A dry ice cleaning method and apparatus

Abstract

The application discloses a dry ice cleaning method and device. The method comprises the following steps: identifying stains on the building surface image of a building to be cleaned; and performing dry ice cleaning based on the stain information after the stains are identified. Since the dry ice cleaning does not need to use a water source and does not need to contact the building surface by using mechanical tools, the stains on the building surface can be quickly removed, secondary pollution is avoided, and the building surface is not damaged, so that the cleaning effect and the integrity of the building surface can be simultaneously considered. In addition, a closed-loop grading recovery mode is adopted to recover the mixed gas in the dry ice spraying area, so that the resource consumption is reduced, and the excessive discharge of carbon dioxide is avoided.

Description

Technical Field

[0001] This invention relates to the field of building cleaning technology, and in particular to a dry ice cleaning method and apparatus. Background Technology

[0002] Building maintenance is crucial for extending the lifespan of buildings, preserving their aesthetics, and ensuring their functionality. Existing building cleaning methods typically rely on high-pressure water jets, chemical cleaning agents, and mechanical tools. These methods are time-consuming, water-intensive, cause environmental pollution, and can easily damage building surfaces. For some special buildings, such as historical buildings, which possess unique cultural and artistic value, it is necessary to avoid damaging their surfaces. However, existing building cleaning methods struggle to simultaneously achieve both effective cleaning and maintain the integrity of the building's surface. Summary of the Invention

[0003] This invention provides a dry ice cleaning method and apparatus to solve the technical problem that existing building cleaning methods cannot simultaneously achieve both cleaning effectiveness and the integrity of building surfaces.

[0004] To address the aforementioned technical problems, a first aspect of this invention provides a dry ice cleaning method, comprising:

[0005] Acquire images of the building surface to be cleaned, and identify stains in the building surface images;

[0006] When stains are detected in the building surface image, stain information is obtained;

[0007] Based on the stain information, determine the current dry ice spray parameters;

[0008] The stain is subjected to dry ice blasting based on the dry ice blasting parameters.

[0009] As a preferred embodiment, the step of acquiring an image of the building surface to be cleaned and identifying stains on the building surface image specifically includes:

[0010] Acquire a multispectral image of the building surface to be cleaned, and use the multispectral image as the building surface image;

[0011] The stain recognition model is used to identify stains in the multispectral image to obtain stain recognition results; wherein, the stain recognition model is obtained by training a multichannel convolutional neural network using multispectral training images with stain marking information;

[0012] When it is determined that any type of stain exists based on the stain identification result, it is determined that there is a stain in the building surface image.

[0013] As a preferred embodiment, the step of using a pre-trained stain recognition model to perform stain recognition on the multispectral image and obtain stain recognition results specifically includes:

[0014] Feature extraction is performed on the multispectral image to obtain multimodal stain features;

[0015] The CBAM module is used to enhance the multimodal stain features to obtain enhanced stain features;

[0016] A classifier is used to identify the enhanced stain features, and a stain probability heatmap is generated.

[0017] Based on the stain probability heatmap, determine the stain bounding box and the target stain type;

[0018] The FAST corner detection algorithm is used to extract a preset number of candidate points within the stain bounding box, and based on the gradient magnitude of each candidate point, several stain feature points are obtained.

[0019] The positions of several stain feature points are iteratively optimized based on the gray-level gradient method until the position change of the stain feature points is less than a preset pixel value. The target stain center position is determined based on the centroid of each stain feature point after iterative optimization.

[0020] The stain identification result is obtained based on the target stain type and the center position of the target stain.

[0021] As a preferred embodiment, obtaining the stain information specifically includes:

[0022] The surface of the building to be cleaned is scanned using a lidar sensor to construct a three-dimensional model of the building surface.

[0023] Based on the point cloud data in the three-dimensional building surface model and the positions of each stain feature point after iterative optimization, the stain area in the three-dimensional building surface model is determined.

[0024] Based on the point cloud data and the stained area, calculate the stained area and the building surface area of ​​the building to be cleaned, and calculate the maximum distance difference between the stained area and the building surface;

[0025] Based on the stain area and the building surface area, calculate the area ratio of the stain area, and based on the preset stain stubbornness levels corresponding to different area ratios and distance differences, determine the area ratio of the stain area and the target stain stubbornness level corresponding to the maximum distance difference.

[0026] The stain information of the stain is determined based on the type of the target stain, the center location of the target stain, and the stubbornness level of the target stain.

[0027] As a preferred embodiment, determining the current dry ice spray parameters based on the stain information specifically includes:

[0028] Based on the preset dry ice blasting parameters corresponding to different stain types and stain stubbornness levels, determine the current dry ice blasting pressure, dry ice particle density, and dry ice blasting time;

[0029] Determine the current dry ice spray angle based on the center position of the target stain;

[0030] The current dry ice injection parameters are determined based on the dry ice injection pressure, the dry ice particle density, the dry ice injection time, and the dry ice injection angle.

[0031] As a preferred embodiment, the dry ice blasting of the stain based on the dry ice blasting parameters specifically includes:

[0032] The dry ice spray nozzle is adjusted according to the dry ice spray angle so that it is aimed at the center of the target stain.

[0033] Based on the aforementioned stain probability heatmap, the stain density at different locations is determined;

[0034] Based on the preset cleaning area boundary information and motion constraint parameters, starting from the center position of the target stain, the path is planned in descending order of stain density to obtain the dry ice spray trajectory.

[0035] Dry ice is sprayed using the dry ice spray nozzle according to the dry ice spray pressure, the dry ice particle density, and the dry ice spray time, and the dry ice spray nozzle is moved according to the dry ice spray trajectory.

[0036] As a preferred embodiment, after dry ice blasting the stain based on the dry ice blasting parameters, the method further includes:

[0037] The mixed gas within the dry ice spray area is recovered; wherein the mixed gas includes dry ice particles, dirt, and carbon dioxide;

[0038] The mixed gas is separated once by a cyclone separator to obtain the dirt and the dry ice particles from the mixed gas, and the dirt and the dry ice particles are collected in a preset dirt storage tank and dry ice storage tank respectively.

