Multi-mode sensing slag salvaging system and method for electrolytic aluminum workshop

Abstract

The invention provides a multi-mode sensing slag salvaging system and method for an electrolytic aluminum workshop, the system comprises an acquisition module, a mapping module, an identification module, a path planning module and an execution module, the acquisition module is configured to acquire multi-mode data in the electrolytic aluminum workshop; the mapping module is configured to perform fusion processing on the multi-modal data and construct a multi-modal-based fusion map; the identification module is configured to match the features of the electrolytic cell from the fusion map based on pre-stored feature data, fit the height of the molten aluminum plane and calculate the distances from all points in the features of the electrolytic cell to the molten aluminum plane so as to obtain slag features; the path planning module is configured to construct a slag hopper geometric model based on the slag hopper size, and generate a slag salvaging path in combination with the slag hopper geometric model, the electrolytic bath characteristics and the slag characteristics; and the execution module is configured to control the crown block and the slag bucket to move based on the slag salvaging path so as to execute slag salvaging operation.

Description

Technical Field

[0001] This invention relates to the field of slag removal technology for molten aluminum in electrolytic cells, and particularly to a multimodal sensing slag removal system and method for use in electrolytic aluminum workshops. Background Technology

[0002] Aluminum electrolysis is the process of producing metallic aluminum by electrolyzing bauxite (primarily composed of alumina, Al₂O₃). This process takes place in an electrolytic cell, a container made of reinforced concrete lined with graphite or other corrosion-resistant materials to withstand the high temperatures and corrosiveness of the molten aluminum. During electrolysis, aluminum ions in the molten aluminum are reduced to metallic aluminum, which is deposited on the cathode (usually made of aluminum alloy), while the graphite anode undergoes an oxidation reaction, producing carbon dioxide and other gases. Simultaneously, slag (mainly composed of carbonates and other impurities) is produced as a byproduct during electrolysis. As electrolysis progresses, slag accumulates on the surface of the electrolytic cell, especially in the anode area. This slag is mainly composed of unreacted impurities, byproducts of the reaction, and the consumption of the graphite anode. Therefore, timely cleaning is necessary. Failure to remove it promptly will not only reduce electrolysis efficiency and affect product quality but may also lead to a series of problems such as abnormal consumption of the anode carbon block and frequent occurrences of the anode effect. The existing slag removal operation still mainly relies on manually controlled overhead cranes. The operation process is limited by the experience of the personnel, which not only results in high labor intensity and low efficiency, but also poses safety hazards such as high temperature and electric shock.

[0003] With the intelligent upgrading of traditional manufacturing, some manufacturers use a combination of machine vision and remote control for automatic slag removal. However, the high temperature and dust in the workshop can lead to low accuracy of machine vision in slag identification. In addition, there are many interference factors when using machines to identify slag, which increases the amount of calculation required for identification. Therefore, it is still necessary to manually identify the slag and mark its position in the molten aluminum so that the overhead crane can drive the slag bucket to remove the slag according to its position. As a result, it is impossible to achieve fully automatic slag removal. In view of this, the present invention is proposed. Summary of the Invention

[0004] To address the problems of high computational load and low recognition accuracy in existing machine time-based slag removal schemes, this invention provides a technical field for slag removal from molten aluminum in electrolytic cells.

[0005] To address the aforementioned technical problems, this invention provides a multimodal sensing slag removal system for an aluminum electrolysis workshop. The system comprises a data acquisition module, a mapping module, a recognition module, a path planning module, and an execution module. The data acquisition module is configured to acquire multimodal data within the aluminum electrolysis workshop. The mapping module is configured to fuse the multimodal data and construct a fused map based on the multimodal data. The recognition module is configured to match electrolytic cell features from the fused map based on pre-stored feature data, fit the height of the molten aluminum plane, and calculate the distances between all points within the electrolytic cell features and the molten aluminum plane to obtain slag features. The path planning module is configured to construct a slag hopper geometric model based on the slag hopper dimensions and generate a slag removal path by combining the slag hopper geometric model, electrolytic cell features, and slag features. The execution module is configured to control the movement of the overhead crane and slag hopper based on the slag removal path to perform the slag removal operation. The multimodal data includes point cloud data, thermal data, and texture data.

[0006] In an embodiment of the present invention, the mapping module includes a correction unit, a filtering unit, a denoising unit, and a fusion unit. The correction unit is configured to: use the coordinates of the point cloud data as a reference and use preset compensation parameters to correct the coordinates of the thermal data and texture data, so that the point cloud data, thermal data, and texture data achieve coordinate alignment. The filtering unit is configured to: filter the point cloud data, thermal data, and texture data respectively to reduce the computational load when constructing the fused map. The denoising unit is configured to: use a statistical method to calculate the average distance from each point cloud to its k nearest neighbors, and remove point clouds with a distance exceeding μ ± α·σ, so as to remove interference in the point cloud data, where μ is the mean, σ is the standard deviation, and α is the threshold coefficient. The fusion unit is configured to: use a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then map the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct the fused map.

