Cabin assembly intelligent detection method and system based on unmanned aerial vehicle

Abstract

The invention provides a cabin assembly intelligent detection method and system based on an unmanned aerial vehicle, and the method comprises the steps: comprehensively collecting the point cloud of the inner surface of a cabin through the unmanned aerial vehicle, and forming an original point cloud data set; extracting all theoretical structural part models from the design file of the cabin and converting the theoretical structural part models into theoretical point cloud models; after an original point cloud data set is preprocessed, point cloud splicing and mapping are carried out to obtain an actual measurement point cloud model; performing multi-scale registration on the actually measured point cloud model and the theoretical point cloud model, and completing accurate alignment in space; and performing multi-level difference calculation on the actually-measured point cloud model and the theoretical point cloud model after accurate alignment to form an assembly detection result. According to the invention, the cost and the safety risk can be effectively reduced in the detection process, and the detection precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the field of assembly detection, and in particular to an unmanned aerial vehicle-based intelligent detection method and system for cabin assembly. Background Art

[0002] With the rapid development of modern industry, the demand for intelligent detection in complex environments has become more urgent, especially in complex and closed scenes such as shipbuilding, energy facilities and underground spaces. These environments usually have complex structures, insufficient light, poor ventilation and other characteristics, which pose huge challenges to traditional detection methods. Take the cabin in shipbuilding as an example. The cabin has a complex structure and a small space, which requires extremely high assembly precision. Its assembly quality and integrity directly affect the safety and service life of the ship. Therefore, how to achieve efficient and accurate detection in such environments has become a problem that needs to be solved urgently.

[0003] At present, cabin inspection mainly relies on manual visual inspection, but there are many limitations. First, manual inspection is costly and requires the construction of auxiliary facilities such as scaffolding, which is not only time-consuming and inefficient, but also poses certain safety risks. Secondly, manual inspection has poor accuracy and the inspection results are affected by human factors. It is difficult to accurately locate the specific location and details of assembly defects (such as component ID), which can easily lead to missed inspections or false inspections. In addition, manual inspection lacks a systematic data management method, making it difficult to effectively manage and trace large-scale inspection data, resulting in inefficient review and comparison inspections.

[0004] As the shipbuilding industry's requirements for detection accuracy and efficiency increase, traditional manual detection methods are difficult to meet. Existing unmanned equipment (such as drones and robots) are flexible in certain environments, but they are still insufficient in positioning, data processing and automated analysis capabilities in complex spaces, and it is difficult to completely replace manual labor to achieve comprehensive and reliable detection. Limited signals in complex environments (such as the lack of GPS signals), flexible control in small spaces, and high-precision requirements further limit the detection application of existing unmanned equipment.

[0005] In summary, shipbuilding urgently needs an innovative detection solution that can improve detection accuracy while reducing costs, and realize systematic management and analysis of detection data to meet the high-precision detection needs of ship assembly. Summary of the invention

[0006] In view of the problems existing in the prior art, a method and system for intelligent detection of cabin assembly based on drones are provided. By using the lidar and IMU equipment carried by the drone, high-precision source data information is collected, and a high-precision digital model is generated through offline mapping. The digital model is compared with the theoretical model to realize automatic detection and analysis of the integrity of the cabin assembly.

[0007] The first aspect of the present invention proposes a method for intelligent detection of cabin assembly based on a drone, comprising:

[0008] The point cloud of the inner surface of the cabin is comprehensively collected by a drone to form an original point cloud data set; the drone is equipped with at least a laser radar and an inertial measurement unit;

[0009] All theoretical structural component models are extracted from the design files of the cabin and converted into theoretical point cloud models; after preprocessing the original point cloud data set, the point cloud is spliced ​​and mapped to obtain the measured point cloud model;

[0010] Perform multi-scale registration of the measured point cloud model and the theoretical point cloud model to achieve precise alignment in space;

[0011] Multi-level difference calculation is performed on the measured point cloud model and the theoretical point cloud model after precise alignment to form the assembly inspection result.

[0012] As a preferred solution, the formation process of the theoretical point cloud model specifically includes:

[0013] Extract all theoretical structural component models with unique IDs from the cabin design files;

[0014] The surfaces of all theoretical structural component models are sampled into dense point clouds to form a theoretical point cloud model.

[0015] As a preferred solution, the process of forming the measured point cloud model specifically includes:

[0016] De-noising the original point cloud dataset;

[0017] All point cloud data are stitched and registered to generate a complete measured point cloud model.

[0018] As a preferred solution, the multi-scale registration specifically includes:

[0019] Perform point cloud denoising on the measured point cloud model and the theoretical point cloud model respectively to remove isolated points and noise;

[0020] Downsample the measured point cloud model and the theoretical point cloud model respectively to reduce the point cloud density of the model;

[0021] According to the data of the inertial measurement unit, the local coordinate system of the measured point cloud model is converted into the same coordinate system as the theoretical point cloud model;

[0022] Normalize the scale of the measured point cloud model and the theoretical point cloud model respectively;

[0023] Extract the feature points of the measured point cloud model and the theoretical point cloud model after scale normalization, and complete the rough registration and alignment by matching the feature points of the two models to form a rough registered measured point cloud model;

[0024] The nearest neighbor point of each point in the measured point cloud model is found in the theoretical point cloud model, and the optimal transformation matrix is ​​iteratively calculated to minimize the square error of the Euclidean distance between the two point clouds. Based on the optimal transformation matrix, precise alignment is completed to form a multi-scale registered measured point cloud model.