[0039] The remaining mixed gas after the first separation process is subjected to a second separation process using an electrostatic precipitator to separate carbon dioxide from the remaining mixed gas. The carbon dioxide is then liquefied and collected in a pre-designated carbon dioxide storage tank, or the carbon dioxide is adsorbed and stored through a molecular sieve.

[0040] As a preferred embodiment, the method further includes:

[0041] When it is determined that there are no stains in the building surface image, the distance between the current position and the edge of the building surface to be cleaned is calculated;

[0042] When the distance between the two sides is greater than the first preset distance, the system moves along the normal of the building surface to be cleaned toward the edge of the building surface and calculates the distance between the two sides in real time. When the distance between the two sides is detected to remain unchanged within a preset time period, the system moves back a preset distance, switches the moving angle, and continues to move toward the edge of the building surface until the distance between the two sides is detected to decrease within the preset time period. Then, the system re-acquires the building surface image of the building to be cleaned and performs stain recognition on the building surface image.

[0043] When the distance between the two sides is less than or equal to the first preset distance and greater than or equal to the second preset distance, the system moves towards the edge of the building surface according to the preset approximation speed corresponding to the building surface material of the building to be cleaned, and updates the distance between the two sides at a frequency of 10Hz during the movement, performs multispectral scanning on the building surface to reacquire the building surface image of the building to be cleaned, and identifies stains on the building surface image.

[0044] When the distance between the two sides is less than the second preset distance, the system moves towards the edge of the building surface until the distance between the two sides reaches the preset minimum distance. A contact sensor array is used to detect changes in the pressure distribution on the building surface, and the standard deviation of the contact pressure on the building surface is calculated based on these changes. When the standard deviation of the contact pressure is greater than a preset pressure standard deviation threshold, a vibration signal is emitted to the building surface to identify whether there are stains on the building surface. If stains are present, the building surface image to be cleaned is re-acquired, and stain identification is performed on the building surface image. If stains are not present, dry ice is sprayed onto the building surface according to preset dry ice cleaning parameters.

[0045] As a preferred embodiment, the method further includes:

[0046] Real-time monitoring of the remaining dry ice level in the dry ice storage tank;

[0047] When the remaining dry ice is detected to be less than a preset threshold, an insufficient dry ice alarm is issued.

[0048] A second aspect of the present invention provides a dry ice cleaning device, including a housing, a moving module, a storage module, a recycling module, a telescopic arm module, a dry ice spraying module, and a control module;

[0049] The moving module is located on both sides of the outer casing; the storage module, the recycling module, and the telescopic arm module are all located on the outer casing.

[0050] The dry ice injection module is located at the end of the telescopic arm module. The dry ice injection module includes a high-pressure air generating module, a dry ice injection nozzle, and a vacuum adsorption module. The input end of the dry ice injection nozzle is connected to the gas output end of the high-pressure air generating module, and the input end of the dry ice injection nozzle is connected to the recovery module through a dry ice delivery pipe. The vacuum adsorption module is positioned close to the dry ice injection nozzle and is connected to the recovery module.

[0051] The storage module includes a dirt storage tank, a dry ice storage tank, and a carbon dioxide storage tank. The recycling module has a dirt separation port, a dry ice separation port, and a carbon dioxide separation port, which are respectively connected to the dirt storage tank, the dry ice storage tank, and the carbon dioxide storage tank.

[0052] The control module is located inside the housing, and the control terminal of the control module is connected to the controlled terminal of the moving module, the controlled terminal of the storage module, the controlled terminal of the recovery module, the controlled terminal of the telescopic arm module, and the controlled terminal of the dry ice spraying module, respectively; the control module is used to perform the dry ice cleaning method as described in any one of the first aspects.

[0053] Compared with the prior art, the beneficial effect of the embodiments of the present invention is that by identifying stains in the image of the building surface to be cleaned, and then performing dry ice cleaning based on the stain information after the stains are identified, since dry ice cleaning does not require the use of water or mechanical tools to contact the building surface, it can quickly remove stains from the building surface without causing secondary pollution or damaging the building surface, thus simultaneously ensuring both cleaning effect and the integrity of the building surface. Attached Figure Description

[0054] Figure 1 This is a schematic flowchart of the dry ice cleaning method in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of dry ice spraying parameters corresponding to different stain types in the embodiments of the present invention;

[0056] Figure 3 This is a schematic diagram of the mixed gas recovery process during dry ice cleaning in an embodiment of the present invention;

[0057] Figure 4 This is a flowchart of the dry ice cleaning method in an embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram of the dry ice cleaning device in an embodiment of the present invention;

[0059] Figure 6 This is an architectural diagram of the dry ice cleaning system in an embodiment of the present invention;

[0060] The components include: 1. Outer shell; 2. Moving module; 3. Dry ice delivery pipe; 4. Telescopic arm module; and 5. Dry ice spraying module. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Please see Figure 1 The first aspect of this invention provides a dry ice cleaning method, comprising the following steps S1 to S4:

[0063] Step S1: Obtain an image of the building surface to be cleaned, and identify stains on the building surface image;

[0064] Step S2: When stains are detected in the building surface image, obtain stain information of the stains;

[0065] Step S3: Determine the current dry ice spraying parameters based on the stain information;

[0066] Step S4: Based on the dry ice spraying parameters, dry ice is sprayed onto the stain.

[0067] It's worth noting that traditional high-pressure water cleaning, chemical cleaning, and mechanical cleaning methods all present several potential risks of damaging building exteriors. Specifically, high-pressure water generates significant impact when washing exterior walls. This impact can damage the surface structure, especially for older buildings made of fragile materials. For example, high-pressure water can cause cracks and peeling on the stone surface, and long-term use may loosen the grout lines, affecting the overall stability of the wall. Chemical cleaning agents typically contain various chemicals that can react with the building's materials. For instance, acidic cleaners can corrode the calcium carbonate in the stone, causing it to lose its luster and become rough; alkaline cleaners can damage the coating, leading to peeling and discoloration. Furthermore, if chemical cleaners are not thoroughly rinsed off, residual chemicals can continue to corrode the exterior, accelerating aging and damage. When using mechanical tools to clean building exteriors, the tools are in direct contact with the surface, easily causing friction. During the cleaning process, mechanical tools may scratch the exterior wall surface, and this damage is more pronounced on buildings with decorative carvings or coatings. For example, tools such as wire brushes may damage the details of the carvings, compromising the building's artistic value, and may also cause the exterior coating to peel off, reducing the protective performance of the exterior wall. Dry ice cleaning, on the other hand, does not require the use of water, avoiding secondary pollution and residue on the wall surface associated with traditional water-based cleaning methods. It also eliminates the need for mechanical tools to come into contact with the building surface, and is non-toxic and non-conductive, causing no environmental pollution and ensuring the integrity of the building surface.