[0007] In an embodiment of the present invention, the filtering unit includes voxel filtering, thermal filtering, and texture filtering. The voxel filtering unit is configured to: divide the point cloud data into several voxels according to a preset size, and retain only one representative point cloud in each voxel. The thermal filtering unit is configured to: divide the color blocks in the thermal data into several thermal fields, compare them in each thermal field based on a preset temperature fluctuation value, and filter out thermal fields with temperature fluctuations smaller than the preset temperature fluctuation value. The texture filtering unit is configured to: perform texture comparison on the texture data from the filtered thermal fields based on pre-stored slag features to avoid slag with aluminum melt temperature fluctuations smaller than the preset temperature fluctuation value being mistakenly filtered out.

[0008] In an embodiment of the present invention, the identification module includes an electrolytic cell identification unit, a point cloud segmentation unit, an aluminum liquid fitting unit, and a slag extraction unit. The electrolytic cell identification unit is configured to: identify the electrolytic cell from the fused map based on pre-stored electrolytic cell texture features; the point cloud segmentation module is configured to: identify the electrolytic cell based on the electrolytic cell boundary and the known aluminum liquid height range [h]. min ,h max Extracting a subset p of the point cloud inside the electrolytic cell C The aluminum liquid fitting unit is configured to: randomly sample three points from the pC and calculate the equation parameters of the aluminum liquid plane using the following formula. d = -n·p1, where n is the unit normal vector of the molten aluminum plane, d is the constant term offset of the molten aluminum plane, and p1, p2, and p3 are the values ​​from p1 to p1 respectively. c Three sample points are randomly sampled; the slag extraction unit is configured to calculate a subset p of the point cloud. C The distance D from all points to the molten aluminum surface i =|n·p i +d|, if D i If the value is ≤ε, it is marked as an interior point. The final set of interior points is the aluminum melt point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i}, where ε is a preset distance threshold.

[0009] In an embodiment of the present invention, the acquisition module includes a lidar, a thermal imager, and a camera mounted on an overhead crane and moving synchronously with the crane. The lidar is configured to acquire point cloud data within the electrolytic aluminum workshop; the thermal imager is configured to acquire temperature field distribution data within the electrolytic aluminum workshop; and the camera is configured to acquire texture image data within the electrolytic aluminum workshop.

[0010] In an embodiment of the present invention, the multimodal perception slag removal system further includes an obstacle detection module, which is configured to: generate an obstacle detection ROI area on the slag removal path, extract all point clouds of the obstacle detection ROI area in the fused map, and fine-tune the path for possible collision locations.

[0011] In an embodiment of the present invention, the obstacle detection module is further configured to: control the lidar to detect in real time whether new dynamic or unknown obstacles appear on the path ahead during the actual movement of the slag hopper along the slag removal path, and trigger the crane to stop moving and issue obstacle warning information when an obstacle is detected.

[0012] To address the problems in the prior art, this invention also provides a multimodal sensing slag removal method for an electrolytic aluminum workshop, applied to the aforementioned multimodal sensing slag removal system. The method includes: a data acquisition module that collects multimodal data within the electrolytic aluminum workshop; a mapping module that fuses the multimodal data and constructs a fused map based on the multimodal data; a recognition module that matches electrolytic cell features from the fused map based on pre-stored feature data, fits the height of the molten aluminum plane, and calculates the distances between all points within the electrolytic cell features and the molten aluminum plane to obtain slag features; a path planning module that constructs a slag hopper geometric model based on the slag hopper dimensions, and generates a slag removal path by combining the slag hopper geometric model, electrolytic cell features, and slag features; and an execution module that controls the movement of the overhead crane and slag hopper based on the slag removal path to perform the slag removal operation. The multimodal data includes point cloud data, thermal data, and texture data.

