, when faced with new point cloud and image data, machine learning models with good generalization ability can produce reasonable disease diagnosis results.
[0060] Step S40: Compare the disease diagnosis results with the preset real disease information of the tunnel model, and check whether the machine learning model meets the preset validity conditions based on the comparison results. The preset real disease information includes the location, type, and three-dimensional dimensions of the real disease. The machine learning model is used to diagnose the apparent diseases of the tunnel.
[0061] Specifically, after setting the simulated apparent defects on the tunnel model, it is necessary to input preset real defect information into the simulation test platform. This preset real defect information includes the location, type, and three-dimensional dimensions of the real defects, characterizing the actual defect situation of the tunnel model. The three-dimensional dimensions can include length, width, and depth. Furthermore, the preset real defect information can also include the shape and extent of the real defects.
[0062] By comparing the disease diagnosis results with the preset real disease information of the tunnel model, the comparison results can be obtained. More specifically, the comparison process involves the following: (1) Location comparison: Compare the location of the diagnosed disease with the location of the real disease. If the locations are consistent, it indicates that the machine learning model is accurate in locating the disease. (2) Type comparison: Compare the type of the diagnosed disease with the type of the real disease. If the types match, it indicates that the machine learning model can correctly identify different types of apparent diseases. (3) Shape comparison: Compare the shape of the diagnosed disease with the shape of the real disease. Consistent shapes indicate that the machine learning model is accurate in capturing the shape of the disease. (4) Size comparison: Compare the size of the diagnosed disease with the size of the real disease. Consistent sizes indicate that the machine learning model can accurately estimate the three-dimensional dimensions of the disease. (5) Severity comparison: Compare the severity of the diagnosed disease with the severity of the real disease. Consistent severity indicates that the machine learning model can accurately assess the severity of the disease.
[0063] The comparison results can be quantified into scores for the aforementioned multiple comparisons, and a comprehensive score is calculated based on these scores. If the comprehensive score reaches a preset score threshold, the machine learning model is determined to meet the preset validity conditions; if the comprehensive score does not reach the preset score threshold, the machine learning model is determined not to meet the preset validity conditions.
[0064] Alternatively, the comparison results can be quantified as the scores of the above multiple comparisons. If the scores of the multiple comparisons reach the preset score thresholds, the machine learning model is determined to meet the preset validity conditions; if the score of any comparison does not reach the preset score thresholds, the machine learning model is determined not to meet the preset validity conditions.
[0065] The specific steps for judging the preset validity conditions mentioned above are only illustrative examples. In fact, after multiple comparisons, the comparison results can also be used to characterize the validity of the machine learning model in the form of Boolean values, text, etc., and other methods besides threshold judgment can be used to check whether the machine learning model meets the preset validity conditions.
[0066] In this embodiment, point cloud data of the tunnel model is acquired through the laser module, and a first appearance state detection result of the tunnel model is obtained based on the point cloud data analysis; and image data of the tunnel model is acquired through the camera module, and a second appearance state detection result of the tunnel model is obtained based on the image data analysis; based on the first appearance state detection result and the second appearance state detection result, the suspected defect area of the tunnel model is located; the point cloud data and image data corresponding to the suspected defect area are input into a preset machine learning model to obtain the defect diagnosis result corresponding to the suspected defect area output by the machine learning model, wherein the defect diagnosis result includes the location, type, and three-dimensional size of the diagnosed defect; the defect diagnosis result is compared with the preset real defect information of the tunnel model, and the machine learning model is used to diagnose tunnel appearance defects based on the comparison result. Based on the proposed solution, a comprehensive inspection of the appearance of a tunnel model is achieved through a simulation test platform. Point cloud data and image data are comprehensively utilized in the simulation environment to verify the accuracy of the machine learning model used to diagnose tunnel appearance defects. This avoids the limitations of real vehicle verification and non-physical verification, effectively improves the verification effect and efficiency of the machine learning model, promotes the performance optimization of the machine learning model, and improves the diagnostic accuracy.
[0067] Furthermore, based on the first embodiment described above, a second embodiment of the method for three-dimensional detection and diagnostic simulation verification of tunnel surface defects of this application is proposed. The detection vehicle is equipped with a computer, and the computer is equipped with a data analysis module. The step S10, "obtaining the first appearance state detection result of the tunnel model based on the point cloud data analysis," is further refined, including:
[0068] Step S11: Using the data analysis module, perform surface grayscale difference analysis and stereo depth difference analysis based on the point cloud data to obtain the first appearance state detection result of the tunnel model.
