A slag removal control method and slag removal device for medium-sized metal ingots

Abstract

The present invention discloses a slag removal control method and device for medium-sized metal ingots, belonging to the field of slag removal equipment for metal ingots. The slag removal control method comprises: obtaining a target photo, preprocessing and extracting texture features of the surface image information of the ingot mold and the molten metal within the ingot mold; judging whether the surface quality of the molten metal in the target photo meets the quality standards based on the texture features of the molten metal surface image; if not, identifying the slag removal area and matching the corresponding slag removal depth based on a slag removal prediction model constructed using a DenseNet model, and simultaneously correcting the slag removal position of the ingot mold; then, configuring multiple slag removal trajectory points based on the determined slag removal area, slag removal depth, and the corrected slag removal position information of the ingot mold, numbering them in sequence and connecting them in series to form a complete slag removal working trajectory, and having a slag removal robot perform slag removal. The slag removal method of the present invention has high slag removal precision and accuracy, and a small slag removal amplitude, which can ensure the quality of the finished metal ingot.

Description

Technical Field

[0001] The invention belongs to the field of slag scooping equipment for metal ingots, and in particular relates to a slag scooping control method and a slag scooping device for medium-sized metal ingots. Background Art

[0002] Non-ferrous metal ingots are semi-finished products made by melting and pouring non-ferrous metals such as copper, aluminum, and zinc, and their alloys, into blocks or ingots of specific shapes. They are widely used in a variety of fields, including aerospace, automotive manufacturing, electronics, construction, and machinery manufacturing. During the ingot processing of non-ferrous metals into molten metals, mechanical or automated equipment is required to separate and remove surface oxide slag impurities from the molten metal through a slag skimming process to ensure the internal quality and surface finish of the ingot.

[0003] The traditional slag removal process relies on manual operation. Workers are required to hold iron hooks or iron clamps in close contact with the high-temperature molten metal, visually judge the thickness of the slag layer, and scrape the liquid surface horizontally to remove the slag. However, this process has low slag removal efficiency and there are safety hazards such as splashing of molten metal and high-temperature burns to the human body. In addition, uneven slag layer thickness or improper timing during slag removal can cause material waste and surface defects after the ingot is formed.

[0004] Currently, intelligent slag removal devices utilize image noise recognition technology to identify the molten metal in the ingot mold and enable slag removal robots to perform the removal. This replaces traditional manual slag removal, eliminates safety threats to operators, and improves slag removal efficiency. However, since non-ferrous metal ingots can be categorized by weight and size as small, medium, large, and extra-large ingots, conventional image noise recognition is not fully compatible with all types of metal ingots. For large and ultra-large ingots, conventional image noise recognition is more applicable because the noise in the target photos obtained is obvious. However, for medium-sized ingots weighing tens to hundreds of kilograms, the noise in the collected images is not obvious, which can easily lead to misjudgment. This makes the technology less applicable and affects the accuracy of slag removal judgment and slag removal position. In addition, when current slag removal robots remove slag from the cooling molten metal in the ingot mold, the large amplitude of the slag removal movement causes the molten metal to fluctuate. As a result, after the slag is removed and the molten metal is cooled to form a metal ingot, the surface of the metal ingot will be left with ripples, affecting the quality of the finished product. Therefore, there is an urgent need to provide a slag removal control method and slag removal device that can solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to address the above-mentioned shortcomings by providing a slag removal control method and slag removal device for medium-sized metal ingots. This method aims to address the problems that current image noise recognition technology is not highly adaptable to the slag removal process of medium-sized ingots, the slag removal precision and accuracy are low, and the large slag removal amplitude can cause ripples on the surface of the finished product. To achieve the above-mentioned objectives, the present invention provides the following technical solutions:

[0006] A method for controlling slag removal of a medium-sized metal ingot comprises the following steps:

[0007] S1: configuring a visual judgment device, which includes a camera, a Gabor filter bank, and a detector; using the camera to obtain a target photo carrying image information of the ingot mold and the surface of the molten metal in the ingot mold and pre-processing it, and then proceeding to step S2;

[0008] S2: Using a Gabor filter bank to extract texture features from the preprocessed molten metal surface image; based on the texture features of the molten metal surface image, determine whether the surface quality of the molten metal in the target photo meets the standards; if not, trigger steps S3 and S4 simultaneously; if it meets the standards, exit the steps;

[0009] S3: Build and train a slag removal prediction model based on the DenseNet model; input the target photo into the slag removal prediction model, identify the slag removal areas that need to be removed, and match the corresponding slag removal depth for each slag removal area, and then proceed to step S5;

[0010] S4: using a detector to detect the real-time position of the ingot mold, and using a Kalman filter algorithm to correct the slag removal position of the ingot mold in the transport state, and then proceeding to step S5;

[0011] S5: Based on the slag removal area, slag removal depth and slag removal position information of the ingot mold, multiple slag removal trajectory points are configured and numbered in order of action; then each slag removal trajectory point is connected in series in the order of numbering to form a complete slag removal work trajectory, and the slag removal robot is enabled to perform slag removal; wherein each slag removal trajectory point includes a fixed motion point and a dynamic adjustment point.