[0013] In an embodiment of the present invention, the step of fusing multimodal data and constructing a multimodal-based fused map further includes: a correction unit using the coordinates of point cloud data as a reference and using preset compensation parameters to correct the coordinates of the thermal data and texture data, so that the point cloud data, thermal data, and texture data achieve coordinate alignment; a filtering unit filtering the point cloud data, thermal data, and texture data respectively to reduce the computational load when constructing the fused map; a denoising unit using statistical methods to calculate the average distance from each point cloud to its k nearest neighbors and removing point clouds with distances exceeding μ ± α·σ, to remove interference in the point cloud data, where μ is the mean, σ is the standard deviation, and α is the threshold coefficient; and a fusion unit using a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then mapping the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct the fused map.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] By collecting multimodal data consisting of point cloud data, thermal data, and texture data, and filtering and denoising the multimodal data before constructing the fused map, the computational load during map construction is reduced, and noise interference in the multimodal data is reduced. This increases the accuracy of the recognition module in identifying slag in the fused map, while also reducing the computational load on the recognition module and improving the slag removal efficiency of the multimodal perception system.

[0016] The path planning module constructs a geometric model of the slag hopper, and combines the slag hopper geometric model, electrolytic cell characteristics, and slag characteristics to generate a slag removal path. The execution module then controls the movement of the overhead crane and slag hopper based on the slag removal path to perform the slag removal operation. This ensures that the slag hopper can complete the slag removal operation of the electrolytic cell with the minimum movement distance and / or the minimum number of slag removals, thereby achieving fully automated slag removal of the electrolytic cell and further improving slag removal efficiency.

[0017] Other features and advantages of the embodiments of the present invention will be described in the following detailed embodiments section. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of this application or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a module framework diagram of the multimodal sensing slag removal system provided in the embodiments of this application;

[0020] Figure 2 This is a flowchart of the multimodal sensing slag removal method provided in the embodiments of this application;

[0021] Figure 3 The flowchart of step S2 in the multimodal sensing slag removal method provided in the embodiments of this application is shown.

[0022] Explanation of reference numerals in the attached figures:

[0023] 1. Multimodal sensing and slag removal system; 11. Data acquisition module; 12. Mapping module; 13. Identification module; 14. Path planning module; 15. Execution module. Detailed Implementation

[0024] Unless otherwise specified, the terms “second direction,” “first direction,” “third direction,” “inner,” and “outer” used in the following descriptions, indicating orientation or positional relationships, are understood to be based on the orientation or positional relationships shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0025] Furthermore, features specified with "first" or "second" for descriptive purposes only should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Features specified with "first" or "second" may explicitly or implicitly include at least one of the specified features. The description of "multiple" generally means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0026] In this application, unless otherwise explicitly specified and limited, terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can be a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0027] In the description of this specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the 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 one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0028] Reference Figure 1To address the problems in the prior art, this invention provides a multimodal sensing slag removal system 1 for an electrolytic aluminum workshop. The multimodal sensing slag removal system 1 includes: a data acquisition module 11, a mapping module 12, an identification module 13, a path planning module 14, and an execution module 15. The data acquisition module 11 is configured to acquire multimodal data within the electrolytic aluminum workshop; the mapping module 12 is configured to perform fusion processing on the multimodal data and construct a fused map based on the multimodal data; the identification module 13 is configured to... The data is matched from the fused map to the electrolytic cell features. The slag features are obtained by fitting the height of the aluminum liquid plane and calculating the distance between all points within the electrolytic cell features and the aluminum liquid plane. The path planning module 14 is configured to: construct a slag hopper geometric model based on the slag hopper size, and generate a slag removal path by combining the slag hopper geometric model, electrolytic cell features, and slag features. The execution module 15 is configured to: control the crane and slag hopper to move based on the slag removal path to perform the slag removal operation. The multimodal data includes point cloud data, thermal data, and texture data.

[0029] By collecting multimodal data consisting of point cloud data, thermal data, and texture data, and fusing the multimodal data, a fused map with geometric, temperature, and texture attributes is obtained. This allows the recognition module 13 to sequentially identify the electrolytic cell, the aluminum liquid plane, and the slag on the fused map. Before constructing the fused map, the multimodal data is filtered and denoised to reduce the computational load during map construction and to reduce noise interference in the multimodal data. This increases the accuracy of the recognition module 13 in identifying slag in the fused map, while also reducing the computational load and improving the slag removal efficiency of the multimodal sensing system.

[0030] During slag removal, the path planning module 14 first constructs a slag hopper geometric model based on the slag hopper size. Then, it generates a slag removal path by combining the slag hopper geometric model, electrolytic cell characteristics, and slag characteristics. That is, when removing slag from electrolytic cells of different specifications, slag hoppers of different sizes can be used. Therefore, it is necessary to construct a slag hopper geometric model that matches the size of the electrolytic cell based on its location and size characteristics. Subsequently, a slag removal path is generated based on the slag hopper geometric model, electrolytic cell characteristics, and identified slag characteristics. The execution module 15 then controls the crane and slag hopper to move based on the slag removal path to perform the slag removal operation. This ensures that the slag hopper can complete the slag removal operation of the electrolytic cell with the least amount of movement distance and / or the fewest slag removal attempts, thereby achieving fully automated slag removal from the electrolytic cell and further improving slag removal efficiency.