[0068] Specifically, in this embodiment, to perform precise dry ice cleaning on the building, it is first necessary to acquire an image of the building surface and identify stains. When stains are detected in the building surface image, the corresponding stain information needs to be obtained, such as stain type, location, size, and stubbornness level. This allows for precise dry ice blasting based on the stain's location, and different dry ice blasting parameters are applied based on the stain's size and stubbornness level to effectively remove the stain from the building surface. Furthermore, after determining the stain information, dry ice blasting parameters need to be determined for the specific stain to be cleaned, such as dry ice blasting angle, dry ice blasting pressure, dry ice particle density, and dry ice blasting time. Dry ice blasting is then performed based on these parameters to achieve the corresponding stain removal effect.

[0069] The dry ice cleaning method provided in this invention identifies stains on the building surface image of the building to be cleaned, and performs dry ice cleaning based on the stain information after the stains are identified. Since dry ice cleaning does not require the use of water or mechanical tools to contact the building surface, it can quickly remove stains from the building surface without causing secondary pollution or damaging the building surface, thus simultaneously ensuring both cleaning effect and the integrity of the building surface.

[0070] Compared to traditional handheld dry ice cleaning machines, which require operators to wear appropriate protective equipment such as gloves and goggles to ensure safety and are more difficult to operate, this invention enables an intelligent and automated dry ice cleaning process, replacing manual labor in high-risk, repetitive, and arduous dry ice cleaning tasks.

[0071] As a preferred embodiment, the step of acquiring an image of the building surface to be cleaned and identifying stains on the building surface image specifically includes:

[0072] Acquire a multispectral image of the building surface to be cleaned, and use the multispectral image as the building surface image;

[0073] The stain recognition model is used to identify stains in the multispectral image to obtain stain recognition results; wherein, the stain recognition model is obtained by training a multichannel convolutional neural network using multispectral training images with stain marking information;

[0074] When it is determined that any type of stain exists based on the stain identification result, it is determined that there is a stain in the building surface image.

[0075] Specifically, this embodiment combines computer vision technology, sensor fusion, and intelligent algorithms to collect multimodal stain data through sensors. For example, a multispectral camera can be used to acquire multispectral images of the building surface to be cleaned. This means that different types of stains on the wall are identified using light at different wavelengths. Different stains exhibit different characteristics in multispectral images. For instance, oil stains have unique reflectance spectra at certain wavelengths, while mold has obvious absorption characteristics at other wavelengths. By analyzing multispectral images, various stains can be quickly distinguished. Furthermore, a pre-trained stain recognition model is used to identify stains in the multispectral images to obtain stain recognition results. In this embodiment, a multispectral training image set with stain marking information can be prepared in advance. This allows the location and type of stains in different multispectral training images to be marked, enabling the multi-channel convolutional neural network to learn the feature information of different types of stains. For example, stain types include oil stains (grease, industrial pollutants), biofilms (mold, algae, moss), dust (dust, PM2.5 deposits), paint residues (paint, graffiti), and mixed stains (i.e., mixtures of the above types of stains).

[0076] As a preferred embodiment, the step of using a pre-trained stain recognition model to perform stain recognition on the multispectral image and obtain stain recognition results specifically includes:

[0077] Feature extraction is performed on the multispectral image to obtain multimodal stain features;

[0078] The CBAM module is used to enhance the multimodal stain features to obtain enhanced stain features;

[0079] A classifier is used to identify the enhanced stain features, and a stain probability heatmap is generated.

[0080] Based on the stain probability heatmap, determine the stain bounding box and the target stain type;

[0081] The FAST corner detection algorithm is used to extract a preset number of candidate points within the stain bounding box, and based on the gradient magnitude of each candidate point, several stain feature points are obtained.

[0082] The positions of several stain feature points are iteratively optimized based on the gray-level gradient method until the position change of the stain feature points is less than a preset pixel value. The target stain center position is determined based on the centroid of each stain feature point after iterative optimization.

[0083] The stain identification result is obtained based on the target stain type and the center position of the target stain.

[0084] Specifically, in this embodiment, during the stain recognition process, feature extraction is first performed on the multispectral image to obtain multimodal stain features. For small stains, such as bird droppings, to achieve accurate stain recognition, this embodiment introduces an attention mechanism for feature enhancement. The CBAM (Convolutional Block Attention Module) enhances local features by introducing attention mechanisms in both channel and spatial dimensions. In the channel dimension, the CBAM module can automatically learn the importance of features in different channels, giving higher weights to channels containing features of small stains. In the spatial dimension, it can focus on the region where the small stain is located, enhancing the feature representation of that region. For example, when bird droppings are detected, the CBAM module highlights the local region where the droppings are located, making the network pay more attention to the small target, thereby improving the ability to recognize and classify small stains. Furthermore, the classifier calculates the probability of each pixel belonging to each stain type based on the enhanced stain features in different bands of the image, thereby generating a stain probability heatmap.

[0085] Furthermore, a threshold segmentation algorithm (such as the Otsu thresholding method) is applied to the generated stain probability heatmap to convert it into a binary image, thereby distinguishing areas where stains may exist from background areas. In the binary image, a contour detection algorithm (such as OpenCV's findcontours function) is used to find the contours of the stain areas. Then, based on these contours, the smallest rectangular bounding box that can enclose the stain areas is calculated; the coordinates of this bounding box are the initial values ​​of the stain bounding box. Based on the calculated initial bounding box values, the dynamic region of interest (ROI) in the image is determined, which is the stain bounding box in this embodiment. Subsequent processing operations will mainly be performed within this stain bounding box, thus reducing computational load and improving processing efficiency.

