Steel ladle lining detection method and system

Abstract

The invention relates to the field of intelligent detection of the smelting technology, in particular to a steel ladle lining detection method which comprises the steps that an initial steel ladle point cloud model of a steel ladle lining and a current steel ladle point cloud model in the circulation process are obtained; segmenting in the current steel ladle point cloud model to obtain a target steel ladle point cloud model matched with the initial steel ladle point cloud model; based on the target steel ladle point cloud model and the initial steel ladle point cloud model, a first transformation matrix is calculated through an iterative nearest point algorithm, and a second transformation matrix is calculated by calculating probability distribution in the target steel ladle point cloud model; obtaining a target matrix according to the first change matrix and the second change matrix, and obtaining a reconstructed steel ladle point cloud model according to the target matrix and the target steel ladle point cloud model; comparing the reconstructed steel ladle point cloud model with the initial steel ladle point cloud model according to the consistency of the circle center axis so as to output steel ladle lining detection data; wherein the steel ladle lining detection data at least comprise lining residual thickness.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection of smelting technology, and in particular to a ladle lining detection method and system. Background Art

[0002] As refractory bricks are used under repeated high temperatures and chemical erosion, their thickness gradually decreases. This change is crucial to the overall strength and safety of the ladle, because once the thickness of the refractory bricks drops below the safety critical value, serious accidents of molten steel penetrating the ladle may occur. Such accidents will not only lead to sudden interruptions of the production line, but may also cause immeasurable damage to on-site personnel and equipment, thus causing major safety hazards.

[0003] Steel mills face multiple challenges in the process of detecting the residual thickness of refractory bricks. On the one hand, the thickness of refractory bricks usually ranges from 180 mm to 230 mm, which requires that the detection means must have extremely high precision to ensure that the service life of refractory bricks is maximized while preventing puncture accidents, thereby controlling production costs. On the other hand, the continuous high temperature environment inside the ladle greatly limits the selection and application of detection methods. Traditional detection methods, such as manual observation, although easy to operate, are highly dependent on the operator's experience and subjective judgment, and it is difficult to ensure the accuracy and consistency of the test results.

[0004] What is more complicated is that the ladle needs to be frequently circulated during the steel production process to meet different production needs. This circulation process makes it extremely difficult to measure the residual thickness; firstly, the posture changes during the circulation process affect the difficulty of its coordinate positioning, and secondly, due to the time urgency during the circulation process and the continuity requirements of the production line, it is difficult to perform detailed refractory brick residual thickness detection on the ladle without affecting the production progress.

[0005] Based on this, this application is filed. Summary of the invention

[0006] In response to the above-mentioned technical problems, the purpose of this application is to provide a ladle lining detection method and system, aiming to solve the problem of difficulty in detecting the residual thickness of refractory bricks due to posture changes and time urgency during the frequent circulation of the ladle; this method and system can accurately and quickly detect the residual thickness of the ladle lining without affecting the production progress, thereby ensuring the safe use of the ladle.

[0007] In a first aspect of the present application, a ladle lining detection method is provided, which comprises: obtaining an initial ladle point cloud model of the ladle lining, and a current ladle point cloud model in a circulation process; segmenting the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model; based on the target ladle point cloud model and the initial ladle point cloud model, calculating a first transformation matrix by an iterative nearest point algorithm, and calculating a second change matrix by calculating the probability distribution in the target ladle point cloud model; obtaining a target matrix according to the first change matrix and the second change matrix, and obtaining a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model; comparing the reconstructed ladle point cloud model with the initial ladle point cloud model according to the consistency of the center axis to output ladle lining detection data; wherein the ladle lining detection data includes at least the residual thickness of the lining.

[0008] In a further scheme of the present application, a target ladle point cloud model that matches the initial ladle point cloud model is segmented in the current three-dimensional point cloud model, including: preprocessing the current ladle point cloud model, the preprocessing including denoising and / or filtering; and classification using the geometric features and / or size features of the ladle to segment the target ladle point cloud model.

[0009] In a further scheme of the present application, based on the target ladle point cloud model and the initial ladle point cloud model, a first transformation matrix is ​​calculated by an iterative nearest point algorithm, including: traversing and acquiring the first feature point of the target ladle point cloud model, acquiring the second feature point corresponding to the first feature point cloud in the initial ladle point cloud model, the first feature point and the second feature point forming a point cloud pair; after each iteration of acquiring the point cloud pair, calculating the first temporary transformation matrix that minimizes the Euclidean distance between all first feature points and the corresponding second feature points; when the Euclidean distance and the first temporary change matrix meet the preset first iteration stop condition, outputting the current first temporary change matrix as the first change matrix.

[0010] In a further scheme of the present application, when a preset first iteration stop condition is met, the current first temporary change matrix is ​​output as the first change matrix, including: when the Euclidean distance difference of all corresponding point cloud pairs is less than or equal to the preset first distance threshold, the current first temporary transformation matrix is ​​output as the first transformation matrix; when the change coefficient of the first temporary transformation matrix is ​​less than the preset first tolerance threshold, the current first temporary transformation matrix is ​​output as the first transformation matrix; if the current number of iterations for iteratively acquiring point cloud pairs reaches the preset first iteration threshold, the iteration is stopped and an error is reported.

[0011] In a further scheme of the present application, based on the target ladle point cloud model and the initial ladle point cloud model, a second change matrix is ​​calculated by calculating the probability distribution in the target ladle point cloud model, including: dividing the target ladle point cloud model into multiple grid bodies, and calculating the probability distribution in each grid body; continuously iteratively adjusting the second temporary transformation matrix with the first transformation matrix as the initial value, so as to minimize the matching error between the initial ladle point cloud model and the grid body after being transformed by the second temporary transformation matrix; when the second temporary transformation matrix meets the preset second iteration stop condition, the second change matrix is ​​output.