[0031] In an embodiment of the present invention, the mapping module includes a correction unit, a filtering unit, a denoising unit, and a fusion unit. The correction unit is configured to: use the coordinates of the point cloud data as a reference and use preset compensation parameters to correct the coordinates of the thermal data and texture data, so that the point cloud data, thermal data, and texture data achieve coordinate alignment. The filtering unit is configured to: filter the point cloud data, thermal data, and texture data respectively to reduce the computational load when constructing the fused map. The denoising unit is configured to: use a statistical method to calculate the average distance from each point cloud to its k nearest neighbors, and remove point clouds with a distance exceeding μ ± α·σ, so as to remove interference in the point cloud data, where μ is the mean, σ is the standard deviation, and α is the threshold coefficient. The fusion unit is configured to: use a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then map the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct the fused map.

[0032] Before constructing the fused map, the coordinates of the multimodal data are first corrected by a correction unit. That is, using the coordinates of the point cloud data as a reference, coordinate compensation is set according to the positions of various hardware components in the acquisition module, and the compensation parameters are pre-stored in the mapping module. This allows the correction unit to perform coordinate correction on the thermal and texture data according to the preset compensation parameters, aligning the coordinates of the point cloud, thermal, and texture data. Subsequently, a filtering unit filters the point cloud, thermal, and texture data respectively to remove non-target and interference items from the multimodal data, reducing the computational load on the mapping module. Finally, a denoising unit is used to further refine the mapping process. Interference items in the point cloud data are further removed to further reduce the computational load of the mapping module. Finally, the fusion unit is used to map the coordinates of all the filtered and denoised point cloud data to the global coordinate system. At the same time, the coordinates of the aligned thermal data and texture data are also mapped to the global coordinate system, thereby constructing a fused map containing geometric, temperature, and texture attributes. This ensures that the features at any coordinate in the fused map correspond to the actual parts in the workshop, and that the geometric, temperature, and texture attributes are consistent with the actual situation, making it easy for the identification module to quickly and accurately identify slag from the fused map.

[0033] In an embodiment of the present invention, the filtering unit includes voxel filtering, thermal filtering, and texture filtering. The voxel filtering unit is configured to: divide the point cloud data into several voxels according to a preset size, and retain only one representative point cloud (such as the center point of the voxel or the point with the highest reflection intensity) within each voxel. The thermal filtering unit is configured to: divide the color blocks in the thermal data into several thermal fields, compare each thermal field based on a preset temperature fluctuation value, and filter out thermal fields with temperature fluctuations less than the preset temperature fluctuation value. That is, calculate the temperature change amplitude of each thermal field. If the temperature fluctuation is less than a preset threshold, it is marked as a "thermal stable zone". A thermal field marked as a "thermal stable zone" can be represented by two results: the first result is that the temperature difference between the slag temperature and the aluminum liquid temperature in the corresponding thermal field is small (less than the preset threshold); the second result is that the corresponding thermal field is full of aluminum liquid and there is no slag. Therefore, for areas marked as "thermal stable areas", a secondary inspection using a texture filtering unit is required to avoid slag being mistakenly filtered out. During the inspection, texture data from the "thermal stable areas" is compared based on pre-stored slag features to detect whether slag exists in the "thermal stable areas". If no slag is detected in the "thermal stable areas", the "thermal stable areas" are completely filtered out and not considered in the constructed fusion map. If slag is detected in the "thermal stable areas", the "thermal stable areas" are restored and added to the constructed fusion map.

[0034] In an embodiment of the present invention, the identification module includes an electrolytic cell identification unit, a point cloud segmentation unit, an aluminum liquid fitting unit, and a slag extraction unit. The electrolytic cell identification unit is configured to: identify the electrolytic cell from the fused map based on pre-stored electrolytic cell texture features; the point cloud segmentation module is configured to: identify the electrolytic cell based on the electrolytic cell boundary and the known aluminum liquid height range [h]. min ,h max Extracting a subset p of the point cloud inside the electrolytic cell C The aluminum liquid fitting unit is configured to: from p C Three random sampling points were taken and the equation parameters of the aluminum liquid plane were calculated using the following formula. d = -n·p1, where n is the unit normal vector of the molten aluminum plane, d is the constant term offset of the molten aluminum plane, and p1, p2, and p3 are the values ​​from p1 to p1 respectively. c Three random sampling points were selected from the sample; the slag extraction unit was configured to calculate a subset p of the point cloud. C The distance D from all points to the molten aluminum surface i =|n·p i +d|, if D i If the value is ≤ε, it is marked as an interior point. The final set of interior points is the aluminum melt point set, and the rest is the point set M containing slag. c By using the point set Mc Perform Euclidean clustering to obtain the slag cluster set {C} i}, where ε is a preset distance threshold.

