Double-arm robot autonomous cleaning system and method for highway tunnel electromechanical facilities

Abstract

The application discloses a two-arm robot autonomous cleaning system and method for highway tunnel electromechanical facilities, comprising the following steps: the two-arm robot approaches the target electromechanical facility and collects a pre-cleaning surface image, extracts high-frequency components, contrast, local gradient and reference image offset features, combines an improved PIDNet dirty area segmentation model to determine the dirt type and degree grade; according to the dirt degree, determine the cleaning necessity, match the cleaning method and cleaning parameters from the parameter library, and generate two-arm cleaning operation instructions; the double mechanical arms perform brushing, spraying, air blowing or squeegeeing actions, re-collect images after cleaning for closed-loop re-inspection, upgrade parameters and supplementary cleaning when not up to standard, and report abnormities when not up to standard after reaching the upper limit of supplementary cleaning; according to the re-inspection result, the parameter library is corrected and a subsequent task queue is generated. The application realizes on-demand cleaning, autonomous execution and closed-loop operation and maintenance of highway tunnel electromechanical facilities.

Description

Technical Field

[0001] This invention relates to the field of special robots and intelligent operation and maintenance technology of electromechanical facilities in highway tunnels, specifically to a dual-arm robot autonomous cleaning system and method for electromechanical facilities in highway tunnels. Background Technology

[0002] Highway tunnels typically house lighting fixtures, surveillance cameras, traffic guidance equipment, and other electromechanical facilities. During long-term operation, the light-receiving, imaging, and exposed functional surfaces of these facilities are susceptible to the accumulation of contaminants such as dust, oil, water stains, and solidified dirt. Especially in environments with continuous vehicle traffic, exhaust particles, dust, and humid air can easily accumulate on the surfaces, affecting the light transmission performance of the lighting fixtures, the image quality of the cameras, and the normal operation of other electromechanical facilities. Therefore, regular cleaning and maintenance of these facilities is necessary.

[0003] Existing highway tunnel electromechanical facilities are typically cleaned through manual inspections followed by centralized cleaning, cleaning scheduled according to fixed cycles, or using mechanized cleaning equipment. Manual cleaning requires workers to enter the corresponding work area, and for facilities located at higher positions, climbing equipment is also needed. The efficiency and safety of the operation are easily limited by the tunnel space and traffic environment. When cleaning is carried out according to fixed cycles, it is difficult to determine whether cleaning is needed based on the actual dirt status of different facilities. This can easily lead to the problem of lightly contaminated facilities being cleaned repeatedly while heavily contaminated facilities are not treated in a timely manner.

[0004] The cleaning methods, cleaning time, contact pressure, brush roller speed, and spray volume of existing cleaning equipment are usually preset or manually set by operators, making it difficult to adjust them according to different types and degrees of dirt, such as floating dust, oil stains, water stains, and solidified dirt. When using the same cleaning parameters to treat different levels of contamination, it can easily lead to excessively long cleaning times and increased resource consumption in cases of light dirt, or insufficient cleaning and the need for repeated operations in cases of heavy dirt.

[0005] Some existing cleaning equipment stops its current operation after cleaning, lacking an automatic re-inspection process for the same facility surface. This makes it impossible to promptly determine whether the post-cleaning level of dirt meets the target requirements. Facilities that do not meet the requirements usually still require manual inspection or re-scheduling for cleaning, making it difficult to form a closed-loop process combining pre-cleaning inspection, cleaning decision-making, cleaning execution, and post-cleaning re-inspection. Furthermore, for the multiple electromechanical facilities distributed along the tunnel, the current method cannot easily create continuous, on-demand cleaning schedules based on the dirt level of each facility, its location, and the robot's operating status.