[0012] In a further scheme of the present application, when the second temporary transformation matrix and the second current number of iterations meet corresponding conditions, a second change matrix is ​​output, including: when the change coefficient of the second temporary transformation matrix is ​​less than a preset second tolerance threshold, outputting the current second temporary transformation matrix as the second change matrix; when the current number of iterations for adjusting the second temporary transformation matrix reaches a preset second iteration threshold, outputting the final second temporary transformation matrix as the second transformation matrix.

[0013] In a further scheme of the present application, a target matrix is ​​obtained according to the first change matrix and the second change matrix, and a reconstructed ladle point cloud model is obtained according to the target matrix and the target ladle point cloud model, including: multiplying the first change matrix and the second change matrix to obtain the target matrix; transforming the target ladle point cloud model through the target matrix to obtain the reconstructed ladle point cloud model.

[0014] In a further scheme of the present application, the second temporary transformation matrix is ​​continuously iteratively adjusted with the first transformation matrix as the initial value, so that the matching error between the initial ladle point cloud model and the grid body after being transformed by the second temporary transformation matrix is ​​minimized, including: using the first transformation matrix as the initial value of the second temporary transformation matrix, projecting the initial ladle point cloud model onto the grid body obtained by dividing the target ladle point cloud model; for each projection point, finding the grid where it is located, and calculating the offset of the position of the point in the grid relative to the grid mean; according to the normal distribution parameters of the projection points in the grid, calculating the probability density of the projection points in the grid, and calculating the likelihood function value from the probability density value. Use an optimization algorithm to adjust the parameters of the second temporary transformation matrix to minimize the likelihood function value, so that the matching error between the initial ladle point cloud model and the grid body after being transformed by the second temporary transformation matrix is ​​minimized.

[0015] In a further scheme of the present application, the reconstructed ladle point cloud model and the initial ladle point cloud model are compared according to the consistency of the center axis to output the ladle lining detection data, including: extracting a number of ladle lining points on the reconstructed ladle point cloud model, and calculating the first radius from the ladle lining point to the center axis; obtaining the second radius from the reference point in the same orientation as the ladle lining point on the initial ladle point cloud model to the center axis; obtaining the corrosion value according to the difference between the first radius and the second radius, and calculating the residual thickness of the lining through the original thickness and the corrosion value.

[0016] In a further solution of the present application, the automatic monitoring method also includes: when the residual thickness of the lining is lower than a preset thickness threshold, stopping the ladle circulation and issuing an early warning, and visualizing the corresponding area where the residual thickness of the lining is lower than the preset thickness threshold.

[0017] The second aspect of the present application also provides a ladle lining detection system, including: a data acquisition module, used to obtain an initial ladle point cloud model of the ladle lining and a current ladle point cloud model in a circulation process; a point cloud matching module, used to segment the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model; a matrix calculation module, based on the target ladle point cloud model and the initial ladle point cloud model, calculates a first transformation matrix through an iterative nearest point algorithm, and calculates the probability distribution in the target ladle point cloud model to obtain a second transformation matrix; a reconstruction module, which obtains a target matrix according to the first transformation matrix and the second transformation matrix, and generates a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model; a comparison and analysis module, which is used to compare the reconstructed ladle point cloud model with the initial ladle point cloud model to output ladle lining detection data, wherein the ladle lining detection data at least includes the residual thickness of the lining; an early warning and visualization module, when it is detected that the residual thickness of the lining is lower than a preset thickness threshold, stops the ladle circulation and issues an early warning signal, and visualizes the corresponding area where the residual thickness of the lining is lower than the preset thickness threshold.

[0018] In summary, the offline calibration device and method provided by the embodiments of the present invention include at least the following technical effects:

[0019] The present application provides a ladle lining detection method, which uses a high-precision point cloud model to capture slight changes in the ladle surface, and can accurately reflect the deformation and wear of the ladle through iterative calculation of feature point distance and probability distribution. In particular, for ladles in circulation, it can eliminate errors caused by shape and position changes and improve the detection accuracy. The second change matrix is ​​calculated considering the probability distribution information, which can further improve the accuracy of local alignment, and is particularly suitable for ladles with complex surfaces and detail features. Finally, based on the unique characteristics of the ladle, its center axis consistency is used as a reference element, and its consistency ensures the accuracy of the detection results. Based on the comparison of the center axis consistency, the wear of the ladle lining is accurately evaluated. The entire detection process is based on digital calculation to achieve automation and reduce manual errors. By comparing the initial and reconstructed point cloud models, intelligent analysis of the deformation and wear of the ladle lining is achieved.

[0020] Other features and advantages of the embodiments of the present invention will be described in the following specific implementation examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flow chart of a ladle lining detection method provided in an embodiment of the present application;

[0023] Figure 2 A flow chart of step 200 of the ladle lining detection method provided in an embodiment of the present application;

[0024] Figure 3 An example of a target ladle point cloud model in the ladle lining detection method provided in the embodiment of the present application;

[0025] Figure 4 This is a flow chart of step S300 in the ladle lining detection method provided in an embodiment of the present application;

[0026] Figure 5 Flow chart of step S303 in the ladle lining detection method provided in the embodiment of the present application

[0027] Figure 6 Another flow chart of step S300 in the ladle lining detection method provided in an embodiment of the present application;

[0028] Figure 7This is a flow chart of step S302' in the ladle lining detection method provided in an embodiment of the present application;

[0029] Figure 8 This is a flow chart of step S303' in the ladle lining detection method provided in an embodiment of the present application;

[0030] Fig. 9 Flow chart of step S400 in the ladle lining detection method provided in the embodiment of the present application

[0031] Fig.10 and Fig.11 Schematic diagrams of the target ladle point cloud model (green), the initial ladle point cloud model (yellow), and the reconstructed ladle point cloud model (red) provided in the embodiments of the present application at two viewing angles respectively;

[0032] Fig.12 This is a flow chart of step S500 in the ladle lining detection method provided in an embodiment of the present application;

[0033] Fig.13 This is a system topology diagram of the ladle lining detection system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The terms "second direction", "first direction", "third direction", "inside", "outside" and the like that appear below to indicate orientation or positional relationships, unless otherwise specified, are to be understood as being based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0035] In addition, if there is a feature defined as "first" or "second", it is only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Features defined as "first" or "second" may explicitly or implicitly include at least one of the defined features. If the description of "plurality" appears, the general meaning is to include at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0036] In this application, unless otherwise clearly specified and limited, the terms such as "installed", "connected", "connected", "fixed" and so on should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integrated connection; it can be a mechanical connection or an electrical connection, it can be a direct connection, or it can be an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances.