[0069] The step S10, "obtaining the second apparent state detection result of the tunnel model based on the image data analysis," is further refined to include:
[0070] Step S12: Using the data analysis module, perform surface grayscale difference analysis and surface spectral value difference analysis on the image data to obtain the second appearance state detection result of the tunnel model.
[0071] Specifically, the inspection vehicle involved in this embodiment is equipped with a computer. Because the computer can process, analyze, and store large amounts of sensor data, execute complex algorithms, and adjust the inspection strategy in real time, it can also integrate with other systems to achieve autonomous decision-making and remote monitoring. Furthermore, the presence of the computer endows the inspection vehicle with intelligent capabilities, enabling it to more effectively cope with different tunnel environments and types of defects, thus improving inspection accuracy and efficiency.
[0072] More specifically, the computer also has a data analysis module, which is mainly responsible for receiving sensor data (point cloud data and image data) and performing further analysis on the sensor data.
[0073] On one hand, the data analysis module performs surface grayscale difference analysis and 3D depth difference analysis based on point cloud data to obtain the first appearance state detection result of the tunnel model. On the other hand, the data analysis module performs surface grayscale difference analysis and surface spectral value difference analysis based on image data to obtain the second appearance state detection result of the tunnel model.
[0074] Area grayscale difference analysis is an analytical method used to process point cloud or image data, aiming to detect grayscale (or brightness) variations between adjacent regions. In image data analysis, this involves comparing the grayscale values of adjacent pixels in an image; in point cloud data analysis, it involves comparing the grayscale values (or other values reflecting appearance features) between adjacent points. Area grayscale difference analysis can help detect differences in color or brightness between regions and is often used to reveal surface defects such as cracks and color variations. In tunnel surface defect detection, area grayscale difference analysis can be used to identify different textures, colors, or brightness variations on the surface of a tunnel model, which may be signs of defects.
[0075] Area domain spectral difference analysis is a method for processing image data that aims to detect changes between adjacent regions by analyzing differences in the frequency domain. Image data analysis involves performing frequency domain analysis on the image, such as through Fourier transform, and then comparing differences between spectral values (amplitude information in the frequency domain). Area domain spectral difference analysis is commonly used to detect changes in texture, structure, or other features between different regions of image data. This analysis can help reveal subtle differences in appearance features, aiding in the identification of defects and the capture of surface textures.
[0076] 3D depth difference analysis is a method for processing point cloud data to detect depth (distance) variations between different regions. In point cloud data analysis, the depth differences of objects can be calculated using the depth values between adjacent points. In tunnel surface defect detection, 3D depth difference analysis can be used to discover irregular changes on the tunnel surface, such as unevenness and potholes, which may be manifestations of defects.
[0077] In this embodiment, the detection vehicle uses a computer and data analysis module to intelligently analyze point cloud data and image data. It analyzes point cloud data using area grayscale differences and 3D depth differences, and analyzes image data using area grayscale differences and area spectral value differences. This embodiment integrates multimodal information from point clouds and images, improving the accuracy and comprehensiveness of appearance condition detection and enabling more effective identification of tunnel defects.
[0078] Furthermore, based on the second embodiment described above, a third embodiment of the method for three-dimensional detection and diagnostic simulation verification of tunnel surface defects is proposed. The laser module is equipped with an encoder and an inertial sensor, and the method further includes:
[0079] Step S011 involves acquiring mileage increment information and speed information through the encoder while the device is in motion; and acquiring speed increment information and angular velocity increment information through the inertial sensor while the device is in motion.
[0080] Specifically, the laser module involved in this embodiment is equipped with an encoder and an inertial sensor. In practical applications, the laser module moves along the centerline of a tunnel model, simulating movement along a track within the tunnel. The encoder, mounted on the wheels, is typically used to measure the position and motion of the laser module. By measuring the rotation of the wheels or moving parts during movement, the encoder can provide mileage increment information, i.e., the distance the laser module has moved in space, and can also provide velocity information. The inertial sensor includes an accelerometer and a gyroscope, used to measure the linear acceleration and angular velocity of the laser module. By detecting acceleration and rotation, the inertial sensor can provide velocity increment information and angular velocity increment information of the laser module during its movement.