[0006] Therefore, how to provide a dual-arm robot autonomous cleaning system and method for electromechanical facilities in highway tunnels is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a dual-arm robotic autonomous cleaning system and method for electromechanical facilities in highway tunnels. This invention involves a dual-arm robot approaching the target electromechanical facility and visually acquiring its illuminated / imaging surface. It extracts quantitative indicators of dirt from the surface image, such as high-frequency components, contrast, local distortion, and color / grayscale shift. An improved PIDNet dirt region segmentation model is used to obtain the type and degree of dirt. Based on the degree of dirt, the necessity of cleaning is determined. Cleaning methods and parameters are matched from a pre-established parameter library according to the type and degree of dirt, generating a dual-arm cleaning operation command. The first and second cleaning robotic arms then execute the corresponding cleaning actions. After cleaning, the surface image of the target electromechanical facility is acquired again for closed-loop re-inspection. For cleaning results that do not meet the target level, parameter upgrade information is provided, and supplementary cleaning is performed. Simultaneously, the facility dirt assessment data, cleaning parameters, and re-inspection results are recorded and uploaded, realizing dirt identification, on-demand decision-making, autonomous cleaning, and closed-loop re-inspection of electromechanical facilities in highway tunnels.

[0008] The dual-arm robot autonomous cleaning system and method for electromechanical facilities in highway tunnels according to embodiments of the present invention includes the following steps: S1. The dual-arm robot approaches the target electromechanical facility, identifies the location of the target electromechanical facility through the environmental camera, and adjusts the robot's posture so that the working camera is aimed at the light-receiving / imaging surface of the target electromechanical facility. S2. The target electromechanical facilities are illuminated by the supplementary lighting unit. One or more frames of surface images of the facilities are captured by the working camera. Images that meet the clarity requirements are selected as evaluation images and associated with the facility image archive. S3. Extract the high-frequency component energy ratio, contrast index, local distortion index and color / grayscale shift index of the evaluation image, and perform fusion processing through the improved PIDNet dirty area segmentation model to obtain the dirty type and dirty degree level. S4. Compare the level of dirtiness with the preset cleaning threshold. If the level is lower than the preset cleaning threshold, record the evaluation result and execute S8. If the level reaches or exceeds the preset cleaning threshold, execute S5. S5. Query the pre-established parameter library according to the type and degree of dirt, match the corresponding cleaning method and cleaning parameters, and generate dual-arm cleaning operation instructions based on the matching results. S6. Adjust the robot's position so that the cleaning execution module is close to the target electromechanical facility, and the first and second cleaning robotic arms execute the corresponding cleaning actions according to the dual-arm cleaning operation instructions. S7. Collect images of the target electromechanical facility surface again and conduct a quantitative assessment of dirtiness. Compare the level of dirtiness after cleaning with the target level. If the target level is not reached, generate parameter upgrade information and perform re-cleaning. S8. Generate and upload corresponding work records based on whether the target electromechanical facilities have been cleaned. After completing the processing of the current target electromechanical facilities, move to the next target electromechanical facilities or return to the standby position.

[0009] Optionally, step S1 includes the following steps: S11. Determine the location of the target electromechanical facility based on the task information, and move along the corresponding pole of the target electromechanical facility to the preset working height using the climbing and walking module; S12. Collect images of the target electromechanical facilities and their surrounding environment using the pole-mounted environmental camera, and identify the spatial orientation of the target electromechanical facilities based on the environmental images. S13. Control the climbing and walking module to rise and fall according to the spatial orientation of the target electromechanical facilities, and adjust the robot's circumferential working position through the pole-turning mechanism; S14. Based on the adjusted robot pose, orient the working camera toward the target electromechanical facility's light-receiving / imaging surface and maintain a working position that meets the requirements for image acquisition and cleaning execution.

[0010] Optionally, step S2 includes the following steps: S21. Control the supplementary lighting unit to supplement light according to the preset brightness curve based on the current brightness of the target electromechanical facility's light-receiving / imaging surface; S22. Control the operation camera to continuously acquire one or more frames of facility surface images, and evaluate the sharpness of each acquired frame of facility surface image. S23. Select the facility surface image with the highest clarity or that meets the preset clarity requirement as the evaluation image, and associate it with the corresponding facility number and acquisition time. S24. During the initial operation, establish a facility image file and associate it with a pre-stored clean status baseline image. Subsequently, retrieve the corresponding clean status baseline image based on the facility number.