[0037] In the description of this specification, if the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" appear, it means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0038] Reference Figure 1 The present application embodiment first provides a ladle lining detection method, comprising at least the following steps:

[0039] Step S100, obtaining an initial ladle point cloud model of the ladle lining and a current ladle point cloud model during circulation;

[0040] Step S200, segmenting the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model;

[0041] Step S300, iteratively calculating a first transformation matrix according to the distances of corresponding feature points of the target ladle point cloud model and the initial ladle point cloud model, and iteratively calculating a second change matrix by calculating the probability distribution in the target ladle point cloud model;

[0042] Step S400, obtaining a target matrix according to the first change matrix and the second change matrix, and obtaining a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model;

[0043] Step S500: Compare the reconstructed ladle point cloud model with the initial ladle point cloud model according to the consistency of the circle center axis to output ladle lining detection data; wherein the ladle lining detection data at least includes the lining residual thickness.

[0044] In the present application, the "initial ladle point cloud model" is the three-dimensional point cloud data of the ladle lining in its initial state, that is, before use, and the point cloud model is composed of a number of three-dimensional coordinate points; the "current ladle point cloud model" is the three-dimensional point cloud data of the ladle during circulation. Due to corrosion during use, the lining area of ​​the model undergoes morphological changes compared with the initial model.

[0045] In the current ladle point cloud model, a portion matching the initial ladle point cloud model is segmented based on shape matching or feature extraction, and this matching portion is defined as a target ladle point cloud model in this application.

[0046] The above-mentioned first transformation matrix is ​​calculated by the feature point distance algorithm. Specifically, the optimal alignment method is iteratively calculated according to the distance between the point clouds through the feature points of the target ladle point cloud model and the feature points of the initial ladle point cloud model; the first transformation matrix describes the rigid body transformation from the initial ladle point cloud model to the target ladle point cloud model.

[0047] It can be understood that the feature points should have obvious geometric or physical features, such as corner points, edge points or key points on the surface; these feature points should be easy to identify and locate in the point cloud model, and at the same time, the feature points should remain relatively stable during the circulation of the ladle, and should not be easily affected by external factors and undergo significant changes, which helps to ensure the accuracy and reliability of the alignment results; the present application achieves high-precision point cloud alignment by iteratively finding the nearest point pairs between two point cloud models (the initial ladle point cloud model and the target ladle point cloud model) and calculating the transformation that minimizes the distance between these point pairs; it enables accurate comparison of the differences between the initial ladle point cloud model and the current ladle point cloud model, thereby accurately detecting the wear of the lining.

[0048] Furthermore, by calculating the probability distribution of points in the target ladle point cloud model, the probability distribution can be robust to noise and small disturbances in the point cloud data, and can process point cloud data with different shapes, sizes and densities. It is particularly suitable for complex scenes such as ladle in circulation, avoiding the change of point cloud data due to the movement, deformation or occlusion of the object. The transformed model is closer to the real situation in shape and features; it helps to further refine the transformation matrix, so as to more accurately evaluate the corrosion degree of the lining.

[0049] It is understandable that the shape and position of the ladle may change during its circulation due to various factors; in order to accurately evaluate the wear of the lining, it is necessary to accurately align the current ladle point cloud model with the initial model; by calculating the transformation matrix, the current ladle point cloud model can be transformed into a coordinate system consistent with the initial model, thereby eliminating errors caused by shape and position changes and improving the accuracy of detection; the general concept of this application calculates the first transformation matrix by iterative nearest point algorithm, which can ensure that the target ladle point cloud model is more consistent with the initial model globally. At the same time, considering the probability distribution information to calculate the second change matrix, the accuracy of local alignment can be further improved, especially for ladles with complex surfaces and detailed features.

[0050] A target matrix is ​​calculated based on the first transformation matrix and the second transformation matrix. This target matrix may combine the information of the two transformations to provide a more accurate transformation description. The target ladle point cloud model is transformed using the target matrix to obtain a reconstructed ladle point cloud model; this model should be closer to the initial ladle point cloud model, but reflects the deformation in actual use.

[0051] Based on the concept of this application and in accordance with the principle of consistency of the central axis of the ladle, by comparing the reconstructed ladle point cloud model with the initial ladle point cloud model, the changes in the ladle lining, such as the degree of wear and residual thickness, can be calculated.

[0052] It is understandable that the ladle, as a key equipment in steel production, usually has a clear center axis during its design and manufacturing process, and the center axis is kept consistent even during the circulation process. Therefore, based on the consistency of the center axis, the center axis can be used as a reference benchmark to calculate the change data of the ladle lining.

[0053] In summary, the present application provides a ladle lining detection method, which uses a high-precision point cloud model to capture subtle changes in the ladle surface, and can accurately reflect the deformation and wear of the ladle through iterative calculation of feature point distance and probability distribution. In particular, for ladles in circulation, it can eliminate errors caused by shape and position changes and improve the detection accuracy. The second change matrix is ​​calculated by considering the probability distribution information, which can further improve the accuracy of local alignment, and is particularly suitable for ladles with complex surfaces and detail features. Finally, based on the unique characteristics of the ladle, its center axis is consistent as a reference element, and its consistency ensures the accuracy of the detection results. Based on the comparison of the center axis consistency, the wear of the ladle lining is accurately evaluated. The entire detection process is based on digital calculation, which realizes automation and reduces manual errors. By comparing the initial and reconstructed point cloud models, intelligent analysis of the deformation and wear of the ladle lining is achieved.

