A rapid three-dimensional reconstruction method of high-temperature smelting furnace lining

Abstract

This invention relates to the field of industrial inspection and 3D reconstruction technology, and discloses a rapid 3D reconstruction method for high-temperature furnace linings, comprising: acquiring original point cloud data; performing point cloud quality assessment and determining filtering parameters based on the assessment results to complete filtering and noise reduction processing; performing point cloud registration and stitching processing on the filtered point cloud data to obtain target point cloud data; performing 3D reconstruction processing to generate a 3D model of the high-temperature furnace lining; dividing the furnace lining area based on the spatial coordinate information of the target point cloud data and calculating the point cloud coverage rate; identifying target areas with point cloud coverage rates lower than a preset threshold as supplementary scanning areas; performing supplementary scanning on the supplementary scanning areas and updating the target point cloud data; and generating a target high-temperature furnace lining 3D model based on the updated target point cloud data. This achieves rapid and high-precision 3D reconstruction under high-temperature conditions, improves point cloud quality and model integrity, and enhances inspection efficiency and the reliability of condition analysis.

Description

Technical Field

[0001] This invention relates to the field of industrial inspection and three-dimensional reconstruction technology, specifically to a rapid three-dimensional reconstruction method for the lining of a high-temperature furnace. Background Technology

[0002] The inspection of the linings of high-temperature furnaces such as electric arc furnaces, converters, and ladles in the steel smelting industry has long relied on manual observation, which has problems such as low detection accuracy and significant safety hazards. Traditional thickness measurement methods cannot achieve intuitive three-dimensional modeling and trend analysis.

[0003] The following technical problems often exist in the existing 3D modeling process of furnaces: First, when collecting point clouds in a high-temperature furnace environment, the equipment undergoes structural deformation due to temperature, resulting in spatial coordinate deviations in the acquired point cloud data, which affects the accuracy of 3D modeling. Second, point cloud data is usually filtered using fixed parameters, which is difficult to adapt to the differences in point cloud density and distribution in different regions, resulting in unstable filtering effect and affecting the quality of point cloud data. Third, during the stitching process and the scanning process of local areas, multi-view point cloud data may have insufficient data alignment and inadequate local coverage, resulting in discontinuous or incomplete 3D models. Summary of the Invention

[0004] The summary section of this invention provides a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] This invention proposes a rapid three-dimensional reconstruction method for the lining of a high-temperature furnace to solve one or more of the technical problems mentioned in the background section above.

[0006] This invention provides a rapid three-dimensional reconstruction method for the inner lining of a high-temperature furnace, comprising: acquiring the original point cloud data of the inner lining of the high-temperature furnace; evaluating the point cloud quality of the original point cloud data; adaptively determining the filtering parameters based on the evaluation results; and performing filtering and noise reduction processing on the original point cloud data. The filtered point cloud data is registered and stitched together to obtain target point cloud data in a unified coordinate system; the target point cloud data is then reconstructed to generate a three-dimensional model of the high-temperature furnace lining. The furnace lining area is divided based on the spatial coordinate information of the target point cloud data to obtain multiple target areas, and the point cloud coverage rate of each target area is calculated separately. Target areas with point cloud coverage below a preset threshold are identified as supplementary scanning areas; supplementary scanning is performed on these areas and the target point cloud data is updated. Based on the updated target point cloud data, a three-dimensional model of the target high-temperature furnace lining is generated.

[0007] Optionally, perform point cloud quality assessment on the raw point cloud data, including: The number of point clouds in the preset spatial neighborhood of the original point cloud data is counted, and the number of point clouds per unit volume is used as the point cloud density. Based on the distance distribution between each point cloud point and its corresponding spatial neighborhood point, the mean and variance of the distance are calculated, and the variance is used to characterize the degree of dispersion of the point cloud. Anomalies are identified based on whether the distance between each point cloud point and the center point of its corresponding spatial neighborhood exceeds a preset distance threshold, and the proportion of anomalies to the total number of points in the point cloud is used as the noise point proportion. The evaluation results of point cloud quality are determined based on point cloud density, point cloud discreteness, and the proportion of noise points.

[0008] Optionally, the filtering parameters can be adaptively determined based on the evaluation results, including: The size parameters of the filter kernel are determined based on the point cloud density; the type of filter algorithm is selected based on the degree of point cloud dispersion; and the filter threshold parameters are determined based on the proportion of noise points.

[0009] Optionally, the furnace lining area is divided based on the spatial coordinate information of the target point cloud data to obtain multiple target areas, and the point cloud coverage rate of each target area is calculated, including: Obtain the reference geometric model of the high-temperature furnace, and spatially register the target point cloud data with the reference geometric model; Based on predefined region labels in the reference geometric model, the target point cloud data is mapped to multiple predefined regions; the predefined regions are then identified as target regions, resulting in multiple target regions. The point cloud coverage rate of each target area is calculated based on the preset number of point clouds in the reference geometric model and the actual number of point clouds in each target area.

[0010] Optionally, supplementary scanning of the supplementary scanning area and updating of the target point cloud data include: Obtain the spatial coordinate range of the supplementary scan area; generate multiple candidate scan poses based on the spatial coordinate range; Select the pose with the lowest overlap with the already scanned pose from multiple candidate scan poses as the supplementary scan pose, and collect point cloud data based on the supplementary scan pose.