[0035] First, the electrolytic cell identification unit identifies the electrolytic cell from the fused map based on the pre-stored electrolytic cell texture features, so that the point cloud segmentation unit can identify the electrolytic cell based on the boundary of the electrolytic cell and the known range of aluminum liquid height [h]. min ,h max Extract the point cloud subset p inside the electrolytic cell. C ; and the point cloud subset p C Unexpected point cloud data is filtered out to reduce the computational load during slag identification, and then the aluminum liquid fitting unit is used to extract data from the point cloud subset p. C Three non-collinear sample points are randomly selected from the sample points, and the formula for calculating the normal vector is used. Calculate the normal vector of the molten aluminum plane, then calculate the position of the molten aluminum plane in the fused map using the plane offset formula d = -n·p1, and finally extract the molten aluminum plane from the point cloud subset p using the slag extraction unit. C Calculate the distance D from all points to the molten aluminum plane. i That is, the point cloud subset p C The distance D between all points protruding on the molten aluminum surface i If D i If the value is ≤ε, it is marked as an interior point (i.e., a point belonging to the aluminum liquid plane). The final set of interior points is the aluminum liquid plane point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i This enables accurate identification of slag within the electrolytic cell, facilitating the path planning module's use of the slag cluster set {C}. i Plan the slag removal path to ensure that the amount of slag removed is maximized in each slag removal operation, thereby improving slag removal efficiency.

[0036] In an embodiment of the present invention, the acquisition module includes a lidar, a thermal imager, and a camera mounted on an overhead crane and moving synchronously with the crane. The lidar is configured to acquire point cloud data within the electrolytic aluminum workshop; the thermal imager is configured to acquire temperature field distribution data within the electrolytic aluminum workshop; and the camera is configured to acquire texture image data within the electrolytic aluminum workshop.

[0037] By mounting lidar, thermal imager, and camera on an overhead crane and moving them synchronously with the crane, the acquisition module collects multi-module data from inside the electrolytic cell. The complementary effect of multi-modal data improves the accuracy of slag identification, reduces the computational load of slag identification, and avoids the problems of low detection accuracy and low detection efficiency caused by using a single sensor.

[0038] In an embodiment of the present invention, the multimodal sensing slag removal system further includes an obstacle detection module, which is configured to: generate an obstacle detection ROI (Region of Interest) on the slag removal path, extract all point clouds of the obstacle detection ROI in the fused map, and fine-tune the path for possible collision locations to avoid collisions with static obstacles when the overhead crane moves the slag bucket on the slag removal path.

[0039] In an embodiment of the present invention, the obstacle detection module is further configured to: control the lidar to detect in real time whether new dynamic or unknown obstacles appear on the path ahead during the actual movement of the slag hopper along the slag removal path, and trigger the overhead crane to stop moving and issue obstacle warning information when an obstacle is detected, thereby avoiding collision between dynamic obstacles appearing on the slag removal path and the slag hopper, and improving the slag removal safety of the multimodal perception system.

[0040] Reference Figure 2 To address the problems in the prior art, this invention also provides a multimodal sensing slag removal method for use in an electrolytic aluminum workshop, applied to the aforementioned multimodal sensing slag removal system. The modal sensing slag removal method includes:

[0041] Step S1: The acquisition module collects multimodal data within the electrolytic aluminum workshop;

[0042] Step S2: The mapping module performs fusion processing on the multimodal data and constructs a fused map based on the multimodal data;

[0043] Step S3: The identification module matches the electrolytic cell features from the fused map based on the pre-stored feature data, and obtains the slag features by fitting the height of the aluminum liquid plane and calculating the distance between all points in the electrolytic cell features and the aluminum liquid plane.

[0044] Step S4: The path planning module constructs a slag hopper geometric model based on the slag hopper size, and generates a slag retrieval path by combining the slag hopper geometric model, electrolytic cell features, and slag features;

[0045] Step S5: The execution module controls the movement of the overhead crane and slag bucket based on the slag removal path to perform the slag removal operation;

[0046] The multimodal data includes point cloud data, thermal data, and texture data.

[0047] By collecting multimodal data consisting of point cloud data, thermal data, and texture data, and fusing the multimodal data, a fused map with geometric, temperature, and texture attributes is obtained. This allows the recognition module to sequentially identify the electrolytic cell, the molten aluminum plane, and the slag on the fused map. Before constructing the fused map, the multimodal data is filtered and denoised to reduce the computational load during map construction and to reduce noise interference in the multimodal data. This increases the accuracy of the recognition module in identifying slag in the fused map, while also reducing the computational load and improving the slag removal efficiency of the multimodal sensing system.