[0054] According to an implementation of the present application, obtaining the current ladle point cloud model in the circulation process in step S100 includes:

[0055] Preset workstations are set up at key locations of the steel production line, which are usually located in the necessary routes for the ladle to circulate or in areas that require key monitoring; the preset workstations should ensure that the ladle can stay stably during operation to facilitate point cloud collection; when the ladle runs to the preset workstation, the point cloud collection equipment is triggered by the control system to start working; the point cloud collection equipment may include high-precision equipment such as laser scanners and / or three-dimensional cameras, which can capture the three-dimensional coordinate information of the ladle surface.

[0056] In order to obtain a complete point cloud model of the ladle, it is necessary to control the point cloud acquisition equipment to collect the ladle from multiple directions; this can be achieved by adjusting the position and angle of the acquisition equipment or using multiple acquisition equipment to work simultaneously. After collecting the image point cloud information in multiple directions, image registration is required, that is, integrating these image point cloud information into a unified coordinate system; after completing the image registration, the image point cloud information in different directions needs to be spliced ​​to form a complete current point cloud model of the ladle.

[0057] like Figure 2In step S200, a target ladle point cloud model matching the initial ladle point cloud model is obtained by segmenting the current three-dimensional point cloud model, including:

[0058] Step S201, preprocessing the current ladle point cloud model, the preprocessing including denoising and / or filtering;

[0059] Remove noise points in point cloud data, which may be caused by equipment accuracy, environmental factors or errors in the data stitching and registration process; denoising can be performed using algorithms such as statistical filters, conditional filters, and radius filters. For example, statistical filters achieve denoising by calculating the average distance from each point to its nearest neighbor and removing points outside the variance based on Gaussian distribution parameters.

[0060] In addition to denoising, filtering can also be used for point cloud smoothing and data compression; commonly used filtering algorithms include Gaussian filtering, bilateral filtering, voxel filtering, etc. Gaussian filtering smoothes point clouds by weighted averaging; bilateral filtering considers both spatial proximity and pixel value similarity to achieve edge-preserving denoising; voxel filtering calculates the centroid or random points of point cloud data to approximate the original point cloud, thereby achieving data compression.

[0061] Step S202: Classify the ladle using its geometric features and / or size features to segment a target ladle point cloud model.

[0062] It can be understood that the current point cloud model of the ladle includes some of the environment and redundant features of the ladle. Since the purpose of this application is to detect the lining, the current collected ladle point cloud model is segmented through the specific geometric shape of the ladle, such as the plane of the bottom, the cylindrical shape of the ladle, or the surface of the ladle may have specific textures, concavity or curvature changes, etc.; and important geometric features such as the size of the ladle (such as diameter, height, etc.), so as to only segment the target ladle point cloud model, so as to reduce the amount of calculation and improve the subsequent matching accuracy.

[0063] In a specific embodiment, using the geometric features and / or size features of the ladle for classification to segment the target ladle point cloud model includes:

[0064] Use the random sampling consensus (RANSAC) algorithm or other cylinder fitting algorithms to fit a cylinder in the current ladle point cloud model;

[0065] The RANSAC algorithm iteratively randomly selects sample points from the point cloud and tries to fit a cylindrical model; iteratively randomly selects sample points from the point cloud and tries to fit a cylindrical model; in each iteration, the residual between the model and the point cloud data is calculated, and the model with the smallest residual is selected as the current best fit; the iterative process is repeated until the preset number of iterations is reached or the best fit model that meets the conditions is found.

[0066] The fitted cylinder is verified by size features to determine the target ladle point cloud model and segment it.

[0067] Since the fitting result may include multiple cylinders, the fitted cylinders are evaluated by size parameters, including diameter, height and other parameters; these parameters are compared with the preset ladle size range to extract possible ladle parts. If there are multiple fitting cylinders, the most likely ladle model can be selected based on the size matching degree and spatial position relationship.

[0068] According to the matching degree between the size parameters (such as diameter and height) of the fitted cylinder and cone and the preset ladle size range, the point cloud is divided into the target ladle point cloud and other point clouds, and the target ladle point cloud after preliminary classification is further finely segmented, so that only the target ladle point cloud model of the ladle part remains, such as Figure 3 shown.

[0069] like Figure 4 As shown, in step S300, the first transformation matrix is ​​iteratively calculated according to the distance between the corresponding feature points of the target ladle point cloud model and the initial ladle point cloud model, including:

[0070] Step S301, traverse and obtain the first feature point of the target ladle point cloud model, obtain the second feature point corresponding to the first feature point cloud in the initial ladle point cloud model, and the first feature point and the second feature point constitute a point cloud pair;

[0071] Step S302: when acquiring point cloud pairs in each iteration, calculate a first temporary transformation matrix that minimizes the Euclidean distance between all first feature points and corresponding second feature points;

[0072] Step S303: When the Euclidean distance and the first temporary change matrix satisfy a preset first iteration stop condition, output the current first temporary change matrix as the first change matrix.

[0073] It can be understood that the current goal is to find a transformation matrix so that the target ladle point cloud model can be aligned with the initial ladle point cloud model after transformation; extract a series of calibrated feature points from the target ladle point cloud model, the feature points can be edge points, corner points or other points with obvious geometric features of the ladle and not affected by the flow; in the alignment of the initial ladle point cloud model, try to find points corresponding to the feature points in the target ladle point cloud model; this can be achieved through shape matching, color matching or other feature matching algorithms.