[0011] Optionally, the pose with the lowest overlap with the already scanned pose can be selected as the supplementary scan pose, including: Calculate the viewpoint difference between each candidate scan pose and the historical scan pose; The candidate scan pose with the largest difference in viewpoint is selected as the supplementary scan pose.

[0012] Optionally, acquire raw point cloud data of the high-temperature furnace lining, including: Acquire temperature data during the scanning process; determine the structural offset of the point cloud acquisition device based on the temperature data; and perform coordinate correction on the original point cloud data according to the structural offset.

[0013] Optionally, the rapid three-dimensional reconstruction method for the lining of a high-temperature furnace according to the present invention further includes: Register the three-dimensional model of the target high-temperature furnace lining with the historical model; calculate the deformation or thickness change of each target region; and generate the state distribution result of the furnace lining based on the deformation or thickness change.

[0014] This invention offers the following advantages: By introducing temperature data to compensate for the structural offset of the point cloud acquisition device, the influence of high-temperature environments on measurement results is effectively eliminated, thereby improving the accuracy of point cloud spatial coordinates. Simultaneously, by comprehensively evaluating point cloud density, point cloud discreteness, and the proportion of noise points, and adaptively adjusting the filter kernel size, filter algorithm type, and filter threshold parameters, the filtering process can be dynamically optimized for different point cloud distribution characteristics, thus improving point cloud data quality. Through multiple iterations of point matching and position adjustment, high-precision alignment of multiple sets of point cloud data in a unified coordinate system is achieved, and the continuity and consistency of the stitched point cloud are improved through point fusion and homogenization processing. Furthermore, by constructing a triangular mesh and performing smoothing, hole filling, and abnormal facet removal, the generated 3D model exhibits continuity. The complete surface structure can accurately reflect the spatial morphology and surface characteristics of the high-temperature furnace lining. Furthermore, by calculating the point cloud coverage of each region and identifying insufficiently covered areas, combined with a candidate scan pose and viewpoint difference selection mechanism, targeted supplementary scanning is achieved, reducing redundant scanning, improving data acquisition efficiency, and enhancing model integrity. By registering the current 3D model with historical models, the deformation or thickness changes of each region are calculated, and the change information is mapped one-to-one with spatial location to form a state distribution result, thereby providing an intuitive expression of the lining erosion degree and change trend. Therefore, this invention can achieve rapid and automated 3D reconstruction and state analysis in high-temperature environments, featuring high detection accuracy and efficiency, and can provide reliable data support for furnace lining maintenance decisions, life prediction, and safety assessment. Attached Figure Description

[0015] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0016] Figure 1 This is a flowchart of a rapid three-dimensional reconstruction method for a high-temperature furnace lining according to the present invention; Figure 2 This is a flowchart of the point cloud quality assessment method of the present invention; Figure 3 This is another flowchart of a rapid three-dimensional reconstruction method for a high-temperature furnace lining according to the present invention. Detailed Implementation

[0017] The invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0018] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0020] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0021] The names of messages or information exchanged between the various devices of this invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] like Figure 1 The diagram shows a flowchart of a rapid three-dimensional reconstruction method for a high-temperature furnace lining according to the present invention, which specifically includes the following steps: Step 101: Obtain the original point cloud data of the high-temperature furnace lining; evaluate the point cloud quality of the original point cloud data, and adaptively determine the filtering parameters based on the evaluation results to perform filtering and noise reduction processing on the original point cloud data. The acquisition of raw point cloud data for the lining of a high-temperature furnace includes: Acquire temperature data during the scanning process; determine the structural offset of the point cloud acquisition device based on the temperature data; and perform coordinate correction on the original point cloud data according to the structural offset.

[0024] In some embodiments, the main execution entity of the rapid three-dimensional reconstruction method for a high-temperature furnace lining according to the present invention is a backend server. Based on this, the backend server establishes a data connection with a point cloud acquisition device through a communication interface, receiving the raw point cloud data generated by the point cloud acquisition device during the scanning process. The point cloud acquisition device is a hardware device used to generate raw point cloud data, including at least a three-dimensional lidar (responsible for distance measurement), an angle encoder (recording the horizontal and vertical scanning angles), a time counting module (recording the laser emission and reception times), and a temperature sensor (collecting the ambient temperature). The point cloud acquisition device scans the high-temperature furnace lining using a three-dimensional lidar. During the scanning process, the three-dimensional lidar emits laser pulses towards the surface of the furnace lining, while the internal timing module records the start time of the laser emission. When the laser irradiates the lining surface, it is reflected, and the reflected signal is received by the receiving unit, at which point the reception time is recorded. The point cloud acquisition device obtains the round-trip propagation time of the laser in space by subtracting the emission time from the reception time. Since the propagation speed of laser in air is a known constant, the device converts this propagation time into the propagation speed and converts the round-trip propagation path into a one-way distance, thereby obtaining the distance information between the scanning point and the sensor. The lining of a high-temperature furnace refers to the refractory material structural layer installed inside the furnace and in direct contact with the high-temperature molten material. During the scanning process, the point cloud acquisition device not only acquires the distance information of each point but also records the corresponding scanning direction information, including the horizontal and vertical scanning angles. Specifically, a sensor coordinate system is pre-established within the point cloud acquisition device, defining the sensor's forward direction, horizontal rotation direction, and vertical deflection direction. For each scanned point, the horizontal deflection position is determined based on its corresponding horizontal scanning angle. The vertical deflection position is determined based on the vertical scanning angle. The directions corresponding to these two angles are combined with the distance information, and the distance is projected along that direction. The components of the point in the forward / backward, left / right, and up / down directions are calculated respectively, thus obtaining the point's position coordinates in three-dimensional space, i.e., spatial coordinate data. This spatial coordinate data is sent as raw point cloud data to the backend server. The backend server receives the raw point cloud data through a communication interface and stores it as a point cloud dataset. Filtering parameters are a set of parameters used to control the neighborhood range, processing method, and anomaly detection conditions in the point cloud filtering process. They include filter kernel size parameters, filtering algorithm type, and filtering threshold parameters.