[0025] As a preferred solution, the multi-level difference calculation specifically includes:

[0026] Calculate the global error between the theoretical point cloud model after multi-scale registration and the measured point cloud model;

[0027] Decompose the model into several independent components based on component ID, and calculate the geometric difference of each component in the two models;

[0028] Construct a topological relationship network for each component’s ID information and calculate the geometric relationship differences with adjacent components;

[0029] For the areas where the error in the component-level geometric difference is greater than the threshold, fine difference calculation of local alignment is performed based on the component ID;

[0030] According to the level and nature of each difference, the difference calculation results based on the component ID are classified and quantified to obtain the difference values ​​of the point cloud.

[0031] As a preferred solution, it also includes a visual display of the assembly inspection results, specifically including: building a three-dimensional visualization model of the cabin and marking the corresponding component IDs; in the three-dimensional visualization model, mapping the point cloud difference values ​​of each component into a color gradient, and using different colors to represent errors of different sizes.

[0032] The second aspect of the present invention proposes a cabin assembly intelligent detection system based on a drone, comprising:

[0033] The online data acquisition and processing system comprises an unmanned aerial vehicle platform and a portable controller; the unmanned aerial vehicle platform is used to fly in the cabin to collect point cloud data of the inner surface of the cabin; the portable controller is used to control the flight of the unmanned aerial vehicle platform;

[0034] The offline intelligent detection and analysis system is used to generate a theoretical point cloud model of the cabin and a measured point cloud model based on the point cloud data collected by the online data acquisition and processing system. It also calculates multi-level differences after multi-scale registration of the theoretical point cloud model and the measured point cloud model, and outputs and displays the difference results of different categories.

[0035] As a preferred solution, the offline intelligent detection and analysis system includes:

[0036] The model reconstruction and generation module is used to generate a theoretical point cloud model with component ID according to the design file of the cabin, and to reconstruct the collected point cloud data of the cabin inner surface into a measured point cloud model;

[0037] Model multi-scale registration module, used to realize the registration and alignment of theoretical point cloud model and measured point cloud model;

[0038] The model multi-level difference calculation module is used to perform difference calculation on the aligned model and output the difference results of different categories;

[0039] Result visualization and marking module, used to display the difference results in the visualization interface and mark the corresponding component ID;

[0040] The data storage and query module is used to provide data storage and query functions.

[0041] As a preferred solution, the specific working process of the model multi-scale registration module includes: first, removing the noise and outliers in the point cloud in the model; then downsampling the point cloud and converting the coordinate system; then normalizing the scale of the point cloud data of the model; then using feature matching or global optimization methods to perform coarse registration and alignment; finally, based on the coarse registration results, an iterative optimization algorithm is used to achieve precise registration and alignment of the theoretical point cloud model and the measured point cloud model.

[0042] As a preferred solution, the specific working process of the model multi-level difference calculation module includes: first, calculating the global error between the theoretical point cloud model after multi-scale alignment and the measured point cloud model; then, decomposing the model into several independent components based on the component ID, and calculating the geometric difference of each component in the two models respectively; then, constructing a topological relationship network for the ID information of each component, and calculating the geometric relationship difference with adjacent components; then, for the local areas with large errors in the component-level geometric differences, performing fine difference calculation of local alignment based on the component ID; finally, classifying and quantifying each difference, and outputting the quantified difference results.

[0043] Compared with the prior art, the beneficial effects of adopting the above technical solution are:

[0044] 1. The present invention can effectively reduce costs and safety risks during the detection process. By using drones for high-precision data collection and processing, it can flexibly cover various areas in the cabin under various lighting conditions and complex spatial structure environments, greatly reducing the detection time and economic costs. At the same time, there is no need for high-altitude operations, which effectively avoids the safety hazards in traditional detection and improves the overall operation safety.

[0045] 2. The present invention significantly improves detection accuracy and reliability. By using high-precision sensing equipment such as LiDAR and IMU, a measured point cloud model of the interior of the cabin is generated, and a multi-level difference comparison analysis is performed with the theoretical point cloud model. Multi-level difference calculation can accurately identify assembly errors and defects, reduce missed detections and false detections, and provide higher accuracy and consistency for assembly quality assessment.

[0046] 3. The present invention realizes standardized storage and query of data. By associating the inspection data with the component ID, a complete data management process can be realized, including the original point cloud data, difference analysis results and related reports, etc., which greatly improves the data traceability and management efficiency, and provides reliable data support for subsequent review and troubleshooting.