[0012] Furthermore, the step S2 is specifically as follows:

[0013] S21: Filtering the pre-processed molten metal surface image in multiple directions and at multiple scales using a Gabor filter bank; each Gabor filter is generated by modulating a Gaussian kernel function with a complex sine function; then, extracting the real and imaginary components of the Gabor filter; wherein the real component uses an even-symmetric filter to extract texture contrast information, and the imaginary component uses an odd-symmetric filter to extract texture phase information;

[0014] S22: Convolving the molten metal surface image with a Gabor filter bank to generate multiple scale response maps, and calculating the mean, variance, energy, and entropy of each scale response map; then, constructing a texture feature vector set of multidimensional texture feature values ​​of the molten metal surface based on the calculated results of the mean, variance, energy, and entropy;

[0015] S23: Each texture feature value in the texture feature vector set is compared with a preset standard texture feature parameter in the visual judgment device to determine whether the quality of the molten metal surface meets the standard. If it does not meet the standard, steps S3 and S4 are triggered simultaneously. If it meets the standard, the process is exited. The preset standard feature parameters are obtained through statistical learning of historical qualified samples.

[0016] Furthermore, the step S21 is specifically as follows:

[0017] Firstly, the product of Gaussian function and complex sine function is used as Gaussian kernel function to construct multi-scale and multi-directional Gabor filter bank.

[0018] The Gaussian kernel function formula is specifically expressed as:

[0019] ; (1)

[0020] ; (2)

[0021] Then, directly extract the real and imaginary components of the Gabor filter. and the imaginary component Respectively expressed as:

[0022] ; (3)

[0023] ; (4)

[0024] Among them, x and y are the coordinate positions of the pixel value on the x-axis and y-axis respectively in the original coordinate system; angle is the direction factor of the filter; and They are the pixel values ​​in the new coordinate system after the rotation transformation, respectively axis, The coordinate position on the axis, wherein the original coordinate system is rotated to align the Gabor filter along the direction of the angle θ; Represents the trigonometric function wavelength, which is used to control the scale of the filter; represents the phase shift of the trigonometric function; represents the standard deviation of the Gaussian function; It represents the spatial aspect ratio, which is used to determine the ellipticity of the Gaussian kernel function; i represents the imaginary unit, which is used to construct the imaginary part of the complex number.

[0025] In step S22, the average value and variance The formulas are expressed as:

[0026] ; (5)

[0027] ; (6)

[0028] Where m and n represent the length and width of the image size respectively; Represents the grayscale value of the pixel; Represents the relevant grid position; v represents the filter direction, and u represents the filter scale;

[0029] Then, according to the calculation results of the average value α and the variance value β, a vector set of the surface texture feature values ​​of the molten metal is constructed. , specifically expressed as:

[0030] ; (7)

[0031] in, Represents the mean amplitude of the response graph at each scale u-1 and direction v-1, Indicates the amplitude variance of the response map at each scale u-1 and direction v-1.

[0032] Furthermore, the step S3 is specifically as follows:

[0033] S31: Build and train a slag removal prediction model based on the DenseNet model; at the same time, use the weighted cross entropy loss function to correct and optimize the prediction results of the slag removal prediction model;

[0034] S32: Divide the inner contour area of ​​the ingot mold in the target photo and number each inner contour area in sequence; calculate each inner contour area using the trained slag removal prediction model to obtain a slag removal prediction value;

[0035] S33: Preset a slag removal judgment interval according to the slag removal prediction model; determine in turn whether the slag removal prediction value of each inner contour area falls within the slag removal judgment interval, and then determine the inner contour area falling within the slag removal judgment interval as the slag removal area;

[0036] S34: Preset a number of slag removal depth intervals using an equal-frequency binning method based on historical data distribution, and different slag removal depth intervals are associated with different slag removal depths; and determine the slag removal depth of each slag removal area based on the slag removal depth interval into which the slag removal prediction value in the slag removal area falls.

[0037] Furthermore, the step S5 is specifically as follows:

[0038] S51: Establishing a three-dimensional coordinate system, and then setting multiple slag scooping trajectory points that the slag scooping robot needs to pass through when scooping slag, and numbering them according to the order of action; wherein the multiple slag scooping trajectory points include one or more preset fixed motion points and one or more dynamically adjusted points;

[0039] S52: Determine the three-dimensional coordinates of each dynamic adjustment point according to the slag removal area determined by the visual judgment device, the predicted slag removal depth, and the predicted slag removal position of the ingot mold;

[0040] S53: connecting the various slag removal trajectory points in series in the order of action to form a complete slag removal work trajectory, and enabling the slag removal robot to perform slag removal when the ingot mold reaches the predicted slag removal position.