[0025] In some embodiments, the backend server simultaneously establishes a communication connection with the temperature sensor via a communication interface to receive temperature data during the scanning process at preset time intervals. The preset time interval is determined based on the temperature change rate and allowable temperature error; specifically, it is the maximum sampling time interval within the allowable error range calculated based on the temperature change rate. First, the backend server receives temperature data uploaded by the temperature sensor via the communication interface and reads the current temperature value as the current ambient temperature. Simultaneously, during the initial system calibration phase, a calibration temperature is pre-recorded as a reference temperature when the device has no structural deformation; this calibration temperature serves as the base temperature for subsequent calculations. Subsequently, the backend server determines the temperature change based on the difference between the current temperature and the calibration temperature. Further, the backend server reads pre-stored device structural parameters, including the initial structural dimensions of the point cloud acquisition device in three spatial directions and the coefficient of thermal expansion corresponding to the device material. These structural parameters are pre-written into the backend server during device manufacturing or system calibration. After obtaining the temperature change and structural parameters, the backend server calculates the changes in structural dimensions in each spatial direction; that is, in each direction, it determines the length change in that direction based on the temperature change and thermal expansion characteristics. Specifically, the thermal expansion characteristic of a material is characterized by the proportional elongation or contraction of its length as temperature changes. This proportional relationship is represented by the material's coefficient of thermal expansion. The backend server acquires the initial structural dimensions for each spatial direction and, combining this with the temperature change and the material's coefficient of thermal expansion, determines the length change as follows: the greater the temperature change, the larger the initial structural dimensions, or the larger the coefficient of thermal expansion, the greater the length change in that direction; conversely, the smaller the length change. In practice, the backend server multiplies the temperature change, the initial structural dimensions for that direction, and the coefficient of thermal expansion to obtain the length change in that direction. The coefficient of thermal expansion is a pre-calibrated constant parameter. The initial structural dimensions for different directions are obtained through device calibration. The backend server directly calls this parameter during calculation and combines it with the current temperature change to obtain the length change in each direction. Since the installation posture of the point cloud acquisition device is fixed, the structural length changes in each direction can be converted one-to-one into coordinate offsets in the corresponding directions of the point cloud coordinate system, thus obtaining the structural offsets in the three spatial directions. After obtaining the structural offset, the backend server performs coordinate correction processing on the point cloud data. Specifically, for each point in the point cloud data, its original coordinate values ​​in each coordinate direction are corrected, that is, the structural offset is subtracted in the corresponding direction to obtain the corrected point cloud coordinate data, so as to eliminate the influence of equipment structural deformation caused by high temperature environment on the point cloud measurement results.

[0026] The point cloud quality assessment of the raw point cloud data includes: Step 201: Count the number of points in the original point cloud data within the preset spatial neighborhood, and use the number of points in the point cloud per unit volume as the point cloud density. Step 202: Based on the distance distribution between each point cloud point and its corresponding spatial neighborhood point, calculate the mean and variance of the distance, and use the variance to characterize the degree of dispersion of the point cloud. Step 203: Determine abnormal points based on whether the distance between each point cloud point and the corresponding spatial neighborhood center point exceeds a preset distance threshold, and use the ratio of the number of abnormal points to the total number of points in the point cloud as the noise point ratio. Step 204: Determine the evaluation results of point cloud quality based on point cloud density, point cloud discreteness, and noise point ratio.

[0027] In some embodiments, such as Figure 2 The diagram illustrates a flowchart of the point cloud quality assessment process according to the present invention. Specifically, after obtaining the corrected point cloud data, the backend server iterates through the point cloud data, calculates the spatial distance between each point and other points, and selects points with a distance less than or equal to a preset radius as neighboring points, thereby constructing a preset spatial neighborhood. The preset spatial neighborhood is a spatial region centered on the point and with a preset radius as its range. Based on this, a point is selected in the point cloud data as the center point. A fixed spatial range (e.g., a spherical or cubic range) is defined centered on this center point. The number of point cloud points contained within this range is counted. The volume of this spatial range is calculated. The number of points is divided by the volume to obtain the number of points per unit volume. This value is used as the point cloud density of the region. Point cloud density refers to the number of point cloud points contained per unit volume within the preset spatial neighborhood, used to characterize whether the point cloud distribution is dense. Simultaneously, the distance from each point in the neighborhood to the center point is calculated, the mean distance is obtained, and the variance of the distance is further calculated to reflect the concentration of the point cloud distribution. Point cloud dispersion refers to the degree of spatial dispersion of point cloud points, specifically the difference in distance from each point in the neighborhood to the center point. Further, if the distance between a point cloud point and the neighborhood center point exceeds a preset distance threshold, that point is considered an outlier. The preset distance threshold is calculated based on the mean and standard deviation of the neighborhood distances. The proportion of outliers to the total number of points in the point cloud is recorded as the noise point ratio. The noise point ratio refers to the proportion of outliers in the point cloud. In one implementation, the above indicators can be normalized and weighted according to preset weights to obtain a point cloud quality score, and the point cloud quality can be divided into different levels based on the score. The point cloud quality assessment result refers to the judgment made on whether the current point cloud data is suitable for 3D reconstruction based on point cloud density, point cloud dispersion, and the noise point ratio.