[0081] The step S11, "obtaining the first appearance state detection result of the tunnel model by performing surface grayscale difference analysis and stereo depth difference analysis based on the point cloud data through the data analysis module," is further refined, including:
[0082] Step S111: The data analysis module performs data fusion based on the point cloud data, the mileage increment information, the speed information, the speed increment information, and the angular velocity increment information to obtain the three-dimensional point cloud model of the tunnel model.
[0083] Step S112: Perform surface grayscale difference analysis and stereo depth difference analysis on the three-dimensional point cloud model to obtain the first appearance state detection result of the tunnel model.
[0084] Specifically, through the data orientation module, point cloud data, mileage increment information, velocity information, velocity increment information, angular velocity increment information, etc., can be integrated to determine the actual position and attitude of the laser module in the tunnel, and further obtain the three-dimensional point cloud model of the tunnel model.
[0085] Then, by utilizing the grayscale values of the point cloud data, the grayscale changes between adjacent areas are analyzed. This helps detect differences in surface color or brightness and identify potential surface defects. Utilizing the depth information of the point cloud data, the depth differences between different points in the point cloud help detect surface elevation changes and identify possible bumps or pits. Finally, the analysis yields the initial appearance condition detection results for the tunnel model.
[0086] In this embodiment, the encoder and inertial sensor of the laser module provide rich data during movement, including mileage increments, velocity, velocity increments, and angular velocity increments. Through data fusion, a more comprehensive three-dimensional point cloud model of the tunnel is formed. This not only improves the accuracy of apparent condition detection but also makes the simulation platform more adaptable to complex tunnel environments, enhancing the realism and effectiveness of the simulation verification.
[0087] Furthermore, based on the second embodiment described above, a fourth embodiment of the method for three-dimensional detection and diagnostic simulation verification of tunnel surface defects in this application is proposed. The step S20, "locating the suspected defect area of the tunnel model based on the first surface condition detection result and the second surface condition detection result," is further refined, including:
[0088] Step S21: The data analysis module performs cross-validation on the first apparent state detection result and the second apparent state detection result to obtain the cross-validation result.
[0089] Step S22: Using the data analysis module, locate the suspected defect areas of the tunnel model based on the cross-validation results.
[0090] Specifically, the first appearance condition detection result includes the appearance three-dimensional structural information of the tunnel model, and the first type of suspected disease area determined based on the appearance three-dimensional structural information; the second appearance condition detection result includes the appearance visual feature information of the tunnel model, and the second type of suspected disease area determined based on the appearance visual feature information.
[0091] To improve the accuracy of locating suspected defect areas, this embodiment uses a data analysis module to cross-validate the first and second apparent state detection results to obtain cross-validation results. The cross-validation process may include: (1) Spatial consistency verification: verifying whether the first and second apparent state detection results are spatially consistent, that is, whether the locations of the first and second suspected defect areas are in the same position on the tunnel model. This can be verified by comparing spatial coordinates or using specific spatial transformations. (2) Feature consistency verification: checking whether the features extracted by the first and second apparent state detection results are consistent. This may include features such as the shape, length, and width of the first and second suspected defect areas. (3) Temporal consistency verification: if there is data at multiple time points, it is possible to check the consistency of the detection results at different time points, which is helpful for capturing the changing trend of tunnel defects.
[0092] Cross-validation helps reduce errors and improve the accuracy of detection results, ensuring a more reliable final diagnosis of the disease. The data analysis module uses the cross-validation results to locate suspected disease areas in the tunnel model.
[0093] In this embodiment, cross-validation effectively detects and corrects potential errors and inconsistencies. This cross-validation mechanism ensures more accurate and reliable disease localization, improving the credibility and effectiveness of simulation verification.
[0094] Furthermore, based on the second embodiment described above, a fifth embodiment of the method for three-dimensional detection and diagnostic simulation verification of tunnel surface defects in this application is proposed. The computer is equipped with a central control system, and the laser module is equipped with a laser scanner. The step S10, "obtaining point cloud data of the tunnel model through the laser module," is further refined, including:
[0095] Step S13: The central control system sends a point cloud acquisition command to the laser module to control the laser scanner of the laser module to acquire the point cloud data of the tunnel model and transmit the point cloud data of the tunnel model to the data analysis module.
[0096] Specifically, the computer involved in this embodiment is equipped with a central control system, which is mainly used to control various moving parts of the detection vehicle.