[0011] Optionally, step S3 includes the following steps: S31. Preprocess and frequency domain transform the evaluation image, calculate the proportion of high-frequency sub-band energy to the total frequency band energy, and obtain the high-frequency component energy ratio. S32. Calculate the contrast and local distortion index of the evaluation image, and compare the evaluation image with the clean state reference image to obtain the color / grayscale shift index. S33. Input the evaluation image into the improved PIDNet dirty region segmentation model, and convert each dirty quantification index into dirty prior features, and fuse them with the multi-branch image features extracted by the model. S34. Obtain the dirty area segmentation result based on the fused features, and determine the dirty type and corresponding dirty degree level based on the dirty area features.

[0012] Optionally, step S4 includes the following steps: S41. Receive the type and level of dirt obtained by S3, and read the preset cleaning threshold corresponding to the target electromechanical facility. S42. Compare the level of dirtiness with the preset cleaning threshold. If the level of dirtiness is lower than the preset cleaning threshold, it is determined that cleaning is not required. S43. For target electromechanical facilities that do not require cleaning, save the assessment images, dirt type, dirt level and assessment time, and directly execute S8; S44. When the level of dirt reaches or exceeds the preset cleaning threshold, it is determined that cleaning is required, and the type of dirt and the level of dirt are sent to the cleaning decision module.

[0013] Optionally, step S5 includes the following steps: S51. Based on the type of dirt, query the pre-established parameter database to determine the dry, wet, or combined cleaning method corresponding to the type of dirt. S52. Match cleaning time, contact pressure, brush roller speed, spray volume, cleaning agent ratio, and number of passes according to the degree of dirtiness; S53. Adjust the cleaning parameters according to the corresponding parameter level based on the degree of dirtiness, with priority given to adjusting the cleaning time, and then adjusting the contact pressure between the brush roller and the facility surface. S54. Generate a work instruction that includes double-arm cleaning actions, corresponding parameters, and start / stop sequence based on the cleaning method and cleaning parameters; when receiving parameter upgrade information, adjust the current parameter level according to the parameter upgrade information and generate a re-cleaning work instruction.

[0014] Optionally, step S6 includes the following steps: S61. Adjust the robot's working posture through the climbing and walking module so that the cleaning execution ends of the first and second cleaning robotic arms approach the surface of the target electromechanical facility. S62. Control the first and second cleaning robotic arms to call one or more of the following components according to the work instructions: brush plate or brush roller, spray component, air blowing component and squeegee component. S63. Control the first cleaning robotic arm and the second cleaning robotic arm to operate in coordination according to the corresponding cleaning actions and start-stop sequence, and execute the matching cleaning parameters; S64. The brush plate or brush roller is made to adhere to the surface of the target electromechanical facility through the universal connection structure, and cleaning is stopped after the set cleaning time and number of operations are completed.

[0015] Optionally, step S7 includes the following steps: S71. After cleaning, collect images of the same target electromechanical facility surface again, and obtain the degree of dirtiness after cleaning according to the dirt quantification assessment method in S3. S72. Compare the level of dirtiness after cleaning with the preset target level. If it is equal to or lower than the target level, the cleaning is deemed qualified. Record the actual parameter combination and the re-inspection results. S73. When the level of dirtiness after cleaning is higher than the target level, parameter upgrade information is generated based on the current parameter combination and fed back to the cleaning decision module to generate a re-cleaning operation instruction. S74. Repeat cleaning and re-inspection according to the cleaning operation instructions. End the cleaning operation when the target level is reached. If the target level is not met after reaching the maximum number of cleaning operations, mark the abnormality and report it.

[0016] Optionally, step S8 includes the following steps: S81. For target electromechanical facilities that do not require cleaning, write the facility number, facility location, assessment image, dirt assessment data and assessment time into the corresponding work record; S82. For the target electromechanical facilities to be cleaned, write the before and after cleaning images, actual cleaning parameters, dirt assessment data and re-inspection results into the corresponding work record. S83. Calculate the first pass rate of different parameter combinations based on the cumulative operation records, adjust the cleaning time and contact pressure level according to the first pass rate and update the parameter library; S84. Upload the corresponding job record, and the cloud-based operation and maintenance platform will generate a subsequent task queue and job sequence, and control the robot to move to the next target electromechanical facility or return to the standby position.