[0041] Furthermore, in step S51, there are five slag scooping trajectory points, which are represented in order of action as follows: pre-slag scooping posture point P1 (x1, y1, z1), slag scooping starting point P2 (x2, y2, z2), slag scooping ending point P3 (x3, y3, z3), slag starting point P4 (x4, y4, z4) and slag scooping recovery point P5 (x5, y5, z5).

[0042] Furthermore, in step S51, the pre-slag removal posture point P1, the slag starting point P4 and the slag recovery point P5 are preset fixed motion points, and the slag removal starting point P2 and the slag removal ending point P3 are dynamic adjustment points that need to be adjusted in position during each slag removal process.

[0043] Furthermore, after the slag collecting robot has completed a complete slag collecting work trajectory, it is reset from the slag collecting recovery point P5 to the pre-slag removing posture point P1.

[0044] A slag removal device adopts the slag removal control method for medium-sized metal ingots as described in any of the above items, comprising a visual judgment device, a slag removal robot and a control system; the visual judgment device is arranged directly above the ingot mold entering the position to be detected; the visual judgment device comprises a camera, a filter group and a detector; the camera is used to take a target photo; the filter group includes a plurality of Gabor filters for identifying and analyzing the texture features of the metal liquid surface in the target photo from multiple directions and multiple scales; the detector is used to detect the position of the ingot mold in a transport state in real time; the control system is electrically connected to the visual judgment device and the slag removal robot respectively; the control system has a built-in data processing module; the data processing module is used to determine whether the ingot mold needs slag removal, and when slag removal is required, further confirm the slag removal area, slag removal depth, and ingot mold slag removal position information to configure the slag removal work trajectory.

[0045] The beneficial effects of the present invention are:

[0046] 1. This invention uses the texture features of the molten metal surface as the basis for slag removal judgment, changing the traditional slag removal judgment method based on image noise recognition. It is more suitable for slag removal scenarios of medium-sized metal ingots. Based on the acquired texture features, the DenseNet large model is used to predict and judge the slag removal area. The position of the ingot mold is corrected to ensure the precision and accuracy of slag removal.

[0047] 2. In the process of establishing a complete slag scooping working trajectory, the present invention sets three fixed motion points whose positions remain unchanged during each slag scooping process and two dynamic adjustment points that need to be adjusted according to the actual production conditions of the production line. During each slag scooping process, only the positions of the two dynamic adjustment points need to be changed, and the positions of the other three points do not change. This greatly reduces the slag scooping amplitude, prevents the surface of the metal ingot from generating ripples due to the large movement amplitude, and can ensure the quality of the finished metal ingot. In addition, this design can also greatly reduce the amount of calculation for the slag scooping trajectory points. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the slag removal control method for a medium-sized metal ingot provided by the present invention;

[0049] Figure 2 It is a foreground extraction image of the metal liquid surface image in the slag removal control method for medium-sized metal ingots provided by the present invention;

[0050] Figure 3 It is the inner contour area division diagram of the ingot mold in step S3 in the slag removal control method for medium-sized metal ingots provided by the present invention; DETAILED DESCRIPTION

[0051] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying 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 limiting the present invention.

[0052] In the description of the present invention, "first feature" or "second feature" may include one or more of the features.

[0053] In the description of the present invention, "plurality" means two or more.

[0054] In the description of the present invention, a first feature being “on” or “under” a second feature may include the first and second features being in direct contact with each other, or the first and second features not being in direct contact with each other but being in contact with each other via another feature therebetween.

[0055] In the description of the present invention, “on”, “above” and “above” a first feature of a second feature include the first feature being directly above and obliquely above the second feature, or simply means that the first feature is horizontally higher than the second feature.

[0056] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," and "some examples" 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 invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but the present invention is not limited to the following embodiments.

[0058] Example 1:

[0059] See attached Figures 1 and 2 A method for controlling slag removal of a medium-sized metal ingot comprises the following steps:

[0060] S1: configuring a visual judgment device, which includes a camera, a Gabor filter bank, and a detector; using the camera to obtain a target photo carrying image information of the ingot mold and the surface of the molten metal in the ingot mold and pre-processing it, and then proceeding to step S2;

[0061] S2: Using a Gabor filter bank to extract texture features from the preprocessed molten metal surface image; based on the texture features of the molten metal surface image, determine whether the surface quality of the molten metal in the target photo meets the standards; if not, trigger steps S3 and S4 simultaneously; if it meets the standards, exit the steps;

[0062] S3: Build and train a slag removal prediction model based on the DenseNet model; input the target photo into the slag removal prediction model, identify the slag removal areas that need to be removed, and match the corresponding slag removal depth for each slag removal area, and then proceed to step S5;

[0063] S4: using a detector to detect the real-time position of the ingot mold, and using a Kalman filter algorithm to correct the slag removal position of the ingot mold in the transport state, and then proceeding to step S5;

[0064] S5: Based on the slag removal area, slag removal depth and slag removal position information of the ingot mold, multiple slag removal trajectory points are configured and numbered in order of action; then each slag removal trajectory point is connected in series in the order of numbering to form a complete slag removal work trajectory, and the slag removal robot is enabled to perform slag removal; wherein each slag removal trajectory point includes a fixed motion point and a dynamic adjustment point.