[0028] Among them, the filtering parameters are adaptively determined based on the evaluation results, including: The size parameters of the filter kernel are determined based on the point cloud density; the type of filter algorithm is selected based on the degree of point cloud dispersion; and the filter threshold parameters are determined based on the proportion of noise points.

[0029] In some embodiments, after obtaining the point cloud quality assessment results, the point cloud density of the entire point cloud or a local region is statistically analyzed. One or more density thresholds are set (e.g., high density threshold, low density threshold). The current point cloud density is compared with the threshold: if the point cloud density is high, it indicates that the point cloud is relatively dense, and a smaller filter kernel size is selected; if the point cloud density is low, it indicates that the point cloud is relatively sparse, and a larger filter kernel size is selected; the selected filter kernel size parameter is used for subsequent filtering operations. A dispersion threshold is set, and the current dispersion is compared with the threshold: if the dispersion is low, it indicates that the point cloud is relatively smooth, and mean filtering can be used for smoothing; if the dispersion is high, it indicates that there are many outliers, and statistical outlier filtering is used to remove outliers; the filtering algorithm type is determined and executed. A noise ratio threshold is set, and the threshold strength is selected according to the ratio: if the noise ratio is high, a stricter filtering threshold is set (easier to remove points); if the noise ratio is low, a looser filtering threshold is set (more points are retained); this threshold parameter is used in the filtering judgment process. After determining the filtering parameters, filtering and noise reduction processing is performed on the point cloud data. Specifically, taking statistical outlier filtering as an example, the average distance to the points in the neighborhood of each point in the point cloud is calculated, and the global average distance and standard deviation are statistically analyzed. When the average distance of a point is greater than the sum of the global average distance and a preset multiple of the standard deviation, the point is identified as an outlier and removed, thus obtaining the filtered point cloud data. Through the above processing, high-quality point cloud data after coordinate correction and filtering and noise reduction is obtained, providing a reliable data foundation for subsequent point cloud registration, stitching, and 3D reconstruction. By statistically analyzing the distribution characteristics of points in a preset spatial neighborhood, the point cloud density, point cloud dispersion, and proportion of noise points are obtained. Based on the comparison results of these three factors with preset thresholds, the point cloud quality is graded and judged, and then the filter kernel size parameters, filter algorithm type, and filter threshold parameters are determined according to preset rules.

[0030] Step 102: Perform point cloud registration and stitching on the filtered point cloud data to obtain target point cloud data in a unified coordinate system; perform three-dimensional reconstruction on the target point cloud data to generate a three-dimensional model of the high-temperature furnace lining. In some embodiments, during the actual scanning process, the point cloud acquisition device is typically mounted on a robotic arm and scans at different positions along a preset path, thus obtaining multiple sets of point cloud data, each corresponding to an independent local coordinate system. The backend server receives and stores multiple sets of filtered point cloud data: the first set, the second set, ..., the nth set. The backend server first acquires the scanning pose information corresponding to each set of point cloud data. This pose information includes the spatial position and orientation of the point cloud acquisition device. The pose information can be provided by the robotic arm control system or recorded by sensors and uploaded to the backend server along with the point cloud data. The backend server sets one set of point cloud data as the reference point cloud and the others as the point clouds to be registered. For each set of point clouds to be registered, the backend server performs the following operations: First, based on the scanning pose information, it performs initial position adjustment on the point clouds to be registered, transforming the point cloud from its original local coordinate system to a spatial position range that roughly overlaps with the reference point cloud, so that the two sets of point clouds have an overlapping area in space. After initial position adjustment, the backend server performs fine alignment processing on the reference point cloud and the point cloud to be registered. Specifically, several spatially evenly distributed points are selected from the reference point cloud as reference points, and points in the point cloud to be registered that are spatially close to the reference points are selected as corresponding points. Based on the positional differences between the reference points and the corresponding points, the backend server calculates the required translation and rotation angles for the entire point cloud to be registered, and applies these translation and rotation angles to adjust its spatial position. After each position adjustment, the backend server recalculates the corresponding points between the two sets of point clouds and recalculates the positional differences, repeating the process of matching points, calculating offsets, and adjusting positions until the overall positional deviation between the two sets of point clouds is less than a preset error range, thus completing the registration of the two sets of point clouds. The backend server repeats the above process for all point cloud data to align all point cloud data to the same coordinate system.