[0017] An autonomous cleaning system for dual-arm robotic systems used in highway tunnel electromechanical facilities, according to an embodiment of the present invention, includes: The climbing and walking module is used to lift, stop, and adjust the working position of the dual-arm robot relative to the target electromechanical facility along the pole. The visual acquisition module includes a working camera and a supplementary lighting unit, used to acquire images of the target electromechanical facility's illuminated / imaging surface before and after cleaning; The dirt assessment module is used to extract the dirt quantification index of facility surface images and determine the dirt type and dirt level through the improved PIDNet dirt area segmentation model. The parameter library is used to store corresponding cleaning methods and cleaning parameters according to the type and degree of dirt, and to adjust the corresponding parameter levels based on the cumulative re-inspection results; The cleaning decision module is used to determine the necessity of cleaning based on the level of dirt, and to match cleaning methods and cleaning parameters from the parameter library according to the type and level of dirt to generate dual-arm cleaning operation instructions; The cleaning execution module includes a first cleaning robotic arm and a second cleaning robotic arm, as well as various components such as a brushing mechanism, a spraying component, an air blowing component, and a squeegee component disposed at the working ends of the two cleaning robotic arms. The re-inspection module is used to conduct a quantitative assessment of the dirt and grime of the target electromechanical facilities after cleaning, trigger additional cleaning based on the re-inspection results, and generate abnormal information for target electromechanical facilities that still fail to meet the standards after reaching the maximum number of additional cleaning cycles. The main control and communication module is connected to the climbing and walking module, vision acquisition module, dirt assessment module, cleaning decision module, parameter library, cleaning execution module and re-inspection module respectively, and is used to coordinate robot positioning, image acquisition, cleaning decision, dual-arm cleaning, closed-loop re-inspection and operation record uploading.

[0018] The beneficial effects of this invention are as follows: By visually collecting and quantitatively assessing the actual dirt status of electromechanical facilities in highway tunnels, this invention transforms cleaning operations from fixed-cycle triggering to on-demand triggering based on the degree of dirt. For facilities that do not reach the cleaning threshold, only their assessment results are recorded without cleaning, thus reducing unnecessary cleaning operations and the resulting losses in water, electricity, and cleaning mechanisms. The improved PIDNet dirt area segmentation model, combined with high-frequency components, contrast, local distortion, and color / grayscale shift indicators, identifies dirt on the facility surface, enabling dirt areas, dirt types, and dirt levels to participate in subsequent cleaning decisions, improving the correspondence between dirt assessment results and actual cleaning needs in complex tunnel environments. Based on the dirt type and dirt level, the invention matches cleaning methods, cleaning time, contact pressure, brush roller speed, spray volume, cleaning agent ratio, and number of passes from a parameter library, and then uses a first and second cleaning robotic arm to perform the work according to the specified parameters. The system executes corresponding cleaning actions, allowing the cleaning process to be adjusted according to the level of contamination, reducing over-cleaning in lightly contaminated conditions and under-cleaning in heavily contaminated conditions. After cleaning, images of the same facility surface are re-acquired and a quantitative assessment of contamination is performed. The level of contamination after cleaning is compared with the target level. For cases that do not meet the target level, parameter upgrade information is generated and automatic re-cleaning is performed. Facilities that still do not meet the target after reaching the maximum number of re-cleaning attempts are marked as abnormal and reported, thus forming a closed-loop operation process that connects cleaning detection, cleaning decision-making, cleaning execution, and re-inspection and re-cleaning. At the same time, by accumulating cleaning operation records and statistically analyzing the first-pass rate of re-inspection for different parameter combinations, the parameters such as cleaning time and contact pressure are adjusted. This allows the parameter library to be continuously adjusted according to the actual cleaning effect. Combined with the facility's contamination level, facility priority, robot position, and battery level, a subsequent task queue is formed, which is conducive to improving the operational efficiency and maintenance management level of continuous autonomous cleaning of multiple electromechanical facilities in highway tunnels. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a block diagram of the system module composition and signal flow proposed in this invention; Figure 2 This is a flowchart of the cleaning decision-making method proposed in this invention; Figure 3 This is a flowchart of the dirt and grime quantitative assessment algorithm proposed in this invention; Figure 4 This is a schematic diagram of the parameter library matching and self-learning correction logic proposed in this invention; Figure 5 This is a schematic diagram of the multi-facility task scheduling proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-5 A dual-arm robot autonomous cleaning system and method for electromechanical facilities in highway tunnels, comprising the following steps: S1. The dual-arm robot approaches the target electromechanical facility, identifies the location of the target electromechanical facility through the environmental camera, and adjusts the robot's posture so that the working camera is aimed at the light-receiving / imaging surface of the target electromechanical facility. S2. The target electromechanical facilities are illuminated by the supplementary lighting unit. One or more frames of surface images of the facilities are captured by the working camera. Images that meet the clarity requirements are selected as evaluation images and associated with the facility image archive. S3. Extract the high-frequency component energy ratio, contrast index, local distortion index and color / grayscale shift index of the evaluation image, and perform fusion processing through the improved PIDNet dirty area segmentation model to obtain the dirty type and dirty degree level. S4. Compare the level of dirtiness with the preset cleaning threshold. If the level is lower than the preset cleaning threshold, record the evaluation result and execute S8. If the level reaches or exceeds the preset cleaning threshold, execute S5. S5. Query the pre-established parameter library according to the type and level of dirt, match the corresponding cleaning method and cleaning parameters, and generate dual-arm cleaning operation instructions based on the matching results. S6. Adjust the robot's position so that the cleaning execution module approaches the target electromechanical facility, and the first and second cleaning robotic arms execute the corresponding cleaning actions according to the dual-arm cleaning operation instructions. S7. Collect images of the target electromechanical facility surface again and conduct a quantitative assessment of dirtiness. Compare the level of dirtiness after cleaning with the target level. If the target level is not reached, generate parameter upgrade information and perform re-cleaning. S8. Generate and upload corresponding work records based on whether the target electromechanical facilities have been cleaned. After completing the processing of the current target electromechanical facilities, move to the next target electromechanical facilities or return to the standby position.