[0065] Currently, in the process of producing medium-sized metal ingots, a conveying mechanism is used to continuously convey the ingot mold along the production process direction. For medium-sized metal ingots, when judging whether to perform the slag removal step, the metal in the ingot mold is in a molten liquid state and is in the cooling process, and some slag remains on the surface of the molten metal. During the cooling process of the molten metal, the surface roughness of the molten metal will increase due to the slag attached to the surface, and even local depressions will occur, thereby producing obvious texture features. Therefore, the surface quality of the molten metal can be judged by extracting the texture features of the molten metal surface image information, thereby determining whether slag removal is required. When the ingot mold is transported to the visual judgment device by the conveying mechanism, the present invention will obtain the molten metal surface image information through the visual judgment device, and through the analysis and comparison of the texture features, it will be judged whether the quality of the molten metal surface in the current ingot mold meets the standards, thereby determining whether slag removal is performed on the ingot mold.

[0066] In step S1, when the ingot mold enters the recognition range of the visual judgment device, the target photo is taken by the camera, and then the image information of the surface of the molten metal cooled in the ingot mold is obtained and pre-processed. The image information of the foreground of the molten metal surface is extracted as follows: Figure 2 The image preprocessing method is a well-known prior art, so it will not be described here in detail.

[0067] For step S2, the present invention utilizes the texture characteristics of the molten metal surface during the cooling process of the metal ingot after casting as a criterion for judging whether the surface quality of the metal ingot meets the standards, thereby changing the conventional method of using image noise to judge the surface quality and being more suitable for small and medium-sized metal ingots.

[0068] For step S3, the present invention obtains target photos containing image information of the ingot mold and the metal liquid surface in the ingot mold generated in the past production line, and performs computational training on the DenseNet model with a large sample size to ensure the accuracy of the predicted sample classification in the work.

[0069] For step S4, since the ingot mold is continuously transported by the conveying mechanism along the production process direction during the slag removal process, it is necessary to correct the position of the ingot mold during the slag removal. The Kalman filter algorithm is an existing technology. Position correction through this algorithm can further ensure the accuracy of the slag removal position.

[0070] For step S5, the slag scooping trajectory points when the slag scooping robot actually performs slag scooping are divided into fixed motion points and dynamic adjustment points. This allows only the positions of the dynamic adjustment points to be changed during each slag scooping process, while the other fixed motion points remain unchanged. This greatly reduces the slag scooping amplitude, prevents the surface of the metal ingot from generating ripples due to the large movement amplitude, and can ensure the quality of the finished metal ingot. In addition, this design can also reduce the amount of calculation on the slag scooping trajectory points.

[0071] Example 2:

[0072] See attached Figure 1 Based on the first embodiment, the step S2 is specifically as follows:

[0073] S21: Filtering the pre-processed molten metal surface image in multiple directions and at multiple scales using a Gabor filter bank; each Gabor filter is generated by modulating a Gaussian kernel function with a complex sine function; then, extracting the real and imaginary components of the Gabor filter; wherein the real component uses an even-symmetric filter to extract texture contrast information, and the imaginary component uses an odd-symmetric filter to extract texture phase information;

[0074] S22: Convolving the molten metal surface image with a Gabor filter bank to generate multiple scale response maps, and calculating the mean, variance, energy, and entropy of each scale response map; then, constructing a texture feature vector set of multidimensional texture feature values ​​of the molten metal surface based on the calculated results of the mean, variance, energy, and entropy;

[0075] S23: Compare each texture feature value in the texture feature vector set with the standard texture feature parameters preset in the visual judgment device to determine whether the quality of the molten metal surface meets the standard; if not, trigger steps S3 and S4 at the same time; if so, exit the step.

[0076] The visual judgment device of the present invention first uses a camera to take a target photo, and then pre-processes the target photo to facilitate subsequent filtering and feature extraction by the Gabor filter group. The Gabor filter group includes a plurality of Gabor filters, and the plurality of Gabor filters can perform multi-directional and multi-scale filtering on the received molten metal surface image, and obtain multi-directional and multi-scale texture feature values ​​through calculation, and judge whether the quality of the molten metal surface meets the standards based on the texture feature values. Specifically, the Gabor filter can adopt the existing 2d-Gabor filter, and the preset standard feature parameters are obtained through statistical learning of historical qualified samples. The method for calculating the energy and entropy values ​​in each scale response diagram is existing technology and will not be repeated in the present invention.