[0086] Furthermore, the FAST corner detection algorithm is used to extract a predetermined number of candidate points within the stain bounding box, and several stain feature points are obtained based on the gradient magnitude of each candidate point. For example, 50 to 100 candidate points are extracted, and the top 20% of high-confidence points with the highest gradient magnitude are retained as stain feature points.

[0087] Furthermore, the positions of several stain feature points are iteratively optimized based on the gray-level gradient method until the position change of the stain feature points is less than a preset pixel value, for example, until the position change of the stain feature points is <0.01 pixels, so that the centroid iteratively approaches, thereby determining the center position of the stain within the stain bounding box.

[0088] As a preferred embodiment, obtaining the stain information specifically includes:

[0089] The surface of the building to be cleaned is scanned using a lidar sensor to construct a three-dimensional model of the building surface.

[0090] Based on the point cloud data in the three-dimensional building surface model and the positions of each stain feature point after iterative optimization, the stain area in the three-dimensional building surface model is determined.

[0091] Based on the point cloud data and the stained area, calculate the stained area and the building surface area of ​​the building to be cleaned, and calculate the maximum distance difference between the stained area and the building surface;

[0092] Based on the stain area and the building surface area, calculate the area ratio of the stain area, and based on the preset stain stubbornness levels corresponding to different area ratios and distance differences, determine the area ratio of the stain area and the target stain stubbornness level corresponding to the maximum distance difference.

[0093] The stain information of the stain is determined based on the type of the target stain, the center location of the target stain, and the stubbornness level of the target stain.

[0094] Specifically, a 3D LiDAR is used to comprehensively scan the building's walls. The LiDAR emits laser beams towards the wall and measures the time it takes for the reflected light to return, calculating the distance between each point and the LiDAR. This yields a large amount of 3D coordinate information for a large number of points on the wall surface, forming point cloud data. This point cloud data records the geometry and spatial location of the wall, enabling the construction of a 3D building surface model. Since the aforementioned steps have confirmed the location of each stain feature point, which is equivalent to a stain area, and considering that the coordinate system of the 3D building surface model is different from that of the 2D image, the stain area location in the 2D image can be converted into a point cloud location in the 3D building surface model. The point cloud location can be directly obtained based on the point cloud data in the 3D building surface model, thus determining the stain area in the 3D building surface model.

[0095] Furthermore, since the location of each point cloud can be directly obtained based on the point cloud data, the stain area and the building surface area of ​​the building to be cleaned can be directly calculated. By calculating the maximum distance difference between the stain area and the building surface, the depth value of the stain area can be determined, thereby reflecting the stubbornness of the stain to a certain extent.

[0096] Furthermore, since different stain area proportions and distance differences between the stain area and the building surface indicate different stain stubbornness levels, this embodiment can preset different stain stubbornness levels corresponding to different area proportions and distance differences, thereby directly determining the target stain stubbornness level corresponding to the current stain area's area proportion and distance difference, and using the target stain type, target stain center location, and target stain stubbornness level together as the stain information.

[0097] Furthermore, this embodiment can also consider other quantification metrics, such as color contrast, which calculates the ΔE value between the stained area and the building surface in the LAB color space. The ΔE value reflects the degree of difference between the two colors; the larger the ΔE value, the higher the color contrast. Texture complexity, which calculates parameters such as energy and entropy based on the gray-level co-occurrence matrix. Energy reflects the uniformity of the image's gray-level distribution, and entropy represents the randomness of the image. These parameters can quantify the texture complexity of the stained area.

[0098] As a preferred embodiment, determining the current dry ice spray parameters based on the stain information specifically includes:

[0099] Based on the preset dry ice blasting parameters corresponding to different stain types and stain stubbornness levels, determine the current dry ice blasting pressure, dry ice particle density, and dry ice blasting time;

[0100] Determine the current dry ice spray angle based on the center position of the target stain;

[0101] The current dry ice injection parameters are determined based on the dry ice injection pressure, the dry ice particle density, the dry ice injection time, and the dry ice injection angle.

[0102] Specifically, different stain types and stain stubbornness levels result in different stain characteristics, which in turn determine the different dry ice blasting parameters needed to achieve the desired cleaning effect. Therefore, this embodiment pre-sets the dry ice blasting parameters corresponding to different stain types and stain stubbornness levels. Taking stain type as an example, the dry ice blasting pressure and dry ice blasting density corresponding to different stain types can be set, such as... Figure 2 As shown.

[0103] Furthermore, the center position of the identified stain in the image has been determined, typically represented by pixel coordinates, such as the image center as the origin, or a specific point as a reference. This position in the two-dimensional image is then converted to a position in three-dimensional space. Combined with the current position and orientation of the dry cleaning unit, the angle that the dry ice spray nozzle needs to be adjusted is calculated. Specifically, first, the pixel coordinates in the image need to be converted to three-dimensional coordinates in the camera coordinate system. This involves the camera's intrinsic parameters (such as focal length and principal point coordinates) and extrinsic parameters (the camera's position and orientation relative to the dry cleaning unit's base). These parameters can be obtained through camera calibration. Then, the three-dimensional points in the camera coordinate system need to be transformed to the dry cleaning unit's base coordinate system. This requires knowing the camera's position and orientation relative to the dry cleaning unit's base, i.e., the extrinsic parameter matrix. Next, the points in the dry cleaning unit's base coordinate system need to be transformed to the dry ice spray nozzle's coordinate system. The position and orientation of the dry ice spray nozzle relative to the dry cleaning unit's base are known, or can be calculated using the kinematic model of the dry cleaning unit's base. Once the position of the center point of the target stain in the coordinate system of the dry ice spray nozzle is determined, the angle that the dry ice spray nozzle needs to be adjusted can be calculated through inverse kinematics so that the spray direction is aligned with the center of the target stain.

[0104] As a preferred embodiment, the dry ice blasting of the stain based on the dry ice blasting parameters specifically includes:

[0105] The dry ice spray nozzle is adjusted according to the dry ice spray angle so that it is aimed at the center of the target stain.