[0048] During slag removal, the path planning module first constructs a geometric model of the slag hopper based on its dimensions. This model, combined with the electrolytic cell characteristics and slag features, generates a slag removal path. This means that different slag hopper sizes can be used to remove slag from electrolytic cells of different specifications. Therefore, a slag hopper geometric model adapted to the location and size characteristics of the electrolytic cell needs to be constructed. Subsequently, a slag removal path is generated based on the slag hopper geometric model, electrolytic cell features, and identified slag features. The execution module then controls the crane and slag hopper to move based on the slag removal path to perform the slag removal operation. This ensures that the slag hopper can complete the slag removal operation on the electrolytic cell using the minimum travel distance and / or the fewest slag removal attempts, thereby achieving fully automated slag removal from the electrolytic cell and further improving slag removal efficiency.

[0049] Reference Figure 3 In an embodiment of the present invention, the step of fusing multimodal data and constructing a multimodal-based fused map further includes:

[0050] Step S21: The correction unit uses the coordinates of the point cloud data as a reference and uses preset compensation parameters to perform coordinate correction on the thermal data and texture data so that the point cloud data, thermal data and texture data achieve coordinate alignment.

[0051] Step S22: The filtering unit filters the point cloud data, thermal data and texture data respectively to reduce the amount of computation when constructing the fused map;

[0052] Step S23: The denoising unit uses a statistical method to calculate the average distance from each point cloud to its k nearest neighbors, and removes point clouds with distances exceeding μ±α·σ to eliminate interference in the point cloud data.

[0053] Where μ is the mean, σ is the standard deviation, and α is the threshold coefficient;

[0054] Step S24: The fusion unit uses a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then maps the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct a fused map.

[0055] Before constructing the fused map, the coordinates of the multimodal data are first corrected by a correction unit. That is, using the coordinates of the point cloud data as a reference, coordinate compensation is set according to the positions of various hardware components in the acquisition module, and the compensation parameters are pre-stored in the mapping module. This allows the correction unit to perform coordinate correction on the thermal and texture data according to the preset compensation parameters, aligning the coordinates of the point cloud, thermal, and texture data. Subsequently, a filtering unit filters the point cloud, thermal, and texture data respectively to remove non-target and interference items from the multimodal data, reducing the computational load on the mapping module. Finally, a denoising unit is used to further refine the mapping process. Interference items in the point cloud data are further removed to further reduce the computational load of the mapping module. Finally, the fusion unit is used to map the coordinates of all the filtered and denoised point cloud data to the global coordinate system. At the same time, the coordinates of the aligned thermal data and texture data are also mapped to the global coordinate system, thereby constructing a fused map containing geometric, temperature, and texture attributes. This ensures that the features at any coordinate in the fused map correspond to the actual parts in the workshop, and that the geometric, temperature, and texture attributes are consistent with the actual situation, making it easy for the identification module to quickly and accurately identify slag from the fused map.

[0056] In an embodiment of the present invention, the step of the identification module matching electrolytic cell features from the fused map based on pre-stored feature data, fitting the height of the molten aluminum plane, and calculating the distance between all points within the electrolytic cell features and the molten aluminum plane to obtain slag features further includes:

[0057] Step S31: The electrolytic cell identification module identifies the electrolytic cell from the fused map based on the pre-stored electrolytic cell texture features;

[0058] Step S32: The point cloud segmentation module segments the data based on the electrolytic cell boundary and the known range of aluminum liquid height [h]. min ,h max Extracting a subset p of the point cloud inside the electrolytic cell C ;

[0059] Step S33: The aluminum liquid fitting unit starts from p C Three random sampling points were taken and the equation parameters of the aluminum liquid plane were calculated using the following formula.

[0060]

[0061] Where n is the unit normal vector of the molten aluminum plane, d is the constant term offset of the molten aluminum plane, and p1, p2, and p3 are the values ​​from the point cloud subset p c Three sample points were randomly selected from the sample.

[0062] Step S34: Calculate the point cloud subset p of the slag extraction unit C The distance of all points from the aluminum liquid surface

[0063] D i =|n·p i +d|,

[0064] If D i If the value is ≤ε, it is marked as an interior point. The final set of interior points is the aluminum melt point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i},

[0065] Where ε is a preset distance threshold.