[0077] Example 3:

[0078] See attached Figure 1 Based on the second embodiment, the step S21 is specifically as follows:

[0079] Firstly, the product of Gaussian function and complex sine function is used as Gaussian kernel function to construct multi-scale and multi-directional Gabor filter bank.

[0080] The Gaussian kernel function formula is specifically expressed as:

[0081] ; (1)

[0082] ; (2)

[0083] Then, directly extract the real and imaginary components of the Gabor filter. and the imaginary component Respectively expressed as:

[0084] ; (3)

[0085] ; (4)

[0086] Among them, x and y are the coordinate positions of the pixel value on the x-axis and y-axis respectively in the original coordinate system; angle is the direction factor of the filter; and They are the pixel values ​​in the new coordinate system after the rotation transformation, respectively axis, The coordinate position on the axis, wherein the original coordinate system is rotated to align the Gabor filter along the direction of the angle θ; Represents the trigonometric function wavelength, which is used to control the scale of the filter; represents the phase shift of the trigonometric function; represents the standard deviation of the Gaussian function; It represents the spatial aspect ratio, which is used to determine the ellipticity of the Gaussian kernel function; i represents the imaginary unit, which is used to construct the imaginary part of the complex number.

[0087] In step S22, the average value and variance The formulas are expressed as:

[0088] ; (5)

[0089] ; (6)

[0090] Where m and n represent the length and width of the image size respectively; Represents the grayscale value of the pixel; Represents the relevant grid position; v represents the filter direction, and u represents the filter scale;

[0091] Then, according to the calculation results of the average value α and the variance value β, a vector set of the surface texture feature values ​​of the molten metal is constructed. , specifically expressed as:

[0092] ; (7)

[0093] in, Represents the mean amplitude of the response graph at each scale u-1 and direction v-1, Indicates the amplitude variance of the response map at each scale u-1 and direction v-1.

[0094] Preferably, in step S2, the Gabor filter bank extracts the texture features of the metal liquid surface image in the target photo in 5 scales and 8 directions, and the kernel function is used to extract the imaginary and real components, and the vector set of texture feature values ​​actually obtained is: The dimension is 160, which ensures the accuracy of judging the position of the oxide slag on the surface of the molten metal.

[0095] The present invention combines the spatial locality of the Gaussian kernel function with the frequency domain selectivity of the complex sine function, and constructs a multi-scale texture feature analysis system for the surface pattern of the molten metal based on a multi-scale and multi-directional Gabor filter group, which can be applied to complex industrial scenarios. Specifically, the present invention forms an amplitude-phase joint analysis model by extracting the real and imaginary features. Compared with the traditional grayscale analysis method, it can greatly improve the accuracy of sample classification; it can also dynamically adjust the wavelength parameters to obtain the desired texture feature. The pixel value of the sensor achieves full-scale coverage from macroscopic slag spots (size > 3mm) to microscopic cracks (size < 0.5mm). Compared with traditional single-scale algorithms, this significantly reduces detection blind spots. In addition, by accurately capturing texture features in different flow directions, the detection rate of oblique defects is improved. Furthermore, while maintaining 0.1mm-level spatial positioning accuracy, it effectively suppresses metal liquid reflections and bubble noise.

[0096] Example 4:

[0097] See attached Figure 1 Based on the third embodiment, the step S3 is specifically as follows:

[0098] S31: Build and train a slag removal prediction model based on the DenseNet model; at the same time, use the weighted cross entropy loss function to correct and optimize the prediction results of the slag removal prediction model;

[0099] S32: Divide the inner contour area of ​​the ingot mold in the target photo and number each inner contour area in sequence; calculate each inner contour area using the trained slag removal prediction model to obtain a slag removal prediction value;

[0100] S33: Preset a slag removal judgment interval according to the slag removal prediction model; determine in turn whether the slag removal prediction value of each inner contour area falls within the slag removal judgment interval, and then determine the inner contour area falling within the slag removal judgment interval as the slag removal area;

[0101] S34: Preset a number of slag removal depth intervals using an equal-frequency binning method based on historical data distribution, and different slag removal depth intervals are associated with different slag removal depths; and determine the slag removal depth of each slag removal area based on the slag removal depth interval into which the slag removal prediction value in the slag removal area falls.