[0031] In some embodiments, after completing the registration of all point cloud data, the backend server merges all registered point cloud data into a single point cloud set. During the merging process, for point cloud points with similar spatial locations, the backend server calculates their spatial distance. If the distance is less than a preset merging distance, multiple points are merged into one point, or only one point is retained, to avoid duplicate data. Simultaneously, the backend server performs a continuity check on the stitched point cloud data. For sparse areas of the local point cloud, the original point distribution is retained; for overly dense areas, appropriate homogenization processing is performed to make the overall point cloud distribution more uniform. Through the above processing, complete point cloud data in a unified coordinate system is obtained, i.e., the target point cloud data. The target point cloud data refers to the set of point cloud data that, after registration and stitching, is in the same coordinate system and can completely represent the spatial morphology of the high-temperature furnace lining. Based on this, the backend server calculates the spatial distribution of each point in the target point cloud data within its neighborhood and determines the orientation information of the surface where the point is located based on the spatial positional relationship of the points within the neighborhood, thus providing a foundation for subsequent surface construction. Subsequently, the backend server selects spatially adjacent points in the point cloud data and connects them according to proximity, gradually constructing a mesh structure composed of multiple triangles. During the construction process, every three spatially adjacent and non-collinear points are connected to form a triangular facet. Multiple triangular facets are pieced together to gradually form a mesh model covering the entire point cloud surface. After the initial mesh model is constructed, the backend server optimizes the mesh. Specifically, this includes: smoothing the mesh surface to reduce local abrupt changes; filling areas with voids by connecting surrounding points to generate new facets; and removing abnormal facets to eliminate erroneous structures caused by residual noise points. After completing the above processing, a continuous and complete three-dimensional mesh model is obtained. The backend server outputs the optimized three-dimensional mesh model as the three-dimensional model of the high-temperature furnace lining. The three-dimensional model can intuitively reflect the spatial shape and surface morphology of the furnace lining and can be used for subsequent deformation analysis, thickness assessment, and life prediction.

[0032] Step 103: Divide the furnace lining area based on the spatial coordinate information of the target point cloud data to obtain multiple target areas, and calculate the point cloud coverage rate of each target area in the multiple target areas respectively; Specifically, the furnace lining area is divided based on the spatial coordinate information of the target point cloud data to obtain multiple target areas, and the point cloud coverage rate of each target area is calculated, including: Step 301: Obtain the reference geometric model of the high-temperature furnace and spatially register the target point cloud data with the reference geometric model; Step 302: Based on the predefined region labels in the reference geometric model, map the target point cloud data to multiple predefined regions; determine the predefined regions as target regions to obtain multiple target regions; Step 303: Calculate the point cloud coverage rate of each target area based on the preset number of point clouds in the reference geometric model and the actual number of point clouds.

[0033] In some embodiments, such as Figure 3The diagram illustrates another flowchart of a rapid 3D reconstruction method for a high-temperature furnace lining according to the present invention. Specifically, the backend server reads a reference geometric model of the high-temperature furnace from a pre-stored model library. This reference geometric model is established during system deployment using the furnace design dimensions and maintains a consistent spatial proportion with the actual furnace. After reading the reference geometric model, the backend server loads the target point cloud data into the same computational space and aligns the target point cloud data with the reference geometric model. Specifically, the backend server selects the overall center position of the target point cloud data as the point cloud center point and simultaneously calculates the geometric center position of the reference geometric model. The point cloud center point is translated to the center position of the reference geometric model, making them coincide in space. After translation, the backend server adjusts the overall orientation of the point cloud, ensuring that the vertical direction of the point cloud data aligns with the vertical direction of the reference geometric model, and simultaneously aligns the principal axis of the point cloud with the principal axis of the model, thus placing the target point cloud data and the reference geometric model in a unified coordinate system. After alignment, the backend server reads the region labels from the reference geometric model. The reference geometric model is constructed by dividing different locations into regions and assigning a unique identifier to each region. The backend server traverses each point in the target point cloud data and performs the following processing for each point: first, it reads the spatial coordinates of the point; then, it locates the spatial position of the coordinates in the reference geometric model; it determines which region within the reference geometric model the point falls into; and it assigns the point to the corresponding region set. After traversal, the backend server divides the target point cloud data into multiple point cloud subsets, each subset corresponding to a region label, thus completing the region division of the furnace lining. Each point cloud subset corresponding to a region label is considered as a target region, resulting in multiple target regions. The backend server performs surface sampling on each region in the reference geometric model to obtain the standard number of points in that region. Specifically, the backend server divides the surface of the region into grids at fixed intervals, places virtual sampling points at the grid intersections, and counts the total number of sampling points, using this number as the preset point cloud quantity for that region. During system initialization, the preset point cloud quantity is calculated and stored in the backend server and is directly read during use. For each target area, the backend server counts the number of point cloud points falling within that area. During the counting process, the backend server handles points that are spatially duplicated or too close together. Specifically, it calculates the spatial distance between points; when the distance between two points is less than a preset distance, only one point is retained for counting, thus avoiding the influence of duplicate points on the statistical results. After processing, the number of point cloud points retained in each area is counted as the actual point cloud count for that area. For each area, the preset point cloud count for that area is read, and the actual point cloud count for that area is also read. Dividing the actual point cloud count by the preset point cloud count yields the point cloud coverage rate for that area.The backend server repeats the above calculation process for all target areas to obtain the point cloud coverage rates for each area. Through the above processing, the backend server divides the target point cloud data into multiple target areas; obtains the point cloud point set corresponding to each target area; and calculates the point cloud coverage rate for each target area. The point cloud coverage rate reflects the completeness of the point cloud data within the area. A higher coverage rate indicates that the area has been scanned more thoroughly, while a lower coverage rate indicates that the area has unscanned or insufficiently scanned areas. The predefined area refers to multiple spatially defined regional units pre-divided on the surface of the high-temperature furnace lining during the reference geometric model construction stage, based on the structural characteristics and operating conditions of the lining. The target area refers to the point cloud data set formed after mapping the points in the target point cloud data to the predefined area. Spatial coordinate information refers to the positional data of each point in the point cloud data in three-dimensional space, calculated from the distance information collected by the point cloud acquisition device and the corresponding horizontal and vertical scanning angles, specifically represented by the three-dimensional coordinate values ​​of each point.