[0106] Based on the aforementioned stain probability heatmap, the stain density at different locations is determined;

[0107] Based on the preset cleaning area boundary information and motion constraint parameters, starting from the center position of the target stain, the path is planned in descending order of stain density to obtain the dry ice spray trajectory.

[0108] Dry ice is sprayed using the dry ice spray nozzle according to the dry ice spray pressure, the dry ice particle density, and the dry ice spray time, and the dry ice spray nozzle is moved according to the dry ice spray trajectory.

[0109] Specifically, this embodiment generates a coverage path (such as a spiral or grid method) based on the stain distribution, and divides the cleaning priority area based on the stain density, prioritizing the treatment of heavily stained areas. It can be understood that the boundary information of the cleaning area is the vertex coordinates of the polygon area, which defines the current dry ice spray area. The stain density can be determined according to the proportion of stain area within each preset area range. The motion constraint parameters include maximum speed, acceleration, and turning radius. Thus, starting from the current target stain center position, the path is planned in descending order of stain density to obtain the dry ice spray trajectory.

[0110] As a preferred embodiment, after dry ice blasting the stain based on the dry ice blasting parameters, the method further includes:

[0111] The mixed gas within the dry ice spray area is recovered; wherein the mixed gas includes dry ice particles, dirt, and carbon dioxide;

[0112] The mixed gas is separated once by a cyclone separator to obtain the dirt and the dry ice particles from the mixed gas, and the dirt and the dry ice particles are collected in a preset dirt storage tank and dry ice storage tank respectively.

[0113] The remaining mixed gas after the first separation process is subjected to a second separation process using an electrostatic precipitator to separate carbon dioxide from the remaining mixed gas. The carbon dioxide is then liquefied and collected in a pre-designated carbon dioxide storage tank, or the carbon dioxide is adsorbed and stored through a molecular sieve.

[0114] Specifically, such as Figure 3 As shown, this embodiment proposes a closed-loop-based graded recovery method to recover the mixed gas in the dry ice spray area, reducing resource consumption and avoiding excessive carbon dioxide emissions. First, since dry ice particles are difficult to distinguish after mixing with dirt, a cyclone separator is used for gas-solid separation, based on the principle of density difference (dry ice density 1.5 g / cm³). 3 (The dirt is usually lighter). Larger dirt particles and dry ice fragments fall into the dirt storage tank and dry ice storage tank respectively. Micron-sized particles are filtered out, and the remaining mixed gas enters the electrostatic precipitator. The electrostatic precipitator can capture micron-sized particles from 0.1 microns to 50 microns, thereby separating carbon dioxide. The separated carbon dioxide is recovered through a compression / adsorption system. The recovered carbon dioxide can be remade into dry ice (requires external equipment) or directly used for secondary spraying (gas-assisted cleaning).

[0115] In addition, this embodiment can also use a filtration and purification device to filter the recovered carbon dioxide with activated carbon to remove odors and volatile organic compounds (VOCs).

[0116] As a preferred embodiment, the method further includes:

[0117] When it is determined that there are no stains in the building surface image, the distance between the current position and the edge of the building surface to be cleaned is calculated;

[0118] When the distance between the two sides is greater than the first preset distance, the system moves along the normal of the building surface to be cleaned toward the edge of the building surface and calculates the distance between the two sides in real time. When the distance between the two sides is detected to remain unchanged within a preset time period, the system moves back a preset distance, switches the moving angle, and continues to move toward the edge of the building surface until the distance between the two sides is detected to decrease within the preset time period. Then, the system re-acquires the building surface image of the building to be cleaned and performs stain recognition on the building surface image.

[0119] When the distance between the two sides is less than or equal to the first preset distance and greater than or equal to the second preset distance, the system moves towards the edge of the building surface according to the preset approximation speed corresponding to the building surface material of the building to be cleaned, and updates the distance between the two sides at a frequency of 10Hz during the movement, performs multispectral scanning on the building surface to reacquire the building surface image of the building to be cleaned, and identifies stains on the building surface image.

[0120] When the distance between the two sides is less than the second preset distance, the system moves towards the edge of the building surface until the distance between the two sides reaches the preset minimum distance. A contact sensor array is used to detect changes in the pressure distribution on the building surface, and the standard deviation of the contact pressure on the building surface is calculated based on these changes. When the standard deviation of the contact pressure is greater than a preset pressure standard deviation threshold, a vibration signal is emitted to the building surface to identify whether there are stains on the building surface. If stains are present, the building surface image to be cleaned is re-acquired, and stain identification is performed on the building surface image. If stains are not present, dry ice is sprayed onto the building surface according to preset dry ice cleaning parameters.

[0121] Specifically, when no stains are detected in the building surface image, this embodiment employs a distance-based control strategy for movement. When the distance between the current position and the edge of the building surface to be cleaned is greater than a first preset distance (e.g., 50cm), it indicates that the current distance is too far, and the core objective should be to quickly approach the edge and avoid path deviation. In this case, the priority of navigation accuracy is higher than cleaning efficiency. Therefore, the map is corrected using the SLAM algorithm, and the straight-line distance between the current position and the edge of the building surface to be cleaned is set as the navigation path. The edge approach mode is enabled (obstacle avoidance sensitivity is reduced to 30%, and movement is prioritized along the wall normal direction). A large-scale wall outline is detected using LiDAR, and an infrared sensor compensates for edge angle deviations. If no change in distance is detected for 5 consecutive seconds, an anti-jamming mechanism is triggered: the distance is reversed by 1m and then switched to a 45° angled approach until the distance is detected to decrease within 5 seconds. The building surface image of the building to be cleaned is then reacquired, and stain identification is performed on the building surface image.

[0122] When the distance between the objects is less than or equal to the first preset distance (e.g., 50cm) and greater than or equal to the second preset distance (e.g., 10cm), the core objective should be to precisely adjust the posture, prepare for contact detection, and prioritize sensor calibration over movement speed. Switch to millimeter-wave radar + visual servo system, updating the distance data at a frequency of 10Hz. The approach speed is dynamically adjusted based on the building surface material. Understandably, for more fragile surface materials, such as plaster, a lower preset approach speed is used to prevent damage to the wall due to excessive speed; for solid surface materials, such as concrete, a relatively higher approach speed can be set to improve cleaning efficiency. Extend a retractable camera (3cm-5cm from the wall) to perform multispectral scanning. If suspected stains (e.g., areas with abnormal reflection) are found, they are marked as points to be confirmed.