[0047] 4. The present invention supports the generation of multi-level analysis reports and data visualization, which is convenient for tracking and maintaining the assembly quality of ships. Detailed global and local error analysis reports can be generated, and component-level geometric differences, topological associations, etc. can be visually annotated. The report can be queried and indexed by component ID, which is convenient for quickly locating and tracking test results in subsequent assembly quality optimization and maintenance, improving the data utilization value and user experience.

[0048] 5. The present invention has good scalability and compatibility, and can be widely used in other quality inspection fields of complex structures. The system is designed with an open API interface and data storage structure, which supports integration with third-party systems, making it not only suitable for ship assembly inspection, but also for energy facilities, underground space and other scenarios, meeting the inspection needs of various complex scenarios and improving the applicability and scalability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of the intelligent detection method for cabin assembly based on drones proposed by the present invention.

[0050] Figure 2 This is a schematic diagram of the composition of the UAV-based cabin assembly intelligent detection system proposed by the present invention. DETAILED DESCRIPTION

[0051] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as limitations on the present application. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0052] The current cabin assembly inspection mainly has the following main problems:

[0053] (1) High cost of manual inspection: Traditional manual inspection requires the construction of auxiliary facilities such as scaffolding, which is costly and poses safety hazards. Especially in complex cabin structures, the time and economic cost of building scaffolding is higher, and there are safety risks in high-altitude operations.

[0054] (2) Poor accuracy of manual inspection: When the light intensity inside the cabin is insufficient, manual inspection (only through visual inspection or image) is easily affected and it is difficult to accurately locate the specific location and details (such as component ID) of assembly defects. The inspection results are easily limited by the experience and viewing angle of the inspectors, resulting in frequent missed inspections and false inspections, especially small or hidden defects that are difficult to find and quantify.

[0055] (3) Difficulty in managing manual inspection data: Manually recorded information is limited and lacks systematic and digital management methods, making it difficult to support effective review and comparative inspection. Traditional manual inspections are mostly based on paper records, with inconsistent data formats and difficulty in systematic analysis, which affects the efficiency of review and data traceability.

[0056] Based on this, the present invention proposes a cabin assembly intelligent detection method based on drones. The drones are used to collect assembly data, and then the assembly difference detection is completed in combination with the theoretical design model. This can effectively reduce the detection cost, improve the detection accuracy, and realize the systematic management of data. Please refer to Figure 1 , the specific plan is as follows:

[0057] Step 1: Use drones to comprehensively collect point clouds of the interior surface of the cabin to form an original point cloud dataset.

[0058] In this embodiment, the drone is equipped with a laser radar and an IMU (inertial measurement unit). The drone platform is remotely controlled by a portable controller, and the drone platform itself integrates the laser radar and IMU data, so that the drone platform can accurately collect real-time data in complex and narrow environments, thereby ensuring that the entire operation process can collect high-quality source data for subsequent work analysis. In actual applications, the drone can also be equipped with a pan-tilt camera and a fill light to further ensure the accuracy of data collection.

[0059] It should be noted that when collecting data, it is necessary to ensure that the attitude of the drone is stable, so as to reduce the spatial error introduced by tilt and drift. In this embodiment, a positioning and navigation system based on SLAM (Simultaneous Localization and Mapping) is used in a complex cabin environment for real-time positioning and attitude control to ensure the high-precision flight stability of the drone in a limited space. In order to avoid overlapping flight paths and improve collection efficiency, the flight route can be planned in advance so that the drone can cover all detection areas inside the cabin as much as possible.

[0060] The multi-source data collection and processing is completed through the laser radar, IMU and gimbal camera carried by the drone to ensure accurate recording of the geometry and image information in the cabin. Among them, the laser radar performs three-dimensional scanning at a fixed angle and frequency to record the point cloud of the surface of the cabin, and the sampling frequency is f s , record point cloud dataset (x j ,y j ,z j However, since the drone has different postures during flight, in this embodiment, the drone posture data (q j ,t j ), including quaternion or Euler angle and position data, and then use the attitude data to compensate for the attitude error during acquisition to obtain the original point cloud data set P B :P B ={(x j ,y j ,z j ,q j ,t j )|j=1,2,…,M}, where M represents the number of collected point clouds.

[0061] Step 2: Extract all theoretical structural component models from the cabin design files and convert them into theoretical point cloud models; after preprocessing the original point cloud data set, the point cloud is spliced ​​and mapped to obtain the measured point cloud model.

[0062] In order to realize difference detection, it is necessary to first establish a measured point cloud model representing the actual scene and a theoretical point cloud model representing the theory. The generation process of the theoretical point cloud model and the measured point cloud model is described in detail below.

[0063] (1) For the theoretical point cloud model

[0064] In this embodiment, a theoretical component model with a unique ID is extracted from the design file of the cabin (in this embodiment, a CAD three-dimensional design model file is used as an example), which contains the geometric information and component ID of all key components in the cabin, and defines the position and direction of each component in a global coordinate system. The global coordinate system is a customizable coordinate system origin for subsequent registration between models.