[0034] Step 104: Identify target areas with point cloud coverage rates below a preset threshold as supplementary scanning areas; perform supplementary scanning on the supplementary scanning areas and update the target point cloud data; This includes performing supplementary scanning of the supplementary scanning area and updating the target point cloud data, including: Obtain the spatial coordinate range of the supplementary scan area; generate multiple candidate scan poses based on the spatial coordinate range; Select the pose with the lowest overlap with the already scanned pose from multiple candidate scan poses as the supplementary scan pose, and collect point cloud data based on the supplementary scan pose.

[0035] Among them, the pose with the lowest overlap with the already scanned pose is selected as the supplementary scan pose, including: Calculate the viewpoint difference between each candidate scan pose and the historical scan pose; The candidate scan pose with the largest difference in viewpoint is selected as the supplementary scan pose.

[0036] In some embodiments, the backend server reads the point cloud coverage rate corresponding to each target region and obtains a pre-set coverage threshold. The coverage threshold is a boundary value used to determine whether the region scanning is sufficient, and is pre-set and stored in the backend server during the system initialization phase. The backend server compares the point cloud coverage rate of each target region with the coverage threshold. When the point cloud coverage rate of a target region is less than the coverage threshold, the target region is marked as a supplementary scanning region, and all marked regions are aggregated to form a supplementary scanning region set. After determining the supplementary scanning regions, the backend server obtains the spatial coordinate range of each supplementary scanning region. Specifically, the backend server reads the three-dimensional coordinates of all point cloud points in the region and calculates the minimum and maximum coordinate values ​​of all points in the X, Y, and Z directions, respectively, to obtain the boundary range of the region in the three directions. The boundary range is used as the spatial coordinate range of the supplementary scanning region. After obtaining the spatial coordinate range, the backend server sets up multiple observation positions around the spatial range outside it. The observation positions are distributed in different directions of the supplementary scanning region, including above, side, and diagonal positions. For each observation position, the backend server calculates the direction of the line connecting that observation position and the center of the supplementary scanning area, and uses this direction as the scanning direction, pointing the scanning direction towards the interior of the supplementary scanning area. The backend server combines the observation position and the scanning direction to form multiple candidate scanning poses, each containing spatial position parameters and scanning direction parameters. After generating candidate scanning poses, the backend server reads the historical scanning pose set. The historical scanning poses are the collection of scanning pose data executed during the preceding scan process, recorded and stored in the backend server after each scan. Subsequently, the backend server calculates the viewpoint difference between each candidate scanning pose and the historical scanning poses. Specifically, the backend server reads the scanning direction of the candidate scanning pose and sequentially reads the scanning directions of each scanning pose in the historical scanning pose set, calculates the angle between the two directions, and uses this angle as the viewpoint difference value. For each candidate scanning pose, the viewpoint difference values ​​between it and all historical scanning poses are statistically processed to obtain the comprehensive viewpoint difference degree of the candidate scanning pose. The preset threshold is a coverage limit value used to determine whether a region is sufficiently scanned. It is set to a fixed value, such as 80%, by the backend server during the system initialization phase and stored in the parameter configuration.

[0037] In some embodiments, the backend server compares the degree of viewpoint difference corresponding to all candidate scanning poses and selects the candidate scanning pose with the largest viewpoint difference as the supplementary scanning pose. This method maximizes the difference between the newly selected scanning direction and the historical scanning direction, thereby reducing redundant scanning and improving scanning efficiency for uncovered areas. After determining the supplementary scanning pose, the backend server sends control commands to the point cloud acquisition device, causing the device to move to the supplementary scanning pose and perform point cloud data acquisition according to the set scanning direction. After completing the scan, the point cloud acquisition device sends the newly acquired point cloud data to the backend server via the communication interface. Upon receiving the new point cloud data, the backend server performs the same coordinate correction and filtering noise reduction processing as in step 101, and converts the processed point cloud data to a unified coordinate system. Subsequently, the backend server merges the new point cloud data with the original target point cloud data, performs fusion processing on points with similar spatial positions, and removes duplicate points to obtain updated target point cloud data. Through the above processing, supplementary scanning and point cloud data updates are achieved for areas with insufficient point cloud coverage, improving the integrity of the overall point cloud data and the reconstruction accuracy of the 3D model. Viewpoint difference refers to the degree of deflection between two scanning poses in the scanning direction, reflecting the magnitude of the difference between the two scanning viewpoints. Scanning pose refers to the spatial state of the point cloud acquisition device during scanning, including position parameters (the device's coordinates in space) and orientation parameters (the device's scanning direction). Candidate scanning poses are multiple alternative scanning states set around the supplementary scanning area, each candidate scanning pose including spatial position parameters and scanning orientation parameters. The supplementary scanning pose is the scanning state selected from the multiple candidate scanning poses for performing the supplementary scanning operation.

[0038] Step 105: Based on the updated target point cloud data, generate a three-dimensional model of the target high-temperature furnace lining.