[0123] When the distance between the two objects is less than the second preset distance (e.g., 10cm), it indicates that the current distance is too close. The core objective is to maintain stable contact. A deep scan is initiated, and the control priority of mechanical stability is higher than that of energy consumption. The object moves towards the edge of the building surface until the distance reaches the preset minimum distance. The pressure distribution change on the building surface is detected using a contact sensor array (e.g., piezoelectric film, conductive rubber). When the standard deviation of the contact pressure is greater than the preset standard deviation threshold of the pressure (e.g., 15%), it indicates that the surface is uneven. A high-frequency vibration mode (50Hz) is initiated to assist in detecting stains. Even if no stains are identified, a cleaning spray is still performed once according to the preset dry ice cleaning parameters. It can be understood that the dry ice cleaning parameters can be set to a fixed and universal dry ice spray pressure, dry ice particle density, and dry ice spray time, and a fixed dry ice spray path can be set, such as a horizontal path, a vertical path, or a spiral path.

[0124] Therefore, the overall dry ice cleaning process can be as follows: Figure 4 As shown.

[0125] As a preferred embodiment, the method further includes:

[0126] Real-time monitoring of the remaining dry ice level in the dry ice storage tank;

[0127] When the remaining dry ice is detected to be less than a preset threshold, an insufficient dry ice alarm is issued.

[0128] Specifically, in this embodiment, a sensor is installed inside the dry ice storage tank to detect the remaining dry ice in the tank in real time. When the remaining dry ice is detected to be less than a preset threshold, an audible and visual alarm signal can be issued to indicate that the remaining amount is insufficient. Alternatively, a low remaining amount alarm message can be sent to relevant terminal devices via communication technology so that relevant personnel can replace the dry ice storage tank in a timely manner.

[0129] Please see Figure 5 The second aspect of the present invention provides a dry ice cleaning device, including a housing 1, a moving module 2, a storage module (not shown in the figure), a recycling module (not shown in the figure), a telescopic arm module 4, a dry ice spraying module 5, and a control module (not shown in the figure);

[0130] The moving module 2 is located on both sides of the outer casing 1; the storage module, the recycling module, and the telescopic arm module 4 are all located on the outer casing 1;

[0131] The dry ice injection module 5 is located at the end of the telescopic arm module 4. The dry ice injection module 5 includes a high-pressure air generating module, a dry ice injection nozzle, and a vacuum adsorption module. The input end of the dry ice injection nozzle is connected to the gas output end of the high-pressure air generating module, and the input end of the dry ice injection nozzle is connected to the recovery module through the dry ice delivery pipe 3. The vacuum adsorption module is positioned close to the dry ice injection nozzle and is connected to the recovery module.

[0132] The storage module includes a dirt storage tank, a dry ice storage tank, and a carbon dioxide storage tank. The recycling module has a dirt separation port, a dry ice separation port, and a carbon dioxide separation port, which are respectively connected to the dirt storage tank, the dry ice storage tank, and the carbon dioxide storage tank.

[0133] The control module is located inside the housing 1, and the control terminal of the control module is connected to the controlled terminal of the moving module 2, the controlled terminal of the storage module, the controlled terminal of the recycling module, the controlled terminal of the telescopic arm module 4, and the controlled terminal of the dry ice spraying module 5, respectively; the control module is used to perform the dry ice cleaning method as described in any embodiment of the first aspect.

[0134] Specifically, the dry ice storage tank is further composed of a vacuum-insulated storage tank and several modular storage tanks. The modular storage tanks play a role in precisely controlling the amount of dry ice sprayed. They are connected to the dry ice separation port of the recovery module to receive the recovered dry ice particles. Their conveying port is connected to the interior of the vacuum-insulated storage tank. The vacuum-insulated storage tank stores granular dry ice, which is connected to the dry ice spray nozzle through the dry ice conveying pipe 3. The inner wall of the conveying pipe needs to be treated with anti-static agents, such as by coating with conductive / antistatic materials to reduce resistivity, dissipate static charge, and prevent dry ice from clumping.

[0135] Furthermore, the high-pressure air generation module generates a high-pressure airflow (pressure range 5 bar to 15 bar) via an air compressor, propelling dry ice particles at high speed to impact the wall surface. The dry ice spray nozzle is designed with an adjustable angle, allowing for flexible control of the dry ice spray trajectory and coverage area. The dry ice spray nozzle is connected to the high-pressure air generation module and the dry ice delivery pipe 3, mixing the high-pressure air and dry ice particles before spraying. When the dry ice particles impact the wall surface, they sublimate instantaneously, creating a micro-explosion effect that peels off dirt. Combined with the low-temperature embrittlement effect, it separates oil stains, paint, and other contaminants from the wall surface. Different spray nozzles are intelligently adjusted according to different cleaning needs to achieve the best cleaning effect; for example, a round nozzle is used for small-area fine cleaning of stains, while a flat nozzle is used for large-area rapid cleaning.

[0136] Furthermore, the vacuum adsorption module is positioned close to the dry ice spray nozzle, so that the remaining mixed gas after dry ice cleaning can be adsorbed and recovered to the recovery module at the same time. It uses a negative pressure suction port and a vacuum pump to recover unsublimated dry ice particles, dirt and carbon dioxide gas. The recovery module further includes a cyclone separator that uses centrifugal force to separate large particles of dirt and dry ice debris, and an electrostatic precipitator that captures micron-sized particles (such as dust). In addition, the recycling module is equipped with a mechanical sorting mechanism, including an airlock ash discharge valve, an eddy current separator, and a vibrating screen. The airlock ash discharge valve alternately opens the dirt separation port and the dry ice separation port by rotating a vane valve (5 rpm to 15 rpm), ensuring that the leakage rate of dirt in the dry ice storage tank is less than 0.1%. The eddy current separator uses an alternating magnetic field (frequency of 50 Hz) to repel non-magnetic metal contaminants, resulting in a metal removal rate of greater than 99%. The vibrating screen uses a double-layer screen (1 mm / 3 mm aperture) to ensure that dry ice particles return to the dry ice storage tank and dirt returns to the dirt storage tank, with a screening efficiency of greater than 95%. Furthermore, this embodiment also includes a sensor network for anti-mixing, including an infrared spectrometer, a capacitive level gauge, and a laser particle size analyzer. The infrared spectrometer is located at the outlet of the cyclone separator to detect the dry ice content in real time. The capacitive level gauge is located inside each modular storage tank to monitor the dry ice storage amount. The laser particle size analyzer is located at the inlet of the electrostatic precipitator to classify micron-sized particles.