[0065] The model of the design file is discretized and the surface of the CAD model is sampled into a dense point cloud (x i ,y i ,z i ), and then form a theoretical point cloud model A:P A ={(x i ,y i ,z i)|i=1,2,…,N}, where N represents the number of point clouds of the theoretical model A. In order to ensure high-precision details, this embodiment generates point clouds with reasonable sampling intervals through equidistant sampling or patch sampling. In the actual sampling process, the sampling resolution Δd can be used to balance the amount of sampled data and accuracy. Based on this, the discretization point cloud generation process of the theoretical point cloud model can also be expressed as: Among them, S k represents the discrete point set of the kth component, and K is the total number of components.

[0066] (2) For the measured point cloud model

[0067] Before generating the measured point cloud model, it is necessary to collect the original point cloud dataset P B Noise elimination is performed and a unified expression of point cloud data in the global coordinate system is achieved, providing high-quality input for subsequent point cloud stitching. In this embodiment, noise points in the acquisition process are mainly removed by statistical filtering (such as K-nearest neighbor algorithm) or voxel filtering method to ensure data reliability.

[0068] Using SLAM algorithm to analyze the point cloud dataset (x j ′,y j ′,z j ′) to stitch and align, and then generate a complete measured point cloud model B: Where L is the total number of merged point clouds. In order to minimize the point cloud stitching error, the drone's pose information can be optimized through nonlinear optimization algorithms (such as ICP or backend processing based on graph optimization). At the same time, the accuracy and consistency of the point cloud model are improved through global constraint optimization (such as closed loop detection).

[0069] Through the above-mentioned generation and pre-processing process of the theoretical point cloud model A and the measured point cloud model B, the consistency of the two in terms of spatial reference, density and scale is ensured. This provides a solid foundation for the subsequent multi-scale model registration and difference calculation, and helps to improve the accuracy and reliability of assembly detection.

[0070] Step 3: Perform multi-scale registration on the measured point cloud model and the theoretical point cloud model to achieve precise alignment in space.

[0071] In this embodiment, multi-scale registration can realize the spatial alignment of the measured point cloud model and the theoretical point cloud model, laying the foundation for difference calculation. Multi-scale registration mainly includes noise reduction, downsampling, coordinate system conversion, point cloud standardization, coarse registration alignment and fine registration alignment to ensure the consistency of the two in spatial reference, scale and density. The following is a description of each step of multi-scale registration.

[0072] (1) Noise reduction

[0073] Denoising the point cloud of the model can remove isolated points and noise. In this embodiment, statistical filtering is used for noise reduction, that is, the average distance D of the k-neighborhood of each point in the model is calculated. i , if the distance exceeds the set threshold d th , it is determined to be a noise point, specifically:

[0074]

[0075] If D i >d th , then filter out point p i , and finally obtain the theoretical point cloud model after denoising And measured point cloud model

[0076] (2) Downsampling

[0077] In order to reduce the amount of point cloud calculation and improve the efficiency of registration calculation, in this embodiment, voxel downsampling is used to reduce the point cloud density, that is, the model point cloud is divided into v sample The center point of each voxel is taken as the representative point.

[0078]

[0079] Finally, the representative point p voxel Theoretical point cloud model after composition downsampling And measured point cloud model

[0080] (3) Coordinate system conversion

[0081] In order to facilitate the registration of the theoretical point cloud model with the measured point cloud model, in this embodiment, according to the IMU data, the local coordinate system of the measured point cloud model B is converted to the same global coordinate system as the theoretical point cloud model A. Specifically, the measured point cloud model is converted into the global coordinate system using the rotation matrix R and the translation vector t. Convert to global coordinates:

[0082]

[0083] Among them, the rotation matrix R and the translation vector t can be calculated from the IMU data or external parameter calibration data, and finally the measured point cloud model aligned with the global coordinates is obtained.

[0084] (4) Point cloud standardization

[0085] In order to provide a unified spatial reference during registration, the theoretical point cloud model and the measured point cloud model need to be consistent in scale. In this embodiment, the theoretical point cloud model and the measured point cloud model are Perform scale normalization to eliminate coordinate scale differences:

[0086]

[0087] in, is the normalized theoretical point cloud model, is the normalized measured point cloud model, μ is the centroid of the point cloud, and σ is the scale factor (such as the standard deviation or maximum and minimum value range of the point cloud).

[0088] (5) Rough registration and alignment

[0089] After completing the scale normalization of the theoretical point cloud model and the measured point cloud model, a feature-based rough registration algorithm (RANSAC is used as an example in this embodiment) is used for preliminary alignment. Specifically, the theoretical point cloud model is extracted by SIFT (Scale Invariant Feature Transform) or ISS (Intrinsic Shape Signature). And measured point cloud model The feature points of the two models are matched using the RANSAC algorithm to estimate the initial transformation matrix T initial :

[0090]

[0091] Based on the transformation matrix T initial , you can generate a roughly aligned measured point cloud model

[0092] (6) Precision alignment

[0093] In this embodiment, the ICP (Iterative Closest Point) algorithm is used to perform fine alignment based on the results of the rough alignment. First, the rough alignment results are searched by nearest neighbor. Find the nearest neighbor of each point in the point cloud A; then iteratively calculate the optimal transformation matrix T final To minimize the squared Euclidean distance error between two point clouds:

[0094]

[0095] When the error change is less than the threshold ∈ or reaches the maximum number of iterations N max Stop iteration when .