[0074] Furthermore, an initial transformation matrix is ​​preset as a unit matrix to represent the relative position relationship between the target ladle point cloud model and the initial ladle point cloud model before any transformation is performed; at the same time, an upper limit of the number of iterations, i.e., the first iteration threshold, and a Euclidean distance threshold for judging whether the transformation is accurate enough are set;

[0075] In each iteration, a pair of feature points (respectively denoted as point P and point Q) are selected from the first feature point of the initial ladle point cloud model and the second feature point of the target ladle point cloud model; starting from the initial transformation matrix, point P is transformed so that it can be as close to point Q as possible, that is, the "first temporary transformation matrix with the smallest Euclidean distance between all first feature points and corresponding second feature points" mentioned above; this transformation matrix can be solved by the least squares method or other optimization algorithms; thus, the current temporary transformation matrix is ​​applied to the target ladle point cloud model to obtain a new target ladle point cloud model; by continuously iteratively optimizing the transformation matrix, the Euclidean distance between the feature point pairs between the initial ladle point cloud model and the target ladle point cloud model is minimized, and accurate alignment of the two point cloud models can be achieved.

[0076] In one example, the initial transformation matrix: is set to the unit matrix I, indicating that before any transformation is performed, the relative position relationship between the two point cloud models is a unit transformation; the upper limit of the number of iterations is set to 50 times, that is, the first iteration threshold. The Euclidean distance threshold is set to 0.1 to determine whether the transformation is accurate enough. The feature points of the initial ladle point cloud model are: P1, P2, ..., Pn; the second feature points of the target ladle point cloud model are: Q1, Q2, ..., Qn. Select feature point P from the initial ladle point cloud model, and select feature point Q from the target ladle point cloud model. Calculate the Euclidean distance from point P to point Q under the current transformation matrix (initially the unit matrix I), and use the least squares method or other optimization algorithms to solve the first temporary transformation matrix T1 that makes point P as close to point Q as possible after transformation; until the Euclidean distance and the first temporary change matrix meet the preset first iteration stop condition, the current first temporary change matrix is ​​output as the first change matrix.

[0077] like Figure 5As shown, in step S303 of the present application, when the preset first iteration stop condition is met, the current first temporary change matrix is ​​output as the first change matrix, including:

[0078] Step S3031: when the Euclidean distance differences of all corresponding point cloud pairs are less than or equal to a preset first distance threshold, output the current first temporary transformation matrix as the first transformation matrix;

[0079] When the Euclidean distance differences of all corresponding point cloud pairs are less than or equal to the preset first distance threshold, the current first temporary transformation matrix is ​​output as the first transformation matrix. This means that the distance differences between the positions of all matching points after transformation and the actual points are very small, indicating that the transformation matrix is ​​accurate enough.

[0080] Step S3032: when the variation coefficient of the first temporary transformation matrix is ​​less than a preset first tolerance threshold, output the current first temporary transformation matrix as the first transformation matrix;

[0081] When the change coefficient of the first temporary transformation matrix is ​​less than the preset first tolerance threshold, the current first temporary transformation matrix is ​​output as the first transformation matrix; in the present application, the change coefficient may refer to a small change in the elements in the transformation matrix (such as rotation angle, translation vector, etc.), which may be determined according to the actual configuration; if the change is already very small, it means that further iterations have reached the optimal solution, and subsequent iterations are unlikely to significantly improve the results.

[0082] Step S3033: If the current number of iterations for iteratively acquiring point cloud pairs reaches a preset first iteration threshold, the iteration is stopped and an error is reported.

[0083] During the entire iterative process, by continuously updating the transformation matrix and applying it to the target point cloud, the error can be gradually reduced and the alignment accuracy can be improved. At the same time, by setting the upper limit of the number of iterations and the Euclidean distance threshold, the algorithm can be ensured to converge within a reasonable time and avoid falling into a local optimal solution.

[0084] like Figure 6 As shown, in step S300, based on the target ladle point cloud model and the initial ladle point cloud model, the probability distribution in the target ladle point cloud model is calculated to iteratively calculate the second change matrix, including:

[0085] Step S301', dividing the target ladle point cloud model into multiple grid volumes, and calculating the probability distribution within each grid volume;

[0086] The target ladle point cloud model is divided into multiple grid volumes using a space partitioning algorithm such as octree, KD tree, etc. For each grid volume, the probability distribution of its internal points is calculated using a point cloud normal estimation algorithm (such as PCA, covariance matrix, etc.). The probability distribution can be expressed as the mean, variance, or other statistics of the normals of all points in the grid volume.

[0087] Step S302', taking the first transformation matrix as the initial value, iteratively adjusting the second temporary transformation matrix continuously, so as to minimize the matching error between the initial ladle point cloud model after being transformed by the second temporary transformation matrix and the grid body;

[0088] Step S303': when the second temporary conversion matrix and the second current iteration number meet corresponding conditions, output the second change matrix.

[0089] In summary, after each iteration, check whether the stopping condition is met: the number of iterations reaches the upper limit or the matching error is less than the threshold; if the stopping condition is met, the current second temporary transformation matrix is ​​output as the second change matrix; if the stopping condition is not met, continue to iterate and adjust the second temporary transformation matrix.

[0090] like Figure 7 In an embodiment corresponding to step S302', the second temporary transformation matrix is ​​iteratively adjusted continuously with the first transformation matrix as the initial value so as to minimize the matching error between the initial ladle point cloud model after being transformed by the second temporary transformation matrix and the grid body, including:

[0091] Step S3021′, using the first transformation matrix as the initial value of the second temporary transformation matrix, projecting the initial ladle point cloud model onto a grid volume obtained by dividing the target ladle point cloud model;

[0092] It can be understood that each point in the initial ladle point cloud model is transformed (rotated and translated) using the first transformation matrix, and then the transformed points are projected onto the grid volume obtained by dividing the target ladle point cloud model.

[0093] Step S3022', for each projection point, find the grid where it is located, and calculate the offset of the position of the point in the grid relative to the grid mean;

[0094] For each projection point, find the grid it is in and calculate the offset of the point's position in the grid relative to the grid mean (i.e., the average position of all points in the grid). For example, for the projection point pi and the mean μg of the grid it is in, the offset Δpi=pi-μg.