[0022] In this embodiment, S1 includes: like Figure 2 As shown, the dual-arm robot receives the current task, determines the target light pole to be processed according to the pre-established inspection task sequence, and moves to the work area corresponding to the target light pole.

[0023] After the robot arrives at the target light pole, it establishes a stable contact with the pole through a clamping structure. The clamping structure can be a clamping arm arranged around the pole, an elastic preload, or other structures that can keep the robot stably attached. By adjusting the clamping force, the robot can maintain a reliable attachment to the pole during climbing, parking, and cleaning with both arms.

[0024] After the drive mechanism is activated, it propels the robot to move along the pole towards the target electromechanical facility at the specified height. During the robot's movement, the pole's environmental camera continuously acquires or acquires environmental images of the pole, the target facility, and their surrounding area at preset time intervals, and sends the acquired environmental images to the robot's main control unit.

[0025] The robot identifies the position of the target electromechanical facility in its current field of view by identifying the environmental image. The robot determines the height of the facility based on its longitudinal offset relative to the image center. The robot determines its circumferential orientation relative to the robot's current working direction based on the imaging relationship of the facility located on the left, right, or other circumferential positions of the pole.

[0026] The target electromechanical facilities can be lighting fixtures, surveillance cameras, and other electromechanical facilities installed on poles or pole-like support structures within highway tunnels. Different target electromechanical facilities can have different installation heights, surface dimensions, and installation orientations. The robot does not need to maintain a fixed cleaning posture in advance, but rather completes the approach and positioning step by step based on the actual environmental images collected.

[0027] Once the robot approaches the target electromechanical facility at its height, it reduces its movement speed along the pole to minimize the impact of robot inertial motion and structural vibration on subsequent visual data acquisition. Small adjustments are made based on the target facility's longitudinal position in the environmental image to gradually bring it within the effective working height range corresponding to the working camera and dual robotic arms.

[0028] When the target electromechanical facility is located outside the robot's current circumferential working direction, the robot is driven to move circumferentially relative to the pole through a rotary structure, so that the vision acquisition part and the cleaning execution part gradually move towards the target facility.