[0096] Based on the optimal transformation matrix T final , you can generate a precisely aligned measured point cloud model at this time

[0097] Through the process of multi-scale registration, the measured point cloud model and the theoretical point cloud model are accurately aligned in space, providing accurate and unified basic data for subsequent difference detection.

[0098] Step 4: Perform multi-level difference calculation on the measured point cloud model and the theoretical point cloud model after precise alignment to form the assembly detection result.

[0099] In this embodiment, the multi-level difference mainly includes global error calculation, component-level geometric difference calculation based on ID, topological correlation analysis, and fine difference analysis of local alignment. By introducing the multi-level error analysis method associated with component ID, it is ensured that each component maintains unique identification in the difference analysis and accurately locates the assembly deviation source. The difference calculation process at each level is explained one by one below.

[0100] (1) Global error calculation

[0101] The global error is mainly used to evaluate the overall assembly deviation, which can be calculated by calculating the average point-to-point distance E global accomplish:

[0102]

[0103] in, and is the matching point pairs between the measured point cloud model and the global point cloud in the theoretical point cloud model after precise alignment, and N is the number of matching point pairs.

[0104] (2) Component-level geometric difference calculation based on ID

[0105] The point cloud model is decomposed into independent components using component IDs, and the geometric differences of each component in the two models are calculated. Specifically, the point clouds of the measured point cloud model and the theoretical point cloud model are grouped by component ID, and the subsets associated with the component ID are and Among them C k represents component k; for each component C k , calculate the geometric difference between the point cloud in the measured point cloud model and the theoretical point cloud model

[0106]

[0107] Among them, N k For component C k The number of point pairs, and are corresponding point pairs within the component.

[0108] Finally, the component difference matrix based on ID is formed

[0109] (3) Topological correlation difference analysis

[0110] A topological relationship network is constructed for each component ID information to analyze the geometric relationship differences between adjacent components to detect possible offset or angle errors in assembly. Specifically, for each component adjacency matrix A based on the component ID, if component C i and C j Adjacent, then A ij =1; otherwise A ij = 0. For adjacent components (C i ,C j ) Calculate the distance deviation D between the center points ij and the angle deviation θ ij :

[0111]

[0112] Finally, the topological difference matrix {D ij ,θ ij}, used to analyze assembly deviations in geometric relationships between components.

[0113] (4) Fine difference analysis of local alignment

[0114] For component level geometry differences In the local area where the error exceeds the threshold γ, in this embodiment, the fine difference calculation of local alignment is performed based on the component ID. Specifically, in the component area with large error, the ICP algorithm is used to perform secondary registration based on the component ID, and the difference E is calculated based on the secondary registration. local :

[0115]

[0116] Where M is the number of point pairs in the local area, and are corresponding point pairs in the local area.

[0117] The local difference matrix of the final generated component {E local}, further evaluate the assembly deviation in the area with larger error.

[0118] Based on the above-mentioned difference calculation results, further classification and quantification are required to accurately evaluate the assembly quality, as follows:

[0119] (1) Global error measurement: By calculating the global error E global The mean and variance of

[0120] (2) Component-level error measurement: difference matrix based on component ID The maximum, minimum and average values ​​of .

[0121] (3) Topological difference measurement: the distance deviation D between components in the adjacency matrix A ij and the angle deviation θ ij Statistical analysis.

[0122] (4) Local difference measure: local difference matrix E local distribution characteristics.

[0123] Based on the results of various difference measurements, a difference classification report, i.e., the assembly test results, is generated, providing comprehensive quantitative assembly deviation data to support quality control and optimized design. This embodiment can accurately identify the assembly deviation of each component by combining the information of the component ID for multi-level difference calculation, and comprehensively analyze various errors that occur in the assembly process from the global to the local, providing key data for assembly quality control and maintenance optimization.

[0124] In order to facilitate users to display multi-level difference calculation results, the assembly detection method also includes visual display of assembly detection results. This process combines component ID to realize difference marking so that users can clearly understand the distribution and characteristics of assembly errors.

[0125] Specifically, the multi-level difference calculation results are mainly displayed by constructing a 3D visualization model of the cabin. In this 3D visualization model, the ID of each component is marked, and the difference of each component is mapped using an error heat map. That is, the point cloud difference value (i.e., the assembly inspection result) is mapped to a color gradient (for example, from blue to red represents small to large errors), and the geometric deviation of each component is mapped to a color gradient (for example, from blue to red represents small to large errors). Display a heat map.

[0126]

[0127] Where f is the color mapping function, p is the point in the component, is the error value of the corresponding component.