[0102] During the casting process of metal ingots, the ingot casting machine transports the ingot mold to the bottom of the smelting furnace through a conveying mechanism such as a chain. The smelting furnace continuously casts the high-temperature molten alloy into multiple ingot molds. Therefore, the visual judgment device needs to make a separate judgment on each ingot mold when making a slag removal judgment. Although the surface characteristics of the molten metal will continue to change during the casting and cooling process, the distribution of oxide slag on the surface of the molten metal changes slowly and is relatively constant. Therefore, the required slag removal area and slag removal depth can be determined by judging the distribution and characteristics of the oxide slag. The present invention obtains nearly 2,000 target photos carrying image information of the ingot mold and the molten metal surface in the ingot mold during past production processes, and performs large-sample computational training on the DenseNet model. At the same time, the weighted cross entropy loss function is used to correct and optimize the prediction results of the model to ensure the accuracy of the predicted sample classification. When the target photo is input into the trained slag removal prediction model, the inner contour area of ​​the ingot mold is first divided. Then, by analyzing the distribution of slag in each area, the slag removal prediction model calculates the corresponding slag removal prediction value for each area. The slag removal prediction value is then used to determine whether to remove slag from the area based on whether it falls within the slag removal judgment interval. Then, based on the distribution of historical data obtained in the past production process, the equal-frequency binning method is used to preset several slag removal depth intervals that match different slag removal judgment intervals. In actual application, the slag removal depth corresponding to each slag removal area is determined based on the slag removal depth interval that the slag removal prediction value falls into, guiding the slag removal device to remove slag. Among them, the equal-frequency binning method is a prior art and will not be described in detail here.

[0103] Regarding step S4, since the ingot mold is being continuously transported on the production line during the judgment and execution of slag removal, although the visual judgment device obtains the mold position information and speed information in real time, the obtained parameters are still affected by factors such as network communication delay, image processing algorithm operation speed, and frame rate. There may even be shaking of the ingot mold during transportation, which may cause the parameters obtained by the visual judgment device to be inconsistent with the current actual state of the target. Therefore, the present invention uses the Kalman filter algorithm to perform position correction. The Kalman filter algorithm is a prior art. Specifically, the current position information and speed information of the ingot mold in transportation can be obtained through the built-in detector of the visual judgment device, and a motion model of the ingot mold can be established. Then, the change of the ingot mold position over time is predicted, and the covariance matrix is ​​used to eliminate errors. Then, the corrected predicted ingot mold slag removal position is obtained, reducing position errors and ensuring the accuracy of subsequent slag removal.

[0104] Embodiment 5:

[0105] See attached Figure 1 Based on the fourth embodiment, the step S5 is specifically as follows:

[0106] S51: Establishing a three-dimensional coordinate system, and then setting multiple slag scooping trajectory points that the slag scooping robot needs to pass through when scooping slag, and numbering them according to the order of action; wherein the multiple slag scooping trajectory points include one or more preset fixed motion points and one or more dynamically adjusted points;

[0107] S52: Determine the three-dimensional coordinates of each dynamic adjustment point according to the slag removal area determined by the visual judgment device, the predicted slag removal depth, and the predicted slag removal position of the ingot mold;

[0108] S53: connecting the various slag removal trajectory points in series in the order of action to form a complete slag removal work trajectory, and enabling the slag removal robot to perform slag removal when the ingot mold reaches the predicted slag removal position.

[0109] In the step S51, there are five slag scooping trajectory points, which are represented in order of action as follows: pre-slag scooping posture point P1 (x1, y1, z1), slag scooping starting point P2 (x2, y2, z2), slag scooping ending point P3 (x3, y3, z3), slag starting point P4 (x4, y4, z4) and slag scooping recovery point P5 (x5, y5, z5).

[0110] In step S51, the pre-slagging posture point P1, the slag starting point P4 and the slag recovery point P5 are preset fixed motion points, and the slag scooping starting point P2 and the slag scooping ending point P3 are dynamic adjustment points that need to be adjusted in each slag scooping process.

[0111] After the slag collecting robot has completed a complete slag collecting working trajectory, it resets from the slag collecting recovery point P5 to the pre-slag removing posture point P1.

[0112] In the process of establishing a complete slag scooping work trajectory, the present invention collects relevant data based on the actual production conditions of the past production line, and presets three fixed motion points whose positions remain unchanged during each slag scooping process and two dynamic adjustment points that need to be adjusted according to the changes in each slag scooping process. Therefore, the present invention only needs to change the positions of two dynamic adjustment points during each slag scooping process, and the other three points do not change. When slag removal is required, the starting point for each slag removal robot is preset based on the specific location of slag removal during transportation of the ingot mold during previous production processes, namely the pre-slag removal posture point P1. Then, based on the slag removal starting point P2 and slag removal ending point P3 calculated before the slag removal, the slag removal components of the slag removal robot, such as the slag removal scraper, are extended into the slag removal starting point P2 below the molten metal surface of the ingot mold to start slag removal and end slag removal at slag removal ending point P3. However, at this time, the slag removal scraper remains below the molten metal surface of the ingot mold until it moves to the preset slag removal point P4. The slag removed by the slag removal scraper is then transported to the slag recovery point P5 for slag recovery. The three-dimensional coordinates of P2 and P3 need to be determined based on the thickness and distribution of the slag in each ingot mold to ensure that each slag removal can be adjusted to the appropriate slag removal point and depth, effectively improving slag removal efficiency and quality. Furthermore, this slag removal trajectory significantly reduces the amplitude of each slag removal operation. This prevents ripples on the surface of the metal ingots formed after the molten metal cools due to the large amplitude of the slag removal movement, or significant variations in the surface texture of the metal ingots formed from different casting molds within the same batch due to differences in the slag removal trajectory. Furthermore, the fixed motion points significantly reduce the computational effort required for the slag removal trajectory.