[0097] The laser module is equipped with a laser scanner, combined with Figure 3 By sending point cloud acquisition commands to the laser module through the central control system, the laser scanner of the laser module can be controlled to move in different directions to fully scan the tunnel segments of the tunnel model, thereby enabling the laser scanner to acquire point cloud data of the tunnel model and transmit the point cloud data of the tunnel model to the data analysis module.
[0098] The step S10, "acquiring image data of the tunnel model through the camera module," is further refined to include:
[0099] Step S14: The central control system sends an image acquisition command to the camera module to control the camera module to acquire image data of the tunnel model and transmit the image data of the tunnel model to the data analysis module.
[0100] Combination Figure 3 The central control system sends image acquisition commands to the camera module, controlling the camera module's vision sensor to capture images of the tunnel model, thereby enabling the camera module to obtain image data of the tunnel model and transmit the image data of the tunnel model to the data analysis module.
[0101] In this embodiment, the central control system precisely controls the laser scanner and camera module to ensure the synchronization and consistency of data acquisition. Through unified command scheduling, the acquisition sequence of point cloud data and image data is closely matched, improving the accuracy of data integration and analysis. This facilitates more accurate and comprehensive acquisition of tunnel surface defects and enhances the credibility and effectiveness of simulation verification.
[0102] Furthermore, based on the fifth embodiment described above, a sixth embodiment of the method for three-dimensional detection and diagnostic simulation verification of tunnel surface defects of this application is proposed. The detection trolley is equipped with a drive module, and the method further includes:
[0103] Step S021: In response to the input moving speed parameters and moving direction parameters, the central control system generates and sends a driving command to the driving module, so that the driving module moves the detection vehicle based on the driving command.
[0104] Specifically, the detection trolley involved in this embodiment is equipped with a drive module. Relevant personnel can manually input the moving speed parameter and moving direction parameter to the detection trolley, wherein: (1) Moving speed parameter: represents the desired speed of the detection trolley in the moving direction. This can be a scalar value used to control the overall speed of the detection trolley. (2) Moving direction parameter: represents the moving direction of the detection trolley. This can be an angle or a direction vector used to determine the direction in which the detection trolley moves forward.
[0105] Accordingly, the detection vehicle responds to the input moving speed and moving direction parameters, generates and sends driving commands to the drive module through the central control system, so that the drive module moves the detection vehicle based on the driving commands.
[0106] The drive module is the component responsible for actually controlling the movement of the detection cart. The drive module can be wheel-driven and specifically includes a motor, sensors, and a control system. It executes drive commands, controls the rotation of the wheels, and ensures the detection cart moves at the expected speed and direction.
[0107] In this embodiment, the precise navigation and movement of the inspection vehicle is achieved by the central control system accurately responding to the moving speed and moving direction parameters, which improves the stability and consistency of the entire inspection process, helps to obtain more accurate surface defect detection data, and improves the credibility and effectiveness of simulation verification.
[0108] Furthermore, based on the fifth embodiment described above, a seventh embodiment of the method for three-dimensional detection and diagnosis simulation verification of tunnel surface defects of this application is proposed. The detection trolley is equipped with an attitude adjustment module, and the attitude adjustment module and the laser module are connected by a transmission connection. The method further includes:
[0109] Step S031: In response to the input attitude adjustment parameters, the central control system generates and sends an attitude adjustment command to the attitude adjustment module, so that the attitude adjustment module adjusts the attitude of the laser module based on the attitude adjustment command.
[0110] Specifically, the detection vehicle involved in this embodiment is equipped with an attitude adjustment module, and relevant personnel can manually input attitude adjustment parameters into the detection vehicle. These attitude adjustment parameters are input parameters used to describe the desired attitude adjustment of the laser module, and may include attitude information such as roll, pitch, and yaw, used to guide the laser module's positioning and orientation in space.
[0111] Accordingly, the detection vehicle responds to the input attitude adjustment parameters, generates attitude adjustment commands through the central control system, and sends them to the drive module, so that the attitude adjustment module adjusts the attitude of the laser module based on the attitude adjustment commands.
[0112] In one possible implementation, the attitude adjustment module supports the laser module, and each of the four bottom corners of the attitude adjustment module has a liftable component. The lifting and lowering operation of these components is controlled by attitude adjustment parameters, giving it a more flexible ability to adjust the lifting range. By precisely adjusting the height of these liftable components, the attitude adjustment module can support the laser module and simulate various tilt states. The laser module is positioned above the attitude adjustment module; this design allows for the simulation of observing tunnel surface defects under different attitudes.