[0137] Furthermore, a filtration and purification device is fixedly installed inside the carbon dioxide storage tank to filter the recovered carbon dioxide gas with activated carbon, removing odors and volatile organic compounds (VOCs).

[0138] Furthermore, the control module can be powered by a lithium battery pack, and the surface of the outer casing 1 is covered with a solar panel (lightweight flexible photovoltaic film) as an auxiliary power source for the control module.

[0139] Furthermore, the moving module 2 may further include tracks, a motor, a drive gear, and a combination gear. The motor drives the tracks and drive gear to rotate at high speed, providing strong and stable friction, enabling the dry ice cleaning device to move vertically, horizontally, or diagonally along the ground. The telescopic arm module 4 is used to clean areas at different heights.

[0140] The dry ice cleaning device provided in this embodiment of the invention identifies stains on the building surface image of the building to be cleaned, and performs dry ice cleaning based on the stain information after the stains are identified. Since dry ice cleaning does not require the use of water or mechanical tools to contact the building surface, it can quickly remove stains from the building surface without causing secondary pollution or damaging the building surface, thus simultaneously ensuring both cleaning effect and the integrity of the building surface.

[0141] In addition, a closed-loop, staged recycling method is used to recover the mixed gas in the dry ice spray area, reducing resource consumption and avoiding excessive carbon dioxide emissions.

[0142] In an optional embodiment, this embodiment can also provide a dry ice cleaning system, such as... Figure 6 As shown, preferably, the dry ice cleaning system may include a dry ice cleaning device and a server. The sensors of the dry ice cleaning device include a camera, an ultrasonic sensor, an infrared sensor, a wall type sensor, an environmental sensor, an obstacle sensor, a collision sensor, etc., and are connected to the main control board.

[0143] Preferably, the server configuration for the dry ice cleaning system should include suitable server hardware, such as CPU, memory, disk, and network interface, and a suitable operating system, such as Linux, should be installed. Then, based on the requirements of the dry ice cleaning system, appropriate server software should be installed, such as ROS (the operating system for cleaning devices), OpenCV (a computer vision library), and programming languages ​​(such as Python, C++, etc.).

[0144] Preferably, the control program of the dry ice cleaning device is written in a programming language. Based on the dry ice cleaning requirements and sensor data, appropriate algorithms are written to realize the dry ice cleaning function, including path planning algorithm, obstacle avoidance algorithm, dynamic adjustment, dry ice balance monitoring, remote control, etc.

[0145] Preferably, the dry ice cleaning device and the server are connected via Bluetooth or wired connection. Connecting the dry ice cleaning device to the server ensures that the sensors and actuators of the dry ice cleaning device can be controlled by the server and that the sensor data is transmitted to the server.

[0146] Preferably, the basic functions of the dry ice cleaning device include metered dry ice injection, path movement, emergency stop, obstacle avoidance, automatic tank filling, and remote control. Specifically, metered dry ice injection is achieved through a program that allows for adjustment of parameters such as the angle and pressure of the dry ice nozzle, the delivery speed of the dry ice particles, and the injection time. The operating mode of the dry ice cleaning device is adjusted based on sensor data feedback regarding the wall material and degree of dirt, adapting to different cleaning objects and environmental needs. Path movement is achieved through a program that controls the movement of the dry ice cleaning device, including basic actions such as forward, backward, turning, and left / right movement. It also features map creation and navigation functions for the dry ice cleaning device, allowing users to create cleaning plans or designate cleaning areas; emergency stop, enabling the device to immediately halt all operations in emergency situations; obstacle avoidance, automatically stopping or bypassing obstacles detected by the device; automatic tank refilling, addressing the issue of continuous dry ice supply by replenishing dry ice without interrupting cleaning based on feedback from dry ice and level sensors; and remote control, enabling real-time monitoring and control of the dry ice cleaning device via remote server access.

[0147] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A dry ice cleaning method, characterized in that, include: Acquire images of the building surface to be cleaned, and identify stains in the building surface images; When stains are detected in the building surface image, stain information is obtained; Based on the stain information, determine the current dry ice spray parameters; The stain is subjected to dry ice blasting based on the dry ice blasting parameters; When it is determined that there are no stains in the building surface image, the distance between the current position and the edge of the building surface to be cleaned is calculated; When the distance between the two sides is greater than the first preset distance, the system moves along the normal of the building surface to be cleaned toward the edge of the building surface and calculates the distance between the two sides in real time. When the distance between the two sides is detected to remain unchanged within a preset time period, the system moves back a preset distance, switches the moving angle, and continues to move toward the edge of the building surface until the distance between the two sides is detected to decrease within the preset time period. Then, the system re-acquires the building surface image of the building to be cleaned and performs stain recognition on the building surface image. When the distance between the two sides is less than or equal to the first preset distance and greater than or equal to the second preset distance, the system moves towards the edge of the building surface according to the preset approximation speed corresponding to the building surface material of the building to be cleaned, and updates the distance between the two sides at a frequency of 10Hz during the movement, performs multispectral scanning on the building surface to reacquire the building surface image of the building to be cleaned, and identifies stains on the building surface image. When the distance between the two sides is less than the second preset distance, the device moves towards the edge of the building surface until the distance between the two sides reaches the preset minimum distance. The device uses a contact sensor array to detect the pressure distribution change on the building surface and calculates the standard deviation of the contact pressure on the building surface based on the pressure distribution change. When the standard deviation of the contact pressure is greater than the preset standard deviation threshold of the pressure, a vibration signal is emitted to the building surface to identify whether there are stains on the building surface; If present, the building surface image of the building to be cleaned is reacquired, and stain identification is performed on the building surface image. If not, dry ice is sprayed onto the building surface according to the preset dry ice cleaning parameters.