[0128] In practical applications, an enlarged view can be provided in the three-dimensional visualization model, and subtle differences can be displayed through dynamic perspective and color enhancement. The visualization display method proposed in this embodiment can provide users with an intuitive difference analysis view. In addition, the difference calculation results and visualization results can also be used to generate analysis reports in various forms, including charts, tables and text descriptions. The display forms of various difference results include but are not limited to the following methods:

[0129] (1) Global error statistics table: Based on the global error E global The statistical analysis generates tables showing indicators such as mean, maximum value and variance.

[0130] (2) Component-level error report: geometric differences for each component ID Provides independent analysis results, including statistical information such as error distribution graph, error range and standard deviation.

[0131] (3) Topological correlation analysis diagram: Generate a topological analysis report based on the component adjacency relationship, showing the distance deviation D between each component. ij and the angle deviation θ ij .

[0132] (4) Local detailed difference diagram: For components with significant local differences, detailed analysis diagrams and comparison results are provided.

[0133] Automatically generate analysis reports containing charts and text, and export them to PDF or Word file formats for easy archiving and sharing.

[0134] In addition, in order to ensure the traceability and convenient query of subsequent data, the assembly inspection method also includes a data storage process, which provides support for assembly quality control, decision optimization and subsequent maintenance through scientific storage and display solutions.

[0135] In this embodiment, the original point cloud data of the theoretical point cloud model and the measured point cloud model, the preprocessed point cloud data, and the calculation results of each step should be stored uniformly to ensure the integrity of the data. When storing, the data involved can be divided into three categories: original data, intermediate data, and difference analysis data for storage. Among them, the original data includes the collected source data, the theoretical design file model data, and the preprocessed point cloud data of the theoretical point cloud model A and the measured point cloud model B; the intermediate data includes the point cloud model data after noise reduction, downsampling, and standardization; the difference analysis data includes the global error E global , component level error matrix Topological difference matrix {D ij ,θ ij}、Local difference matrix {E local}, image or 3D file format of visualization results and analysis report, etc. Numerical data can use standardized data formats (such as CSV, JSON or HDF5). The data storage method provided in this embodiment generally generates a structured data storage file with complete version information and file association, which can ensure long-term preservation and convenient access to data.

[0136] Based on the data storage process, this embodiment also provides corresponding data retrieval and query functions, supports query functions based on component ID and query functions based on analysis steps, and at the same time, through the design of standardized API interfaces, supports system integration and data call, so that third-party systems can access and utilize stored data; through the design of data visualization interfaces, provides a real-time visualization display interface for query results, supports rotation, zooming and multi-dimensional screening, and realizes in-depth analysis of data.

[0137] Please refer to Figure 2 The embodiment of the present invention also provides a cabin assembly intelligent detection system based on a drone, which mainly includes two parts: an online data acquisition and processing system and an offline intelligent detection and analysis system. The online data acquisition and processing system includes a drone platform and a portable controller; the drone platform is used to fly in the cabin to collect cabin inner surface point cloud data; the portable controller is used to control the flight of the drone platform; the offline intelligent detection and analysis system is used to generate a theoretical point cloud model of the cabin and a measured point cloud model based on the point cloud data collected by the online data acquisition and processing system, and calculate multi-level differences after multi-scale registration of the theoretical point cloud model and the measured point cloud model, and output and display the difference results of different categories.

[0138] Specifically, the UAV platform mainly includes a flight subsystem, a payload subsystem, and a computing subsystem. Among them, the flight subsystem includes the body structure, flight control module and power module, which mainly realizes the flight function of the UAV platform; the payload subsystem mainly includes laser radar, IMU (inertial measurement unit), gimbal camera and fill light, etc., which realizes the data collection function of the UAV platform for the cabin; the computing subsystem mainly includes data collection module, data processing module and data storage module, which mainly realizes the acquisition and storage of data collected by the payload subsystem. In the data collection stage, the user remotely controls the UAV platform carrying laser radar and IMU to make it fly stably in complex environments such as cabins (such as lack of GPS signals and narrow spaces), and cooperates with the fill light effect to overcome the interference caused by insufficient light and narrow space, thereby realizing efficient collection of accurate data.

[0139] The offline intelligent detection and analysis system is mainly composed of a model reconstruction and generation module, a model multi-scale registration module, a model multi-level difference calculation module, a result visualization and marking module, and a data storage and query module. By importing the source data collected by the drone system and the theoretical three-dimensional model with ID into the various modules of the intelligent detection and analysis system, each module performs corresponding processing and analysis on the two models, and finally obtains the detection results and can realize the management and query of the detection data system, thereby achieving the purpose of reducing detection costs, improving detection accuracy and efficiency, and providing intelligent and digital support for ship maintenance. The following is a detailed description of the functions implemented by each module in the offline intelligent detection and analysis system:

[0140] Model reconstruction and generation module, this module mainly realizes the generation of theoretical point cloud model and the reconstruction of measured point cloud model. Among them, the generation of theoretical point cloud model includes: converting the three-dimensional theoretical model (design file) with component ID into theoretical point cloud model A in point cloud format, ensuring that the model has unique identification information. The reconstruction of measured point cloud model includes: using high-precision algorithm to reconstruct the source data collected by the online data acquisition and processing system into measured point cloud model B, using point cloud format to express, and realizing high-precision presentation of measured data.