[0066] First, the electrolytic cell identification unit identifies the electrolytic cell from the fused map based on the pre-stored electrolytic cell texture features, so that the point cloud segmentation unit can identify the electrolytic cell based on the boundary of the electrolytic cell and the known range of aluminum liquid height [h]. min ,h max Extract the point cloud subset p inside the electrolytic cell. C ; and the point cloud subset p C Unexpected point cloud data is filtered out to reduce the computational load during slag identification, and then the aluminum liquid fitting unit is used to extract data from the point cloud subset p. C Three non-collinear sample points are randomly selected from the sample points, and the formula for calculating the normal vector is used. Calculate the normal vector of the molten aluminum plane, then calculate the position of the molten aluminum plane in the fused map using the plane offset formula d = -n·p1, and finally extract the molten aluminum plane from the point cloud subset p using the slag extraction unit. C Calculate the distance D from all points to the molten aluminum plane. i That is, the point cloud subset p C The distance D between all points protruding on the molten aluminum surface i If D i If the value is ≤ε, it is marked as an interior point (i.e., a point belonging to the aluminum liquid plane). The final set of interior points is the aluminum liquid plane point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i This enables accurate identification of slag within the electrolytic cell, facilitating the path planning module's use of the slag cluster set {C}. iPlan the slag removal path to ensure that the amount of slag removed is maximized in each slag removal operation, thereby improving slag removal efficiency.

[0067] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still adjust the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these adjustments or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A multimodal sensing slag removal system for an electrolytic aluminum workshop, characterized in that, The multimodal sensing and slag removal system includes: a data acquisition module, a mapping module, a recognition module, a path planning module, and an execution module. The acquisition module is configured to acquire multimodal data within the electrolytic aluminum workshop; The mapping module is configured to: perform fusion processing on the multimodal data and construct a fused map based on the multimodal data; The identification module is configured to: match the electrolytic cell features from the fused map based on pre-stored feature data, and obtain slag features by fitting the height of the aluminum liquid plane and calculating the distance between all points within the electrolytic cell features and the aluminum liquid plane; The path planning module is configured to: construct a slag hopper geometric model based on the slag hopper size, and generate a slag retrieval path by combining the slag hopper geometric model, electrolytic cell features, and slag features; The execution module is configured to control the movement of the overhead crane and slag bucket based on the slag removal path to perform slag removal operations; The multimodal data includes point cloud data, thermal data, and texture data.

2. The multimodal sensing slag removal system according to claim 1, characterized in that, The mapping module includes a correction unit, a filtering unit, a noise reduction unit, and a fusion unit. The correction unit is configured to: use the coordinates of the point cloud data as a reference and use preset compensation parameters to perform coordinate correction on the thermal data and texture data, so that the point cloud data, thermal data and texture data achieve coordinate alignment; The filtering unit is configured to filter the point cloud data, thermal data and texture data respectively to reduce the amount of computation when constructing the fused map; The denoising unit is configured to: calculate the average distance from each point cloud to its k nearest neighbors using a statistical method, and remove point clouds whose distance exceeds μ±α·σ, in order to eliminate interference in the point cloud data. Where μ is the mean, σ is the standard deviation, and α is the threshold coefficient; The fusion unit is configured to: use a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then map the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct a fused map.

3. The multimodal sensing slag removal system according to claim 2, characterized in that, The filtering unit includes voxel filtering, thermal filtering, and texture filtering. The voxel filtering unit is configured to divide the point cloud data into several voxels according to a preset size, and retain only one representative point cloud in each voxel. The thermal filtering unit is configured to: divide the color blocks in the thermal data into several thermal fields, compare them in each thermal field based on a preset temperature fluctuation value, and filter out thermal fields with temperature fluctuations smaller than the preset temperature fluctuation value. The texture filtering unit is configured to perform texture comparison on the texture data from the filtered thermal field based on pre-stored slag features, so as to avoid slag with aluminum liquid temperature fluctuations less than a preset temperature fluctuation value being mistakenly filtered out.

4. The multimodal sensing slag removal system according to claim 1, characterized in that, The identification module includes an electrolytic cell identification unit, a point cloud segmentation unit, an aluminum liquid fitting unit, and a slag extraction unit. The electrolytic cell identification unit is configured to identify electrolytic cells from the fused map based on pre-stored electrolytic cell texture features; The point cloud segmentation module is configured to: segment the point cloud based on the electrolytic cell boundary and the known aluminum liquid height range [h] min ,h max Extracting a subset p of the point cloud inside the electrolytic cell C ; The aluminum liquid fitting unit is configured to: from p C Three random sampling points were taken and the equation parameters of the aluminum liquid plane were calculated using the following formula. Where n is the unit normal vector of the molten aluminum plane, d is the constant term offset of the molten aluminum plane, and p1, p2, and p3 are the values ​​from the point cloud subset p c Three sample points were randomly selected from the middle. The slag extraction unit is configured to: compute a subset p of the point cloud. C The distance of all points from the aluminum liquid surface D i =|n·p i +d|, If D i If the value is ≤ε, it is marked as an interior point. The final set of interior points is the aluminum liquid plane point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i }, Where ε is a preset distance threshold.