[0095] Step S3023', calculating the probability density of the projection point in the grid according to the normal distribution parameters of the projection point in the grid, and calculating the likelihood function value from the probability density value;

[0096] According to the normal distribution parameters (mean μg and covariance matrix Σg) of the projection points in the grid, the probability density of the projection points in the grid is calculated using the probability density function of the normal distribution. Then the probability density values ​​of all the projection points are multiplied (or added after taking the logarithm in practice) to obtain the likelihood function value.

[0097] For example, the probability density function of the normal distribution is Where x is the projection point, μ is the grid mean, and Σ is the covariance matrix. The likelihood function value is Where N is the number of projection points.

[0098] Step S3024', using an optimization algorithm to adjust the parameters of the second temporary conversion matrix to minimize the likelihood function value, so as to minimize the matching error between the initial ladle point cloud model after being converted by the second temporary conversion matrix and the grid body.

[0099] The parameters of the second temporary transformation matrix are adjusted using an optimization algorithm to minimize the likelihood function value.

[0100] In the present application, the optimization algorithm may be Newton's method, Gauss-Newton method or Levenberg-Marquardt algorithm, etc.

[0101] In summary, by converting the point cloud into a probability density function, the algorithm avoids point-by-point matching, thereby improving the computational efficiency. Moreover, since the normal distribution is used to describe the local properties of the point cloud, the above algorithm has a certain tolerance to noise and outliers in the point cloud, and is particularly suitable for complex scenes such as ladles in circulation.

[0102] like Figure 8 As shown, in step S303', when the second temporary conversion matrix and the current number of iterations meet corresponding conditions, the second change matrix is ​​output, including:

[0103] Step S3031′: when the change coefficient of the second temporary transformation matrix is ​​less than a preset second tolerance threshold, output the current second temporary conversion matrix as a second change matrix;

[0104] As above, the coefficient of change may refer to the amount by which an element in the transformation matrix changes, such as a small change in a rotation angle, translation vector, or scaling factor. If this change is very small, further iterations are unlikely to significantly improve the result, so it can be considered that a good enough transformation matrix has been found.

[0105] Step S3032′: when the second iteration number reaches a preset second iteration threshold, output the final second temporary conversion matrix as the second transformation matrix.

[0106] If the number of iterations reaches the preset upper limit but the condition in step S3031' is not met, the iteration cannot further converge to a more accurate solution. It is a reasonable choice to stop the iteration and output the current best solution (ie, the second temporary transformation matrix).

[0107] It should be noted that since the first change matrix can already achieve a considerable degree of alignment, when the number of subsequent normal vector iterations reaches the preset upper limit, if other conditions for stopping iterations are not met, directly using the last second temporary transformation matrix as the locally optimal second transformation matrix output is a strategy to avoid overfitting.

[0108] It is understandable that the ladle is a container for holding and transporting molten steel in steel production. Its shape is usually complex, and there may be various irregular protrusions, depressions or welds on the surface; these features make the point cloud data of the ladle in the circulation process vulnerable to noise, occlusion or data loss during the collection and processing process. Therefore, when the ladle is reconstructed or aligned in three dimensions, a more accurate and robust algorithm is needed to ensure the accuracy of the result. The reason why the present application provides an iterative matching method using the matching error of the grid body after the point-to-point iterative matching is that the point-to-point matching can achieve alignment, but it may be affected by the irregular features and noise data on the surface of the ladle; and the iterative matching using the probability distribution of the grid body can further process these features and data, and by iteratively adjusting the transformation matrix, the matching error between the initial ladle point cloud model and the grid body after conversion is minimized, and the irregular features and noise data on the surface of the ladle can be better processed, because the division of the grid body can average or smooth these features, thereby reducing the impact on the matching results. Improve the accuracy and stability of alignment. By combining these two methods, accurate alignment and 3D reconstruction of the ladle point cloud data can be achieved.

[0109] like Fig. 9 As shown, further in step S400 of the present application, a target matrix is ​​obtained according to the first change matrix and the second change matrix, and a reconstructed ladle point cloud model is obtained according to the target matrix and the target ladle point cloud model, including:

[0110] Step S401, multiplying the first change matrix and the second change matrix to obtain a target matrix;

[0111] The first change matrix and the second change matrix are subjected to matrix multiplication operation to obtain the target matrix; if the first change matrix is ​​set to A and the second change matrix is ​​set to B, then the target matrix C can be calculated by C=A*B. It should be noted that the multiplication here follows the rule of matrix multiplication, that is, multiplying rows by columns to obtain new element values.

[0112] Step S402: transform the target ladle point cloud model through the target matrix to obtain a reconstructed ladle point cloud model.

[0113] If the point in the target ladle point cloud model is P, then the point P' after the transformation by the target matrix C can be calculated by P'=C*P (here P and P' are usually points represented by homogeneous coordinates, and C is a 4x4 transformation matrix). By transforming each point in the point cloud model, the entire reconstructed ladle point cloud model can be obtained. In a physical sense, the target matrix calculated previously is applied to the actual point cloud data, and a new reconstructed ladle point cloud model is obtained through transformation, which has a certain correspondence (such as position, posture, etc.) with the initial ladle point cloud model; an exemplary example is Fig.10 and Fig.11 The two angle shots shown are as follows: the yellow one is the initial ladle point cloud model to be matched, the green one is the target ladle point cloud model before matching, and the red one is the reconstructed ladle point cloud model.

[0114] like Fig.12 As shown, in step S500, the reconstructed ladle point cloud model and the initial ladle point cloud model are compared according to the consistency of the circle center axis to output the ladle lining detection data, including:

[0115] Step S501, extracting a plurality of ladle lining points on the reconstructed ladle point cloud model, and calculating a first radius from the ladle lining points to the center axis;

[0116] From the reconstructed ladle point cloud model, select several points that can represent the shape of the ladle lining; these points should be evenly distributed on the surface of the ladle lining to ensure that the corrosion of the lining can be fully reflected; for each selected point, calculate its distance to the center axis, which is called the first radius. The center axis is usually the symmetry axis of the ladle and can be determined by the initial ladle point cloud model or design data.