[0141] Model multi-scale registration module, this module mainly performs the complete process from data denoising to refined alignment to achieve precise registration and alignment of the theoretical point cloud model and the measured point cloud model. First, advanced denoising technology is used to remove noise and outliers in the point cloud to improve data quality; then the point cloud is downsampled and the coordinate system is converted to ensure data consistency and processing efficiency; then in the standardization stage, the point cloud data is normalized to facilitate subsequent rough registration; rough registration uses feature matching or global optimization methods for preliminary alignment, laying the foundation for subsequent fine registration; finally, based on the denoising and rough registration results, an iterative optimization algorithm is used to achieve high-precision alignment to ensure accurate matching of the theoretical point cloud model A and the measured point cloud model B.

[0142] The model multi-level difference calculation module is mainly used to accurately identify and locate the assembly deviation between the theoretical point cloud model after multi-scale registration and the measured point cloud model. The main process includes: first, the overall assembly deviation is evaluated through global error calculation to provide a preliminary analysis basis. Then, the model is decomposed into independent components using component-level geometric difference analysis, and the difference of each component is carefully analyzed to generate a difference matrix. Subsequently, in the topological correlation analysis, an adjacency matrix is ​​constructed to identify the geometric relationship differences between adjacent components, thereby detecting potential assembly offsets. In the local fine difference analysis stage, the local ICP algorithm is applied to the area with large errors for high-precision secondary registration, and the fine difference calculation of local alignment is performed based on the component ID. Finally, the difference calculation results are classified and quantified to provide comprehensive assembly deviation data support. Through the model multi-level difference calculation module, not only can various errors in the assembly process be comprehensively analyzed, but also key data support is provided for assembly quality control and optimization, ensuring effectiveness and reliability in practical applications.

[0143] Result visualization and marking module, this module is mainly used to visualize the difference results into three-dimensional display results and corresponding component ID marks, to help users clearly identify the distribution and characteristics of assembly errors. In this module, the point cloud differences are displayed in the form of a three-dimensional heat map, color mapping is used to highlight the errors, and the component ID is marked for quick positioning. For areas with large local deviations, a magnified view is provided to show subtle differences. In addition, this module also supports the automatic generation of analysis reports in various forms, including global error statistics, component-level error analysis, topological correlation reports, and local fine difference maps. The analysis report can be exported to PDF or Word format for easy archiving and sharing.

[0144] Data storage and query module, this module mainly realizes all data generated during the assembly inspection process to ensure the integrity and traceability of the data. At the same time, based on the data storage results, a data retrieval system is established, combined with database indexing and file management functions, to achieve rapid retrieval and query of various analysis results and original data. The data retrieval system supports the following two query methods:

[0145] (1) Query based on component ID: Users can retrieve the geometric difference, topological difference and local error analysis report of a specific component by component ID.

[0146] (2) Query by analysis steps: Query by category according to the analysis steps (such as global error, component-level error, local difference, etc.) to quickly locate the required analysis results.

[0147] In addition, the module also provides data visualization interface and API interface. Among them, the data visualization interface is used to provide a query interface for visualization results, supporting rotation, zooming and multi-dimensional screening to achieve in-depth analysis of data. The API interface is a standardized API interface designed to support system integration and data call, enabling third-party systems to access and use stored data. The data storage and query module provides a friendly data retrieval method to achieve structured query, retrieval and display of data, improve data reuse rate and utilization efficiency, and provide technical support for later decision support.

[0148] It should be supplemented that the functions of each module of the offline intelligent detection and analysis system in the UAV-based cabin assembly intelligent detection system proposed in this embodiment can be achieved through the aforementioned UAV-based cabin assembly intelligent detection method, which will not be elaborated here.

[0149] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program includes a program code for executing the method shown in the flowchart.

[0150] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device or device or used in combination with it. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0151] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0152] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0153] As another aspect, the present application further provides a computer program product or a computer program, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the intelligent detection method for cabin assembly based on a drone as described in the above embodiment.

[0154] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the intelligent detection method for cabin assembly based on a drone described in the above embodiment.

[0155] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0156] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0157] For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific situations; the drawings in the embodiments are used to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0158] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An intelligent detection method for cabin assembly based on drones, characterized in that: include: The point cloud of the inner surface of the cabin is comprehensively collected by a drone to form an original point cloud data set; the drone is equipped with at least a laser radar and an inertial measurement unit; All theoretical structural component models are extracted from the design files of the cabin and converted into theoretical point cloud models; after preprocessing the original point cloud data set, the point cloud is spliced ​​and mapped to obtain the measured point cloud model; Perform multi-scale registration of the measured point cloud model and the theoretical point cloud model to achieve precise alignment in space; Multi-level difference calculation is performed on the measured point cloud model and the theoretical point cloud model after precise alignment to form the assembly inspection result.