[0039] In some embodiments, the backend server first reads the updated target point cloud data and loads it into the 3D modeling processing module. Each point in the point cloud data contains 3D spatial coordinate information. The backend server traverses the point cloud data, reading the spatial coordinates of each point point one by one. Subsequently, the backend server constructs a neighborhood range around each point point. For the current point point, the backend server calculates the spatial distance between that point and other points, filters out points whose distance is less than a preset neighborhood radius, and uses the set of filtered points as the neighborhood point set of that point. After obtaining the neighborhood point set, the backend server analyzes the spatial distribution of points within the neighborhood range. Specifically, it reads the spatial coordinates of the neighborhood points, analyzes the distribution direction of each point in space, and uses the main extension direction of the neighborhood points as the orientation information of the surface where the point is located. After completing the surface orientation calculation for each point, the backend server constructs a triangular mesh structure in the point cloud data. In the specific implementation process, the backend server selects a point in the point cloud data as the starting point, and selects two spatially adjacent points that are not on the same straight line within its neighborhood. These three points are connected to form a triangular patch, which is then added to the mesh set. Subsequently, the backend server repeats the above operation for other points in the point cloud data, gradually forming multiple triangular patches between adjacent points. These multiple triangular patches are then connected to form a mesh structure covering the surface of the point cloud data. After completing the initial mesh construction, the backend server optimizes the mesh structure. First, the backend server reads the neighboring vertices of each vertex in the mesh and calculates the average position of the neighboring vertices. The current vertex is then moved towards this average position to reduce local surface abrupt changes. Second, the backend server searches for areas in the mesh not covered by triangles and establishes connections between the boundary points of these areas, filling the empty areas by constructing new triangular patches. Finally, the backend server detects the triangular patches in the mesh, deleting those with an area smaller than a preset threshold or abnormal shapes. After mesh optimization, the backend server combines all triangular facets to form a complete 3D mesh structure, which is then stored or output as a 3D model of the high-temperature furnace lining. Through this process, discrete point cloud data is transformed into a continuous 3D surface structure, resulting in a 3D model that reflects the spatial shape and surface features of the high-temperature furnace lining. The target high-temperature furnace lining 3D model refers to the 3D geometric model data that fully describes the spatial shape and surface structure of the high-temperature furnace lining, formed based on updated target point cloud data through mesh construction and optimization.

[0040] The rapid three-dimensional reconstruction method for high-temperature furnace lining of the present invention further includes: Register the three-dimensional model of the target high-temperature furnace lining with the historical model; calculate the deformation or thickness change of each target region; and generate the state distribution result of the furnace lining based on the deformation or thickness change.

[0041] In some embodiments, the backend server first reads historical models from a historical model dataset. Historical models are 3D model data obtained after scanning and modeling the lining of the same high-temperature furnace at historical time points. After each 3D modeling operation, the backend server stores the corresponding 3D model along with time information, thus forming a historical model dataset. After reading the current 3D model and the historical model, the backend server performs spatial alignment. The backend server reads the spatial coordinates of all mesh vertices in both the current and historical models, calculates their geometric center positions, and translates the historical model as a whole so that its center position coincides with that of the current 3D model. Subsequently, the backend server adjusts the orientation of the historical model to align its vertical direction with that of the current 3D model, while also aligning the model's principal axis. After initial alignment, the backend server selects vertices with similar spatial positions in the current and historical models as corresponding points, calculates the spatial position difference between these points, and performs overall position adjustment on the historical model. By repeatedly performing corresponding point matching and position adjustment operations, the overall positional deviation between the two models is gradually reduced until it falls below a preset error range, thus completing model registration. After model registration is completed, the backend server maps the target region to the current 3D model and the historical model. Specifically, the backend server reads the spatial range corresponding to each target region, and filters the mesh vertices falling within this spatial range in the current 3D model, and simultaneously filters the mesh vertices falling within the same spatial range in the historical model, thus obtaining the vertex set of that region in the current model and the historical model, respectively. Subsequently, the backend server calculates the deformation of each target region. Specifically, a vertex is selected from the region vertex set in the current model, and the vertex with the closest spatial position in the corresponding region of the historical model is found. The spatial distance between the two is calculated, and this distance is used as the deformation of that vertex. The above operation is repeated for all vertices in the region, and the deformation of all vertices is statistically processed. The statistical result is used as the overall deformation of the target region. At the same time, the backend server calculates the thickness variation of each target region. The backend server first reads the design thickness data of the corresponding region in the reference geometric model. In the current 3D model, the distance from the inner lining surface to the reference plane is measured along the surface normal direction, which is used as the current thickness; the same measurement is performed in the historical model to obtain the historical thickness. The backend server calculates the difference between the current thickness and the historical thickness, using this difference as the thickness change. After obtaining the deformation or thickness change of each target region, the backend server performs state distribution display processing on the 3D model. Specifically, the backend server reads the numerical value corresponding to each target region and converts this value into color information, where regions with larger values ​​correspond to higher brightness or higher contrast colors, and regions with smaller values ​​correspond to lower brightness or lower contrast colors.Subsequently, the backend server assigns color information to the corresponding surface areas in the 3D model and displays or outputs the entire 3D model in a color-coded format. Through this process, the spatial changes of the high-temperature furnace lining under different time conditions can be compared and analyzed, and the deformation degree or thickness change of each region can be presented in a visual manner, thus obtaining the state distribution results of the furnace lining and providing a basis for subsequent maintenance and life assessment. The state distribution results refer to the spatial distribution data reflecting the overall state of the high-temperature furnace lining, formed by corresponding the deformation or thickness change of each target region to its position in 3D space and organizing it in a structured data format.