[0117] Step S502, obtaining a second radius from a reference point at the same position as the ladle lining point of the initial ladle point cloud model to the center axis;

[0118] In the initial ladle point cloud model, find reference points corresponding to the points selected in step S501; these reference points should be located at the same position of the initial ladle lining; for each reference point, also calculate its distance to the center axis, which is called the second radius.

[0119] Step S503: obtaining a corrosion value according to the difference between the first radius and the second radius, and calculating the lining residual thickness by using the original thickness and the corrosion value.

[0120] For each point, the difference between the first radius and the second radius is calculated, and this difference represents the degree of corrosion of the ladle lining at the point, that is, the corrosion value. The larger the corrosion value, the more serious the lining corrosion at the point; according to the original design thickness of the ladle lining and the corrosion value calculated in step S503, the residual thickness of the lining at the point is calculated, and the residual thickness of the lining is equal to the original thickness minus the corrosion value.

[0121] In a specific scheme, the reconstructed ladle point cloud model is evenly divided into n layers of two-dimensional point clouds along its depth direction (i.e., the direction perpendicular to the center axis). For each layer of two-dimensional point cloud, starting from the center axis, a number of ladle lining points are selected uniformly or distributed on demand along the radial direction (i.e., from 0 degrees to 360 degrees); the selection of these points should be able to fully cover and represent the shape of the lining layer and the possible corrosion conditions; for each selected point, the shortest distance to the center axis is calculated, and this distance is defined as the first radius. In the initial ladle point cloud model, a reference point with the same orientation (i.e., radial position and angle relative to the center axis) as the point selected in step 2 is found; for each reference point found, the shortest distance to the center axis is also calculated, and this distance is defined as the second radius; this distance reflects the distance from the lining to the center axis in the initial state (i.e., before corrosion occurs); finally, for each point, the corrosion value is calculated by subtracting the second radius from the first radius, indicating the loss or corrosion degree of the lining at this point relative to the initial state.

[0122] The automatic monitoring method also includes: step S600, when the residual thickness of the lining is lower than a preset thickness threshold, stopping the ladle circulation and issuing an early warning, and visualizing the corresponding area where the residual thickness of the lining is lower than the preset thickness threshold.

[0123] After completing all the steps mentioned above (such as comparison of point cloud models, calculation of corrosion values, etc.), the lining residual thickness calculated at each point is compared with the preset thickness threshold. The preset thickness threshold is determined based on industry standards, safety regulations and the original design thickness of the ladle lining to assess whether the lining has reached a level of corrosion that requires immediate attention.

[0124] If the residual thickness of the lining at any point is found to be lower than the preset thickness threshold, the flow of the ladle will be stopped immediately to prevent further corrosion or potential leakage risks. The operation of stopping the flow may involve sending a signal to the control system to automatically stop the relevant production line or transportation equipment. At the same time, the early warning mechanism is triggered to send a notification to relevant personnel (such as maintenance personnel, management personnel, etc.) to inform them that there is a safety hazard in the ladle lining.

[0125] At the same time, the corresponding areas where the residual thickness of the lining is lower than the preset thickness threshold are visualized in the point cloud model or 2D / 3D graphic interface; this is achieved by highlighting, color coding or other visual markings, so that relevant personnel can intuitively understand the location and scope of the corrosion area.

[0126] Furthermore, those skilled in the art should understand that if the ladle lining detection system 100 provided in the embodiment of the present application; all or part of the sub-modules involved in each product are combined and replaced through embodiments such as fusion, simple change, mutual conversion, etc., such as the placement and movement of each component; or a detachable design; or the product constituted by it is set as one, such as an integrated design; all the combined components can form equipment / devices / systems with specific functions, and using such equipment / devices / systems to replace the corresponding components of the present application also falls within the protection scope of the present application.

[0127] like Fig.13 As shown, in a second aspect, the present application also provides an automatic monitoring system 100 for a ladle, comprising:

[0128] A data acquisition module 10 is used to acquire an initial ladle point cloud model of the ladle lining and a current ladle point cloud model during circulation;

[0129] A point cloud matching module 20 is used to segment the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model;

[0130] The matrix calculation module 30 iteratively calculates a first transformation matrix according to the distance between corresponding feature points of the target ladle point cloud model and the initial ladle point cloud model, and calculates the probability distribution in the target ladle point cloud model to obtain a second transformation matrix;

[0131] A reconstruction module 40 obtains a target matrix according to the first transformation matrix and the second transformation matrix, and generates a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model;

[0132] A comparison and analysis module 50, for comparing the reconstructed ladle point cloud model with the initial ladle point cloud model to output ladle lining detection data, wherein the ladle lining detection data at least includes the lining residual thickness;

[0133] The early warning and visualization module 60 stops the ladle circulation and issues an early warning signal when it detects that the residual thickness of the lining is lower than a preset thickness threshold, and visualizes the corresponding area where the residual thickness of the lining is lower than the preset thickness threshold.

[0134] The various technical features described above can be combined arbitrarily. Although all possible combinations of these technical features are not described, any combination of these technical features should be considered to be covered by this specification as long as there is no contradiction in such combination.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still adjust the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these adjustments or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting ladle lining, characterized in that: include: Obtain the initial ladle point cloud model of the ladle lining and the current ladle point cloud model during the circulation process; Segmenting the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model; Based on the target ladle point cloud model and the initial ladle point cloud model, a first transformation matrix is ​​calculated by an iterative closest point algorithm, and a second change matrix is ​​calculated by calculating a probability distribution in the target ladle point cloud model; Obtaining a target matrix according to the first change matrix and the second change matrix, and obtaining a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model; The reconstructed ladle point cloud model is compared with the initial ladle point cloud model according to the consistency of the circle center axis to output the ladle lining detection data; wherein the ladle lining detection data at least includes the lining residual thickness.