2. The intelligent detection method for cabin assembly based on drone according to claim 1 is characterized in that: The formation process of the theoretical point cloud model specifically includes: Extract all theoretical structural component models with unique IDs from the cabin design files; The surfaces of all theoretical structural component models are sampled into dense point clouds to form a theoretical point cloud model.

3. The intelligent detection method for cabin assembly based on drone according to claim 1 or 2, characterized in that: The formation process of the measured point cloud model specifically includes: De-noising the original point cloud dataset; All point cloud data are stitched and registered to generate a complete measured point cloud model.

4. The intelligent detection method for cabin assembly based on drone according to claim 1 is characterized in that: The multi-scale registration specifically includes: Perform point cloud denoising on the measured point cloud model and the theoretical point cloud model respectively to remove isolated points and noise; Perform point cloud downsampling on the measured point cloud model and the theoretical point cloud model respectively to reduce the point cloud density of the model; According to the data of the inertial measurement unit, the local coordinate system of the measured point cloud model is converted into the same coordinate system as the theoretical point cloud model; Normalize the scale of the measured point cloud model and the theoretical point cloud model respectively; Extract the feature points of the measured point cloud model and the theoretical point cloud model after scale normalization, and complete the rough registration and alignment by matching the feature points of the two models to form a rough registered measured point cloud model; The nearest neighbor point of each point in the measured point cloud model is found in the theoretical point cloud model, and the optimal transformation matrix is ​​iteratively calculated to minimize the square error of the Euclidean distance between the two point clouds. Based on the optimal transformation matrix, precise alignment is completed to form a multi-scale registered measured point cloud model.

5. The intelligent detection method for cabin assembly based on drone according to claim 1 is characterized in that: The multi-level difference calculation specifically includes: Calculate the global error between the theoretical point cloud model after multi-scale registration and the measured point cloud model; Decompose the model into several independent components based on component ID, and calculate the geometric difference of each component in the two models; Construct a topological relationship network for each component’s ID information and calculate the geometric relationship differences with adjacent components; For the areas where the error in the component-level geometric difference is greater than the threshold, fine difference calculation of local alignment is performed based on the component ID; According to the level and nature of each difference, the difference calculation results based on the component ID are classified and quantified to obtain the difference values ​​of the point cloud.

6. The intelligent detection method for cabin assembly based on drone according to claim 1 is characterized in that: It also includes a visual display of the assembly inspection results, specifically including: building a three-dimensional visualization model of the cabin and marking the corresponding component IDs; in the three-dimensional visualization model, mapping the point cloud difference values ​​of each component into a color gradient, and using different colors to represent errors of different sizes.

7. An intelligent detection system for cabin assembly based on drones, characterized in that: include: An online data acquisition and processing system includes an unmanned aerial vehicle platform and a portable controller; the unmanned aerial vehicle platform is used to fly in the cabin to collect point cloud data of the inner surface of the cabin; The portable controller is used to control the flight of the UAV platform; The offline intelligent detection and analysis system is used to generate a theoretical point cloud model of the cabin and a measured point cloud model based on the point cloud data collected by the online data acquisition and processing system. It also calculates multi-level differences after multi-scale registration of the theoretical point cloud model and the measured point cloud model, and outputs and displays the difference results of different categories.

8. The intelligent detection system for cabin assembly based on drone according to claim 7 is characterized in that: The offline intelligent detection and analysis system comprises: The model reconstruction and generation module is used to generate a theoretical point cloud model with component ID according to the design file of the cabin, and to reconstruct the collected point cloud data of the cabin inner surface into a measured point cloud model; Model multi-scale registration module, used to realize the registration and alignment of theoretical point cloud model and measured point cloud model; The model multi-level difference calculation module is used to perform difference calculation on the aligned model and output the difference results of different categories; Result visualization and marking module, used to display the difference results in the visualization interface and mark the corresponding component ID; The data storage and query module is used to provide data storage and query functions.

9. The intelligent detection system for cabin assembly based on drone according to claim 8 is characterized in that: The specific working process of the model multi-scale registration module includes: first, removing the noise and outliers in the point cloud of the model; then downsampling the point cloud and converting the coordinate system; then normalizing the scale of the point cloud data of the model; and then using feature matching or global optimization methods to perform coarse registration and alignment; finally, based on the coarse registration results, an iterative optimization algorithm is used to achieve precise registration and alignment of the theoretical point cloud model and the measured point cloud model.

10. The intelligent detection system for cabin assembly based on drone according to claim 8 or 9, characterized in that: The specific working process of the model multi-level difference calculation module includes: firstly, calculating the global error between the theoretical point cloud model after multi-scale registration and the measured point cloud model; then, decomposing the model into several independent components based on the component ID, and calculating the geometric difference of each component in the two models respectively; then, constructing a topological relationship network for the ID information of each component, and calculating the geometric relationship difference with the adjacent components; then, for the local area with large error in the component-level geometric difference, performing fine difference calculation of local alignment based on the component ID; finally, classifying and quantifying each difference, and outputting the quantified difference result.

Images