[0113] In this embodiment, the central control system precisely responds to attitude adjustment parameters, adjusting the laser module's attitude in real time to simulate various tilting conditions. This helps simulate surface defect detection scenarios in complex environments in actual engineering, improving the realism and reliability of the simulation verification. Through attitude adjustment, the performance of the laser module in multiple directions can be more comprehensively evaluated, providing more representative experimental data for integrated surface defect diagnosis.
[0114] Furthermore, based on the first embodiment described above, the eighth embodiment of the method for three-dimensional detection and diagnosis simulation verification of tunnel surface defects proposed in this application further refines step S30, which involves "inputting the point cloud data and image data corresponding to the suspected defect area into a preset machine learning model to obtain the defect diagnosis result corresponding to the suspected defect area output by the machine learning model," including:
[0115] Step S31: Extract feature information of the suspected disease area from the point cloud data and image data corresponding to the suspected disease area;
[0116] Step S32: Input the feature information of the suspected disease area into the machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model.
[0117] Specifically, feature information of the suspected disease areas is extracted from the point cloud data and image data corresponding to these areas. This feature information represents the suspected disease areas and may include features such as the shape, density, and color of the point cloud, as well as disease-related textures and edges in the image data. The purpose of this step is to transform the complex data of the suspected disease areas into a feature representation that a machine learning model can understand.
[0118] Furthermore, since the machine learning model here is a pre-trained model, it can output disease diagnosis results based on the input feature information. The disease diagnosis results include the location, type, and three-dimensional dimensions of the disease.
[0119] In one possible implementation, the machine learning model can employ a dual-driven training method, utilizing both knowledge graphs and a large amount of real-world data. The knowledge graph provides prior knowledge, while the machine learning model is trained by combining the knowledge graph and actual measurement data. This can be seen as a method that integrates theory and practice, effectively improving the output accuracy and generalization ability of the machine learning model.
[0120] In this embodiment, by extracting feature information from point cloud data and image data of suspected disease areas, the apparent characteristics of the disease can be described more comprehensively and in detail. Such feature information includes not only geometric structure but may also include information on color, texture, and other aspects. Inputting these comprehensive features into a machine learning model helps improve the model's ability to identify and analyze diseases, making disease diagnosis results more accurate and reliable.
[0121] Combination Figure 4 , Figure 4 This is a schematic diagram of the comprehensive process of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects of this application, and it presents the ninth embodiment of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects of this application.
[0122] Specifically, the detection vehicle acquires point cloud data and image data of the tunnel model through the laser module and camera module, respectively.
[0123] For the acquired point cloud data of the tunnel model: the point cloud data is preprocessed, and point cloud registration and tunnel appearance model (i.e., the 3D point cloud model of the tunnel model) are generated based on the preprocessed point cloud data. Then, surface grayscale difference analysis and stereo depth difference analysis are performed on the 3D point cloud model to obtain the first appearance state detection result of the tunnel model. The first appearance state detection result includes the appearance 3D structural information of the tunnel model, as well as the first type of suspected defect areas determined based on the appearance 3D structural information.
[0124] For the acquired tunnel model image data: Area grayscale difference analysis and area spectral value difference analysis were performed on the image data to obtain the second appearance state detection results of the tunnel model. The second appearance state detection results include the appearance visual feature information of the tunnel model, as well as the second type of suspected defect areas identified based on the appearance visual feature information.
[0125] Cross-validating the results of the first and second apparent state detections yields more accurate cross-validation results. Using these cross-validation results, suspicious defect areas in the tunnel model can be precisely located (equivalent to a highly accurate and reliable area location).
[0126] After identifying suspected defect areas in the tunnel model, these areas can be divided into several unit regions. A pre-defined clustering algorithm is then used to perform cluster analysis on these unit regions, filtering out the defective unit regions. Feature extraction is then performed on the point cloud data and image data corresponding to these defective unit regions to obtain their corresponding feature information. This feature information can include 3D dimensions, gradients, and shapes, forming a fused feature set. The purpose of this step is to transform complex data into a feature representation that a machine learning model can understand.