[0042] In these embodiments, by introducing temperature data to compensate for the structural offset of the point cloud acquisition device, the influence of high-temperature environment on measurement results is effectively eliminated, thereby improving the accuracy of point cloud spatial coordinates. Simultaneously, by comprehensively evaluating point cloud density, point cloud discrepancy, and the proportion of noise points, and adaptively adjusting the filter kernel size, filter algorithm type, and filter threshold parameters, the filtering process can be dynamically optimized for different point cloud distribution characteristics, thereby improving point cloud data quality. Through multiple iterations of point matching and position adjustment, high-precision alignment of multiple sets of point cloud data in a unified coordinate system is achieved, and the continuity and consistency of the stitched point cloud are improved through point fusion and homogenization processing. Furthermore, by constructing a triangular mesh and performing smoothing, hole filling, and abnormal facet removal processing, the generated 3D model has continuity and completeness. The complete surface structure can accurately reflect the spatial morphology and surface characteristics of the high-temperature furnace lining. Furthermore, by calculating the point cloud coverage of each region and identifying insufficiently covered areas, combined with a candidate scan pose and viewpoint difference selection mechanism, targeted supplementary scans are achieved, reducing redundant scans, improving data acquisition efficiency, and enhancing model integrity. By registering the current 3D model with historical models, the deformation or thickness changes of each region are calculated, and the change information is mapped one-to-one with spatial location to form a state distribution result, thereby providing an intuitive expression of the lining erosion degree and change trend. Therefore, this invention can achieve rapid and automated 3D reconstruction and state analysis in high-temperature environments, featuring high detection accuracy and efficiency, and can provide reliable data support for furnace lining maintenance decisions, life prediction, and safety assessment.

[0043] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to specific combinations of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for rapid three-dimensional reconstruction of a high-temperature furnace lining, characterized in that, include: Obtain raw point cloud data of the lining of a high-temperature furnace; evaluate the point cloud quality of the raw point cloud data, and adaptively determine the filtering parameters based on the evaluation results to perform filtering and noise reduction processing on the raw point cloud data; The filtered point cloud data is registered and stitched together to obtain target point cloud data in a unified coordinate system. The target point cloud data is subjected to three-dimensional reconstruction processing to generate a three-dimensional model of the high-temperature furnace lining; Based on the spatial coordinate information of the target point cloud data, the furnace lining area is divided into multiple target areas, and the point cloud coverage rate of each target area is calculated. Target areas with point cloud coverage rates below a preset threshold among the multiple target areas are identified as supplementary scanning areas; supplementary scanning is performed on the supplementary scanning areas and the target point cloud data is updated; Based on the updated target point cloud data, a three-dimensional model of the target high-temperature furnace lining is generated.

2. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 1, characterized in that, The point cloud quality assessment of the original point cloud data includes: The number of point clouds in the preset spatial neighborhood of the original point cloud data is counted, and the number of point clouds per unit volume is used as the point cloud density. Based on the distance distribution between each point cloud point and its corresponding spatial neighborhood point, the mean and variance of the distance are calculated, and the variance is used to characterize the degree of dispersion of the point cloud. Anomalies are identified based on whether the distance between each point cloud point and the center point of its corresponding spatial neighborhood exceeds a preset distance threshold, and the proportion of anomalies to the total number of points in the point cloud is used as the noise point proportion. The evaluation results of point cloud quality are determined based on the point cloud density, point cloud dispersion, and noise point ratio.

3. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 2, characterized in that, The adaptive determination of filtering parameters based on the evaluation results includes: The size parameters of the filter kernel are determined based on the point cloud density; the filter algorithm type is selected based on the point cloud discreteness; and the filter threshold parameters are determined based on the noise point ratio.

4. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 1, characterized in that, The furnace lining area is divided based on the spatial coordinate information of the target point cloud data to obtain multiple target areas, and the point cloud coverage rate of each target area is calculated, including: A reference geometric model of a high-temperature furnace is obtained, and the target point cloud data is spatially registered with the reference geometric model. Based on the predefined region labels in the reference geometric model, the target point cloud data is mapped to multiple predefined regions; the predefined regions are then determined as target regions, resulting in multiple target regions. The point cloud coverage rate of each target area is calculated based on the preset number of point clouds in the reference geometric model and the actual number of point clouds in each target area.

5. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 1, characterized in that, The step of performing supplementary scanning on the supplementary scanning area and updating the target point cloud data includes: Obtain the spatial coordinate range of the supplementary scanning area; generate multiple candidate scanning poses based on the spatial coordinate range; Select the pose with the lowest overlap with the already scanned pose from the multiple candidate scan poses as the supplementary scan pose, and collect point cloud data based on the supplementary scan pose.

6. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 5, characterized in that, The selection of the pose with the lowest overlap with the already scanned pose as the supplementary scan pose includes: Calculate the viewpoint difference between each candidate scan pose and the historical scan pose; The candidate scan pose with the largest difference in viewpoint is selected as the supplementary scan pose.

7. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 1, characterized in that, The acquisition of the raw point cloud data of the high-temperature furnace lining includes: Acquire temperature data during the scanning process; determine the structural offset of the point cloud acquisition device based on the temperature data; and perform coordinate correction on the original point cloud data according to the structural offset.

8. The rapid three-dimensional reconstruction method for a high-temperature furnace lining according to claim 1, characterized in that, Also includes: The three-dimensional model of the target high-temperature furnace lining is registered with the historical model; Calculate the deformation or thickness change of each target region; The state distribution result of the furnace lining is generated based on the deformation or thickness change.

Images