2. The automatic monitoring method according to claim 1, characterized in that: The target ladle point cloud model that matches the initial ladle point cloud model is obtained by segmenting the current three-dimensional point cloud model, including: Preprocessing the current ladle point cloud model, wherein the preprocessing includes denoising and / or filtering; The geometric features and / or size features of the ladle are used for classification to segment the target ladle point cloud model.

3. The automatic monitoring method according to claim 1, characterized in that: The first transformation matrix is ​​calculated based on the target ladle point cloud model and the initial ladle point cloud model by iterative closest point algorithm, including: Traversing and acquiring a first feature point of a target ladle point cloud model, acquiring a second feature point corresponding to the first feature point cloud in the initial ladle point cloud model, wherein the first feature point and the second feature point constitute a point cloud pair; After obtaining a point cloud pair in each iteration, calculating a first temporary transformation matrix that minimizes the Euclidean distance between all the first feature points and the corresponding second feature points; When the Euclidean distance and the first temporary change matrix satisfy a preset first iteration stop condition, the current first temporary change matrix is ​​output as the first change matrix.

4. The automatic monitoring method according to claim 1, characterized in that: When the preset first iteration stop condition is met, outputting the current first temporary change matrix as the first change matrix includes: When the Euclidean distance differences of all corresponding point cloud pairs are less than or equal to a preset first distance threshold, outputting the current first temporary transformation matrix as the first transformation matrix; When the change coefficient of the first temporary transformation matrix is ​​less than a preset first tolerance threshold, outputting the current first temporary transformation matrix as the first transformation matrix; If the current number of iterations for iteratively acquiring point cloud pairs reaches the preset first iteration threshold, the iteration is stopped and an error is reported.

5. The automatic monitoring method according to claim 4, characterized in that: The method of calculating a second change matrix by calculating the probability distribution in the target ladle point cloud model based on the target ladle point cloud model and the initial ladle point cloud model comprises: Dividing the target ladle point cloud model into a plurality of grid volumes, and calculating the probability distribution within each grid volume; The second temporary conversion matrix is ​​continuously and iteratively adjusted with the first conversion matrix as the initial value, so as to minimize the matching error between the initial ladle point cloud model after being converted by the second temporary conversion matrix and the grid body; When the second temporary conversion matrix meets a preset second iteration stop condition, a second change matrix is ​​output.

6. The automatic monitoring method according to claim 5, characterized in that: When the second temporary conversion matrix and the second current iteration number meet corresponding conditions, outputting a second change matrix comprises: When the change coefficient of the second temporary transformation matrix is ​​less than a preset second tolerance threshold, outputting the current second temporary transformation matrix as a second change matrix; When the current iteration number of adjusting the second temporary conversion matrix reaches a preset second iteration threshold, the final second temporary conversion matrix is ​​output as the second transformation matrix.

7. The automatic monitoring method according to claim 5, characterized in that: The method of continuously and iteratively adjusting the second temporary conversion matrix with the first conversion matrix as the initial value so as to minimize the matching error between the initial ladle point cloud model after being converted by the second temporary conversion matrix and the grid body comprises: Using the first transformation matrix as the initial value of the second temporary transformation matrix, the initial ladle point cloud model is projected onto a mesh volume obtained by dividing the target ladle point cloud model; For each projection point, find the grid where it is located, and calculate the offset of the point's position in the grid relative to the grid mean; According to the normal distribution parameters of the projection points in the grid, the probability density of the projection points in the grid is calculated, and the likelihood function value is calculated from the probability density value. The parameters of the second temporary transformation matrix are adjusted using an optimization algorithm to minimize the likelihood function value, so that the matching error between the initial ladle point cloud model after transformation by the second temporary transformation matrix and the grid body is minimized.

8. The automatic monitoring method according to claim 5, characterized in that: The step of obtaining a target matrix according to the first change matrix and the second change matrix, and obtaining a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model comprises: Multiplying the first change matrix and the second change matrix to obtain a target matrix; The target ladle point cloud model is transformed through the target matrix to obtain a reconstructed ladle point cloud model.

9. The automatic monitoring method according to any one of claims 1 to 8, characterized in that: The reconstructed ladle point cloud model is compared with the initial ladle point cloud model according to the consistency of the circle center axis to output the ladle lining detection data, including: Extracting a plurality of ladle lining points on the reconstructed ladle point cloud model, and calculating a first radius from the ladle lining points to the center axis; Obtain a second radius from a reference point at the same position as the ladle lining point of the initial ladle point cloud model to the central axis of the circle; The corrosion value is obtained according to the difference between the first radius and the second radius, and the residual thickness of the lining is calculated by the original thickness and the corrosion value.

10. A ladle lining detection system, characterized in that: include: A data acquisition module is used to obtain an initial ladle point cloud model of the ladle lining and a current ladle point cloud model during the circulation process; A point cloud matching module, used for segmenting the current ladle point cloud model to obtain a target ladle point cloud model that matches the initial ladle point cloud model; A matrix calculation module, based on the target ladle point cloud model and the initial ladle point cloud model, calculates a first transformation matrix by an iterative closest point algorithm, and calculates a probability distribution in the target ladle point cloud model to obtain a second transformation matrix; A reconstruction module, which obtains a target matrix according to the first transformation matrix and the second transformation matrix, and generates a reconstructed ladle point cloud model according to the target matrix and the target ladle point cloud model; A comparison and analysis module, used for comparing the reconstructed ladle point cloud model with the initial ladle point cloud model to output ladle lining detection data, wherein the ladle lining detection data at least includes the lining residual thickness; The early warning and visualization module stops the ladle circulation and issues an early warning signal when it detects that the residual thickness of the lining is lower than a preset thickness threshold, and visualizes the corresponding area where the residual thickness of the lining is lower than the preset thickness threshold.

Images