[0127] Furthermore, the feature information is input into a pre-obtained adaptive multi-task sparse digital image correlation model. More specifically, in the adaptive multi-task sparse digital image correlation model, A is the baseline fusion feature matrix, Y is the fusion feature matrix of the suspected region, and C... k This is a sparse correlation mapping matrix, containing the correlation coefficients between the baseline fusion feature matrix A and the fusion feature matrix Y of the suspected disease area. λ is an adaptive adjustment factor, and k is the iteration step number. The formula for the adaptive multi-task sparse digital image correlation model is shown below:
[0128]
[0129] stC≥0
[0130]
[0131] Solving for C yields a more comprehensive and accurate diagnostic result. Specifically, A is equivalent to a feature dictionary, and C is equivalent to the index number of Y in A. By finding the index number, more comprehensive and accurate information about Y can be obtained. This information can be used to add a supervised learning model to the unsupervised learning method of clustering algorithms, further optimizing the analysis results.
[0132] Alternatively, feature information can be input into a pre-prepared traditional machine learning model such as a random forest or neural network. Specifically, taking a random forest, a method that avoids overfitting, as an example, the model is trained by using feature information as input and type and grading level as output. When new feature information is input into the model, it automatically outputs the corresponding type and grading level, and can also output the accuracy rate.
[0133] Based on the proposed solution, a comprehensive inspection of the tunnel model's appearance is achieved through a simulation testing platform. Point cloud data and image data are integrated within the simulation environment to verify the accuracy of the machine learning model used for diagnosing tunnel surface defects. This avoids the limitations of real-vehicle verification and non-physical verification, effectively improving the verification effect and efficiency of the algorithm model. When the machine learning model is found to be unsatisfactory, improvements are made, thereby promoting model optimization and enhancing diagnostic performance.
[0134] Furthermore, this application also proposes a simulation test platform, which performs the steps of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described above.
[0135] Since this simulated tunnel defect diagnosis and verification program employs all the technical solutions of all the aforementioned embodiments when executed by the processor, it possesses at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be elaborated upon here.
[0136] Furthermore, this application also proposes a computer-readable storage medium storing a simulated tunnel defect diagnosis and verification program. When the simulated tunnel defect diagnosis and verification program is executed by a processor, it implements the steps of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described above.
[0137] Since this simulated tunnel defect diagnosis and verification program employs all the technical solutions of all the aforementioned embodiments when executed by the processor, it possesses at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be elaborated upon here.
[0138] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0139] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a simulation test platform (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of this application.
[0141] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A three-dimensional detection and diagnostic simulation verification method for tunnel surface defects, characterized in that, The method is applied to a simulation test platform, which includes a detection vehicle and a tunnel model. The detection vehicle is equipped with a laser module and a camera module, and the detection vehicle is equipped with a computer. The computer is equipped with a data analysis module. The method includes: The laser module acquires point cloud data of the tunnel model, and based on the point cloud data analysis, the first apparent state detection result of the tunnel model is obtained, including: The data analysis module performs surface grayscale difference analysis and stereo depth difference analysis on the point cloud data to obtain the first appearance state detection result of the tunnel model. The surface grayscale difference analysis is used to detect the grayscale value changes between adjacent point cloud regions to identify different textures, colors or brightness changes on the surface of the tunnel model. The stereo depth difference analysis is used to detect the depth value differences between different point cloud regions to identify irregular changes on the surface of the tunnel model. Image data of the tunnel model is acquired through the camera module, and a second apparent state detection result of the tunnel model is obtained based on the image data analysis, including: The data analysis module performs area grayscale difference analysis and area spectral value difference analysis on the image data to obtain the second appearance state detection result of the tunnel model. The area spectral value difference analysis detects the changes in texture and structure between different regions in the image data by performing frequency domain analysis on the image data. Based on the first apparent state detection result and the second apparent state detection result, the suspected defect areas of the tunnel model are located, including: The data analysis module performs cross-validation on the first apparent state detection result and the second apparent state detection result to obtain the cross-validation result. The data analysis module locates the suspected defect areas of the tunnel model based on the cross-validation results. The cross-validation process includes: (1) spatial consistency verification, verifying whether the first apparent state detection result and the second apparent state detection result are spatially consistent; (2) feature consistency verification, checking whether the features extracted by the first apparent state detection result and the second apparent state detection result are consistent, the features including the shape, length and width of the first type of suspected disease area and the second type of suspected disease area; (3) temporal consistency verification, when there is