[0113] Embodiment 5:

[0114] See attached Figure 1 Another aspect of the present invention further discloses a slag scooping device, which adopts the slag scooping control method for a medium-sized metal ingot as described in any one of the first to fourth embodiments.

[0115] A slag removal device comprises a visual judgment device, a slag removal robot and a control system; the visual judgment device comprises a camera, a filter group and a detector; the camera is used to take a target photo when an ingot mold passes directly below the camera; the filter group comprises a plurality of Gabor filters for identifying and analyzing the texture features of the metal liquid surface in the target photo from multiple directions and multiple scales; the detector is used to detect the position of the ingot mold in a transport state in real time; the control system is electrically connected to the visual judgment device and the slag removal robot respectively; the control system has a built-in data processing module; the data processing module is used to determine whether the ingot mold needs slag removal, and if slag removal is required, further confirm the slag removal area, slag removal depth, and slag removal position information of the ingot mold to configure the slag removal work trajectory.

[0116] It can be seen from the above structure that the ingot mold is continuously transported along the production line through the conveying mechanism during the production process of metal ingots. A shooting station is set on the ingot mold conveying route, and the camera is set directly above the shooting station. When the ingot mold is conveyed to directly below the camera, the camera takes a target photo carrying image information of the ingot mold and the molten metal surface in the ingot mold, and then the filter group is used to identify and analyze the texture features of the molten metal surface in the target photo from multiple directions and scales, and then transport it to the control system. The data is processed by the built-in data processing module of the control system to determine whether the molten metal in the current ingot mold needs slag removal. If slag removal is required, the slag removal area, slag removal depth, and slag removal position information of the ingot mold are further confirmed, and then the slag removal trajectory points are configured to form a complete slag removal work trajectory. Finally, the control system is used to control the slag removal robot to perform slag removal according to the slag removal work trajectory.

[0117] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A slag removal control method for a medium-sized metal ingot, characterized in that: The following steps are involved: S1: configuring a visual judgment device, which includes a camera, a Gabor filter bank, and a detector; using the camera to obtain a target photo carrying image information of the ingot mold and the surface of the molten metal in the ingot mold and pre-processing it, and then proceeding to step S2; S2: Use Gabor filter bank to extract texture features from the pre-processed metal liquid surface image; Based on the texture features of the molten metal surface image, determine whether the surface quality of the molten metal in the target photo meets the standards; if not, trigger steps S3 and S4 simultaneously; if it meets the standards, exit the steps; S3: Build and train a slag removal prediction model based on the DenseNet model; input the target photo into the slag removal prediction model, identify the slag removal areas that need to be removed, and match the corresponding slag removal depth for each slag removal area, and then proceed to step S5; S4: using a detector to detect the real-time position of the ingot mold, and using a Kalman filter algorithm to correct the slag removal position of the ingot mold in the transport state, and then proceeding to step S5; S5: Based on the slag removal area, slag removal depth and slag removal position information of the ingot mold, multiple slag removal trajectory points are configured and numbered in order of action; then each slag removal trajectory point is connected in series in the order of numbering to form a complete slag removal work trajectory, and the slag removal robot is enabled to perform slag removal; wherein each slag removal trajectory point includes a fixed motion point and a dynamic adjustment point.

2. The slag removal control method for medium-sized metal ingots according to claim 1, characterized in that: The step S2 is specifically as follows: S21: Filter the pre-processed molten metal surface image in multiple directions and at multiple scales using a Gabor filter bank; each Gabor filter is generated by modulating a Gaussian kernel function and a complex sine function; Then, the real and imaginary components of the Gabor filter are extracted; the real component is used to extract texture contrast information, and the imaginary component is used to extract texture phase information; S22: Convolving the molten metal surface image with a Gabor filter bank to generate multiple scale response maps, and calculating the mean, variance, energy, and entropy of each scale response map; then, constructing a texture feature vector set of multidimensional texture feature values ​​of the molten metal surface based on the calculated results of the mean, variance, energy, and entropy; S23: Compare each texture feature value in the texture feature vector set with the standard texture feature parameters preset in the visual judgment device to determine whether the quality of the molten metal surface meets the standard; if not, trigger steps S3 and S4 at the same time; if so, exit the step.