[0029] During the rotation and adjustment process, environmental images are continuously acquired, and the deviation between the robot's current working direction and the target facility's orientation is determined based on the lateral position change of the target facility in the image. Once the target facility gradually moves into the preset area of ​​the corresponding field of view, the large-scale rotation stops, and a small circumferential adjustment is made to allow the working camera to capture a more complete picture of its main surface to be cleaned.

[0030] After the height and circumferential position are adjusted, the working camera will capture images of the current facility to confirm whether the target electromechanical facility is located in the effective acquisition area.

[0031] When the light-receiving or imaging surface of the target electromechanical facility is within the effective field of view of the working camera, and both the first and second cleaning robotic arms have working space close to the surface of the facility, the robot maintains its current position and proceeds to the next operation.

[0032] When the target facility only partially enters the field of view, or when the current posture of the robotic arm cannot meet the working distance requirements on the facility surface, it continues to make corrections through small lifting and circumferential rotation until the working camera and cleaning execution module are simultaneously in a suitable position for operation.

[0033] When multiple target electromechanical facilities are set on the same pole, the height and circumferential position information of each facility can be established separately. After the current facility is processed, the robot does not need to completely leave the pole. Instead, it continues to perform lifting and circumferential adjustments based on the position information of the next facility, thereby reducing the time consumption caused by repeated pole lowering, re-clamping, and repositioning.

[0034] Figure 2 The target light pole is used as a specific illustration of the pole-mounted facility to represent the process of the robot climbing, ascending, descending, rotating, and initially positioning the target facility along the pole. For facilities installed on monitoring poles, support poles, or other pole-like supporting components in highway tunnels, the positioning principle is the same as described above.

[0035] like Figure 1 As shown, the autonomous cleaning system in this embodiment includes a vision acquisition module, a defect assessment module, a re-inspection module, a rod-shaped environmental camera, a cleaning decision module, a parameter library, a material movement module, a cleaning execution module, and a main control and communication module. It also interacts with the cloud-based operation and maintenance platform through the main control and communication module.

[0036] Figure 1 The “Defect Assessment Module” shown is used to perform the dirt assessment function in this embodiment. Its processing object is the abnormal coverage state formed after the surface of the target electromechanical facility is contaminated. Specifically, it is used to analyze the dirty area, dirt type and dirt level of the target facility surface.

[0037] Figure 1The "material walking module" shown corresponds to the motion execution part of the robot that performs clamping, climbing, parking, and circumferential adjustment along the rod, and includes a clamping structure, a driving structure, and a rotating structure. The names are used to indicate the corresponding functional units in the attached drawings, and their specific functions correspond to the robot's climbing, walking, and position adjustment functions in this embodiment.

[0038] Figure 1 The "rod-shaped environmental camera" shown is used to complete the robot's climbing navigation and initial positioning of the target electromechanical facilities. By collecting images of the robot's surrounding environment and target facilities, it provides visual information for adjusting the robot's height and circumferential position.

[0039] Figure 1 The term "accessory cleaning robot" refers to the collection of functional modules contained in the robot body. In this embodiment, the dual-arm cleaning robot for highway tunnel electromechanical facilities is composed of functions such as visual acquisition, environmental perception, dirt assessment, cleaning decision-making, dual-arm cleaning execution, re-inspection, and communication control.

[0040] The vision acquisition module includes at least a working camera and may further include a supplementary lighting unit. The working camera is mounted on the robot body or near the cleaning execution module, with its imaging direction facing the illuminated / imaging surface of the target electromechanical facility.

[0041] Figure 1 The card-type unit shown can be used as an integrated method for visual acquisition components, which centrally installs the work camera, supplementary lighting unit and corresponding acquisition interface, so that it can be disassembled and adjusted when maintenance or when it is necessary to adapt to different facilities' work fields of view.

[0042] The supplementary lighting unit can use an LED ring light. When the overall lighting inside the tunnel is low, the facilities are backlit, or the facilities themselves emit light, resulting in large differences in brightness in local areas, the intensity of the supplementary lighting is adjusted according to the brightness of the images captured by the working camera, so as to form relatively stable imaging conditions for subsequent dirt assessment.