2. The dry ice cleaning method as described in claim 1, characterized in that, The process of acquiring an image of the building surface to be cleaned and identifying stains on the building surface image specifically includes: Acquire a multispectral image of the building surface to be cleaned, and use the multispectral image as the building surface image; The stain recognition model is used to identify stains in the multispectral image to obtain stain recognition results; wherein, the stain recognition model is obtained by training a multichannel convolutional neural network using multispectral training images with stain marking information; When it is determined that any type of stain exists based on the stain identification result, it is determined that there is a stain in the building surface image.

3. The dry ice cleaning method as described in claim 2, characterized in that, The step of using a pre-trained stain recognition model to identify stains in the multispectral image and obtaining stain recognition results specifically includes: Feature extraction is performed on the multispectral image to obtain multimodal stain features; The CBAM module is used to enhance the multimodal stain features to obtain enhanced stain features; A classifier is used to identify the enhanced stain features, and a stain probability heatmap is generated. Based on the stain probability heatmap, determine the stain bounding box and the target stain type; The FAST corner detection algorithm is used to extract a preset number of candidate points within the stain bounding box, and based on the gradient magnitude of each candidate point, several stain feature points are obtained. The positions of several stain feature points are iteratively optimized based on the gray-level gradient method until the position change of the stain feature points is less than a preset pixel value. The target stain center position is determined based on the centroid of each stain feature point after iterative optimization. The stain identification result is obtained based on the target stain type and the center position of the target stain.

4. The dry ice cleaning method as described in claim 3, characterized in that, The process of obtaining stain information specifically includes: The surface of the building to be cleaned is scanned using a lidar sensor to construct a three-dimensional model of the building surface. Based on the point cloud data in the three-dimensional building surface model and the positions of each stain feature point after iterative optimization, the stain area in the three-dimensional building surface model is determined. Based on the point cloud data and the stained area, calculate the stained area and the building surface area of ​​the building to be cleaned, and calculate the maximum distance difference between the stained area and the building surface; Based on the stain area and the building surface area, calculate the area ratio of the stain area, and based on the preset stain stubbornness levels corresponding to different area ratios and distance differences, determine the area ratio of the stain area and the target stain stubbornness level corresponding to the maximum distance difference. The stain information of the stain is determined based on the type of the target stain, the center location of the target stain, and the stubbornness level of the target stain.

5. The dry ice cleaning method as described in claim 4, characterized in that, The step of determining the current dry ice spray parameters based on the stain information specifically includes: Based on the preset dry ice blasting parameters corresponding to different stain types and stain stubbornness levels, determine the current dry ice blasting pressure, dry ice particle density, and dry ice blasting time; Determine the current dry ice spray angle based on the center position of the target stain; The current dry ice injection parameters are determined based on the dry ice injection pressure, the dry ice particle density, the dry ice injection time, and the dry ice injection angle.

6. The dry ice cleaning method as described in claim 5, characterized in that, The process of dry ice blasting the stain based on the dry ice blasting parameters specifically includes: The dry ice spray nozzle is adjusted according to the dry ice spray angle so that it is aimed at the center of the target stain. Based on the aforementioned stain probability heatmap, the stain density at different locations is determined; Based on the preset cleaning area boundary information and motion constraint parameters, starting from the center position of the target stain, the path is planned in descending order of stain density to obtain the dry ice spray trajectory. Dry ice is sprayed using the dry ice spray nozzle according to the dry ice spray pressure, the dry ice particle density, and the dry ice spray time, and the dry ice spray nozzle is moved according to the dry ice spray trajectory.

7. The dry ice cleaning method as described in claim 1, characterized in that, After dry-ice blasting the stain based on the dry ice blasting parameters, the method further includes: The mixed gas within the dry ice spray area is recovered; wherein the mixed gas includes dry ice particles, dirt, and carbon dioxide; The mixed gas is separated once by a cyclone separator to obtain the dirt and the dry ice particles from the mixed gas, and the dirt and the dry ice particles are collected in a preset dirt storage tank and dry ice storage tank respectively. The remaining mixed gas after the first separation process is subjected to a second separation process using an electrostatic precipitator to separate carbon dioxide from the remaining mixed gas. The carbon dioxide is then liquefied and collected in a pre-designated carbon dioxide storage tank, or the carbon dioxide is adsorbed and stored through a molecular sieve.

8. The dry ice cleaning method as described in claim 7, characterized in that, The method further includes: Real-time monitoring of the remaining dry ice level in the dry ice storage tank; When the remaining dry ice is detected to be less than a preset threshold, an insufficient dry ice alarm is issued.

9. A dry ice cleaning device, characterized in that, It includes a housing, a moving module, a storage module, a recovery module, a telescopic arm module, a dry ice spraying module, and a control module; The moving module is located on both sides of the outer casing; the storage module, the recycling module, and the telescopic arm module are all located on the outer casing. The dry ice injection module is located at the end of the telescopic arm module. The dry ice injection module includes a high-pressure air generating module, a dry ice injection nozzle, and a vacuum adsorption module. The input end of the dry ice injection nozzle is connected to the gas output end of the high-pressure air generating module, and the input end of the dry ice injection nozzle is connected to the recovery module through a dry ice delivery pipe. The vacuum adsorption module is positioned close to the dry ice injection nozzle and is connected to the recovery module. The storage module includes a dirt storage tank, a dry ice storage tank, and a carbon dioxide storage tank. The recycling module has a dirt separation port, a dry ice separation port, and a carbon dioxide separation port, which are respectively connected to the dirt storage tank, the dry ice storage tank, and the carbon dioxide storage tank. The control module is located inside the housing, and the control terminal of the control module is connected to the controlled terminal of the moving module, the controlled terminal of the storage module, the controlled terminal of the recovery module, the controlled terminal of the telescopic arm module, and the controlled terminal of the dry ice spraying module, respectively; the control module is used to perform the dry ice cleaning method as described in any one of claims 1 to 8.

Images