5. The multimodal sensing slag removal system according to claim 1, characterized in that, The acquisition module includes a lidar, a thermal imager, and a camera, all mounted on the overhead crane and moving synchronously with it. The lidar is configured to collect point cloud data within the electrolytic aluminum workshop. The thermal imager is configured to collect temperature field distribution data within the electrolytic aluminum workshop. The camera is configured to acquire texture image data within the aluminum electrolysis workshop.

6. The multimodal sensing slag removal system according to claim 5, characterized in that, The multimodal perception slag removal system also includes an obstacle detection module, which is configured to: generate an obstacle detection ROI area on the slag removal path, extract all point clouds of the obstacle detection ROI area in the fused map, and fine-tune the path for possible collision locations.

7. The multimodal sensing slag removal system according to claim 6, characterized in that, The obstacle detection module is also configured to: control the lidar to detect in real time whether new dynamic or unknown obstacles appear on the path ahead during the actual movement of the slag hopper along the slag removal path, and trigger the crane to stop moving and issue obstacle warning information when an obstacle is detected.

8. A multimodal sensing slag removal method for use in an electrolytic aluminum workshop, applied to the multimodal sensing slag removal system as described in any one of claims 1-7, characterized in that, The modality sensing slag removal method includes: The data acquisition module collects multimodal data within the aluminum electrolysis workshop; The mapping module performs fusion processing on multimodal data and constructs a fused map based on multimodal data; The identification module matches the electrolytic cell features from the fused map based on the pre-stored feature data, and obtains the slag features by fitting the height of the aluminum liquid plane and calculating the distance between all points in the electrolytic cell features and the aluminum liquid plane. The path planning module constructs a geometric model of the slag hopper based on its dimensions, and generates a slag retrieval path by combining the slag hopper geometric model, electrolytic cell features, and slag features. The execution module controls the movement of the overhead crane and slag hopper based on the slag removal path to perform slag removal operations; The multimodal data includes point cloud data, thermal data, and texture data.

9. The multimodal sensing slag removal method according to claim 8, characterized in that, The steps of fusing multimodal data and constructing a multimodal-based fused map further include: The correction unit uses the coordinates of the point cloud data as a reference and uses preset compensation parameters to perform coordinate correction on the thermal data and texture data so that the point cloud data, thermal data and texture data achieve coordinate alignment. The filtering unit filters the point cloud data, thermal data, and texture data respectively to reduce the computational load when constructing the fused map; The denoising unit uses statistical methods to calculate the average distance from each point cloud to its k nearest neighbors, and removes point clouds with distances exceeding μ±α·σ to eliminate interference in the point cloud data. Where μ is the mean, σ is the standard deviation, and α is the threshold coefficient; The fusion unit uses a transformation matrix to map the coordinates of the filtered and denoised point cloud data to the global coordinate system, and then maps the coordinates of the coordinate-aligned thermal data and texture data to the global coordinate system to construct a fused map.

10. The multimodal sensing slag removal method according to claim 8, characterized in that, The step of the identification module matching electrolytic cell features from the fused map based on pre-stored feature data, fitting the height of the molten aluminum plane, and calculating the distance between all points within the electrolytic cell features and the molten aluminum plane to obtain slag features further includes: Electrolytic cell identification is based on pre-stored electrolytic cell texture features to identify electrolytic cells from the fused map; The point cloud segmentation module is based on the electrolytic cell boundary and the known range of aluminum liquid height [h] min ,h max Extracting a subset p of the point cloud inside the electrolytic cell C ; Aluminum liquid fitting unit from p C Three random sampling points were taken and the equation parameters of the aluminum liquid plane were calculated using the following formula. Where n is the unit normal vector of the molten aluminum plane, d is the constant term offset of the molten aluminum plane, and p1, p2, and p3 are the values ​​from p1 to p2 respectively. c Three sample points were randomly selected from the sample. Slag extraction unit calculation point cloud subset p C The distance of all points from the aluminum liquid surface D i =|n·p i +d|, If D i If the value is ≤ε, it is marked as an interior point. The final set of interior points is the aluminum melt point set, and the rest is the point set M containing slag. c By using the point set M c Perform Euclidean clustering to obtain the slag cluster set {C} i }, Where ε is a preset distance threshold.

Images