3. The slag removal control method for medium-sized metal ingots according to claim 2, characterized in that: The step S21 is specifically as follows: Firstly, the product of Gaussian function and complex sine function is used as Gaussian kernel function to construct multi-scale and multi-directional Gabor filter bank. The Gaussian kernel function formula is specifically expressed as: ；（1） ； （2） Then, directly extract the real and imaginary components of the Gabor filter. and the imaginary component Respectively expressed as: ；（3） ；（4） Among them, x and y are the coordinate positions of the pixel value on the x-axis and y-axis respectively in the original coordinate system; angle is the direction factor of the filter; and They are the pixel values ​​in the new coordinate system after the rotation transformation, respectively axis, Coordinate position on the axis; represents the trigonometric wavelength; represents the phase shift of the trigonometric function; represents the standard deviation of the Gaussian function; represents the spatial aspect ratio; i represents the imaginary unit.

4. The slag removal control method for a medium-sized metal ingot according to claim 2, wherein: In step S22, the average value and variance The formulas are expressed as: ；（5） ；（6） Where m and n represent the length and width of the image size respectively; Represents the grayscale value of the pixel; Represents the relevant grid position; v represents the filter direction, and u represents the filter scale; Then, according to the calculation results of the average value α and the variance value β, a vector set of the surface texture feature values ​​of the molten metal is constructed. , specifically expressed as: ；（7） in, Represents the mean amplitude of the response graph at each scale u-1 and direction v-1, Indicates the amplitude variance of the response map at each scale u-1 and direction v-1.

5. The slag removal control method for medium-sized metal ingots according to claim 1, characterized in that: The step S3 is specifically as follows: S31: Build and train a slag removal prediction model based on the DenseNet model; at the same time, use the weighted cross entropy loss function to correct and optimize the prediction results of the slag removal prediction model; S32: Divide the inner contour area of ​​the ingot mold in the target photo and number each inner contour area in sequence; calculate each inner contour area using the trained slag removal prediction model to obtain a slag removal prediction value; S33: Preset a slag removal judgment interval according to the slag removal prediction model; determine in turn whether the slag removal prediction value of each inner contour area falls within the slag removal judgment interval, and then determine the inner contour area falling within the slag removal judgment interval as the slag removal area; S34: Preset a number of slag removal depth intervals using an equal-frequency binning method based on historical data distribution, and different slag removal depth intervals are associated with different slag removal depths; and determine the slag removal depth of each slag removal area based on the slag removal depth interval into which the slag removal prediction value in the slag removal area falls.

6. The slag removal control method for medium-sized metal ingots according to claim 1, characterized in that: The step S5 is specifically as follows: S51: Establishing a three-dimensional coordinate system, and then setting multiple slag scooping trajectory points that the slag scooping robot needs to pass through when scooping slag, and numbering them according to the order of action; wherein the multiple slag scooping trajectory points include one or more preset fixed motion points and one or more dynamically adjusted points; S52: Determine the three-dimensional coordinates of each dynamic adjustment point based on the slag removal area determined by the visual judgment device, the predicted slag removal depth, and the predicted slag removal position of the ingot mold; S53: connecting the various slag removal trajectory points in series in the order of action to form a complete slag removal work trajectory, and when the ingot mold reaches the predicted slag removal position, enabling the slag removal robot to perform slag removal.

7. The slag removal control method for medium-sized metal ingots according to claim 6, characterized in that: In the step S51, there are five slag scooping trajectory points, which are represented in order of action as follows: pre-slag scooping posture point P1 (x1, y1, z1), slag scooping starting point P2 (x2, y2, z2), slag scooping ending point P3 (x3, y3, z3), slag starting point P4 (x4, y4, z4) and slag scooping recovery point P5 (x5, y5, z5).

8. The slag removal control method for medium-sized metal ingots according to claim 7, characterized in that: In step S51, the pre-slagging posture point P1, the slag starting point P4 and the slag recovery point P5 are preset fixed motion points, and the slag scooping starting point P2 and the slag scooping ending point P3 are dynamic adjustment points that need to be adjusted in each slag scooping process.

9. The slag removal control method for medium-sized metal ingots according to claim 7, characterized in that: After the slag collecting robot has completed a complete slag collecting working trajectory, it resets from the slag collecting recovery point P5 to the pre-slag removing posture point P1.

10. A slag removal device, using the slag removal control method for a medium-sized metal ingot according to any one of claims 1 to 9, characterized in that: The system comprises a visual judgment device, a slag removal robot and a control system; the visual judgment device comprises a camera, a filter group and a detector; the camera is used to take a target photo when the ingot mold passes directly below the camera; the filter group comprises a plurality of Gabor filters, which are used to identify and analyze the texture features of the metal liquid surface in the target photo from multiple directions and multiple scales; the detector is used to detect the position of the ingot mold in a transport state in real time; the control system is electrically connected to the visual judgment device and the slag removal robot respectively; the control system has a built-in data processing module; the data processing module is used to judge whether the ingot mold needs slag removal, and if slag removal is required, further confirm the slag removal area, slag removal depth and slag removal position information of the ingot mold to configure the slag removal work trajectory.

Images