data at multiple time points, checking the consistency of the first apparent state detection result and the second apparent state detection result at different time points; The suspected disease area is divided into several unit areas. The several unit areas are clustered and analyzed by a preset clustering algorithm to obtain disease unit areas. Feature extraction is performed on the point cloud data and image data corresponding to the disease unit areas to obtain the corresponding feature information. The feature information is input into a preset machine learning model to obtain the disease diagnosis result corresponding to the suspected disease area output by the machine learning model. The disease diagnosis result includes the location, type, and three-dimensional size of the diagnosed disease. The machine learning model is an adaptive multi-task sparse digital image correlation model. The adaptive multi-task sparse digital image correlation model includes a baseline fusion feature matrix, a fusion feature matrix of the suspected disease area, a sparse correlation mapping matrix, the correlation coefficient between the baseline fusion feature matrix and the fusion feature matrix of the suspected disease area, and an adaptive adjustment factor. The disease diagnosis results are compared with the preset real disease information of the tunnel model, including at least one of the following: By comparing the location of the diagnosed disease with the actual location of the disease, if the locations match, it indicates that the machine learning model is accurate in locating the disease. By comparing the type of the diagnosed disease with the type of the actual disease, if the types are consistent, it indicates that the machine learning model can correctly identify different types of apparent diseases. By comparing the shape of the diagnosed disease with the shape of the actual disease, if the shapes match, it indicates that the machine learning model is accurate in capturing the shape of the disease. By comparing the size of the diagnosed disease with the size of the actual disease, if the sizes are consistent, it indicates that the machine learning model can accurately estimate the three-dimensional size of the disease. By comparing the severity of the diagnosed disease with the severity of the actual disease, if the severity is consistent, it indicates that the machine learning model can accurately assess the severity of the disease. The machine learning model is tested based on the comparison results to see if it meets the preset validity conditions. The preset real disease information includes the location, type, and three-dimensional dimensions of the real disease. The machine learning model is used to diagnose tunnel surface diseases.
2. The three-dimensional detection and diagnostic simulation verification method for tunnel surface defects as described in claim 1, characterized in that, The laser module is equipped with an encoder and an inertial sensor, and the method further includes: The encoder acquires mileage increment information and speed information while in motion; and the inertial sensor acquires speed increment information and angular velocity increment information while in motion. The step of obtaining the first appearance state detection result of the tunnel model by performing surface grayscale difference analysis and stereo depth difference analysis based on the point cloud data through the data analysis module includes: The data analysis module performs data fusion based on the point cloud data, the mileage increment information, the speed information, the speed increment information, and the angular velocity increment information to obtain a three-dimensional point cloud model of the tunnel model. The first appearance state detection result of the tunnel model is obtained by performing surface grayscale difference analysis and stereo depth difference analysis on the three-dimensional point cloud model.
3. The three-dimensional detection and diagnostic simulation verification method for tunnel surface defects as described in claim 1, characterized in that, The computer is equipped with a central control system, and the laser module is equipped with a laser scanner. The step of acquiring the point cloud data of the tunnel model through the laser module includes: The central control system sends a point cloud acquisition command to the laser module, controls the laser scanner of the laser module to acquire the point cloud data of the tunnel model, and transmits the point cloud data of the tunnel model to the data analysis module. The step of acquiring image data of the tunnel model through the camera module includes: The central control system sends image acquisition commands to the camera module, controls the camera module to acquire image data of the tunnel model, and transmits the image data of the tunnel model to the data analysis module.
4. The three-dimensional detection and diagnostic simulation verification method for tunnel surface defects as described in claim 3, characterized in that, The detection vehicle is equipped with a drive module, and the method further includes: In response to the input moving speed and moving direction parameters, the central control system generates and sends a driving command to the driving module, so that the driving module moves the detection vehicle based on the driving command.
5. The three-dimensional detection and diagnostic simulation verification method for tunnel surface defects as described in claim 3, characterized in that, The detection vehicle is equipped with an attitude adjustment module, which is connected to the laser module via a transmission connection. The method further includes: In response to the input attitude adjustment parameters, the central control system generates and sends attitude adjustment commands to the attitude adjustment module, so that the attitude adjustment module adjusts the attitude of the laser module based on the attitude adjustment commands.
6. A simulation test platform, characterized in that, The simulation test platform performs the steps of implementing the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described in any one of claims 1-5.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a simulated tunnel defect diagnosis and verification program, which, when executed by a processor, implements the steps of the three-dimensional detection and diagnosis simulation verification method for tunnel surface defects as described in any one of claims 1-5.