[0043] The defect assessment module receives images of the facility surface, performs image preprocessing, calculates multiple types of dirt indicators, and conducts deep learning fusion analysis to output the current dirt type and dirt level of the facility.

[0044] The cleaning decision module determines whether cleaning is needed based on the level of dirtiness, and when cleaning is needed, it queries the parameter library based on the type and level of dirtiness to obtain the current cleaning method and corresponding parameter combination.

[0045] The parameter library pre-establishes the parameter correspondence between different types of dirt, different levels of dirt, and different cleaning methods, and corrects them based on actual re-inspection results during long-term robot operation.

[0046] The cleaning execution module includes a first cleaning robotic arm and a second cleaning robotic arm. The working ends of the two cleaning robotic arms can be equipped with one or more of the following: brush plate or brush roller, spray component, air blowing component and squeegee component.

[0047] The brush plate or brush roller is connected to the corresponding cleaning robotic arm through a universal connection structure, so that when the brushing mechanism comes into contact with lampshades, camera protective covers or other facility surfaces with different curvatures, it can adapt the angle change according to the surface orientation of the facility to improve the degree of contact between the brushing mechanism and the target surface.

[0048] The spray unit can have both a clean water path and a cleaning agent path, delivering clean water or cleaning agent to the target area through atomizing nozzles. The air blowing unit can be used to treat loosely attached dust or to blow away residual water droplets from the facility surface after wet cleaning; the squeegee unit is used to further remove residual water film.

[0049] The first and second cleaning robotic arms are not limited to a fixed master-slave relationship. Depending on the actual cleaning method, one robotic arm can be responsible for spraying or immersion, while the other robotic arm is responsible for brushing; alternatively, the two robotic arms can handle different areas of the target facility separately; for cleaning tasks that only require a single operation, only one robotic arm can be used, while the other robotic arm remains in a position that does not interfere with visual acquisition and cleaning operations.

[0050] The re-inspection module is used to re-acquire images of the target facility surface after cleaning and call the same or corresponding dirt assessment method as before cleaning to determine the degree of dirt after cleaning.

[0051] In this embodiment, achieving the target level means that the degree of dirtiness after cleaning is equal to or lower than the preset target level. For example, if the target level is set to level 1, achieving level 1 or level 0 after cleaning both constitute achieving the target cleaning requirement.

[0052] The main control and communication module is used to coordinate robot movement, image acquisition, cleaning decisions, dual-arm execution, and re-inspection processes, and upload facility status and operation results to the cloud operation and maintenance platform.

[0053] Figure 1 The cloud-based operations and maintenance platform shown is used to perform remote monitoring and alarms, task queue management, remote scheduling, and multi-robot collaboration. Figure 5 The cloud scheduling platform in the text refers to the part of the cloud operation and maintenance platform that is responsible for task sorting and robot scheduling. Therefore, the two refer to the overall operation and maintenance platform and the scheduling function within it, respectively.

[0054] In this embodiment, S2 includes: After the robot completes the location of the target facility, the vision acquisition module acquires images of the target electromechanical facility before cleaning.

[0055] The camera first acquires an image of the current surface of the facility, and then determines whether the current imaging conditions meet the requirements for dirt assessment based on the overall brightness and local brightness distribution in the image.

[0056] When the overall brightness of the target surface is lower than the preset acquisition range, the output intensity of the supplementary light unit is increased; when the target facility itself has the ability to emit light or the surface has strong specular reflection, the supplementary light intensity is reduced to reduce the impact of local overexposure on image texture and dirt boundary information.

[0057] After completing the supplementary lighting adjustment, the camera can capture one frame of image or capture multiple frames of image continuously.

[0058] The preferred method is to use multi-frame continuous acquisition to reduce errors caused by factors such as robot micro-vibration, changes in vehicle lighting in the tunnel, local reflection changes, and instantaneous occlusion.

[0059] After acquiring multiple frames of facility surface images, the sharpness of each image is calculated. Images with obvious motion blur, strong light occlusion, or local defocus are excluded. Among the remaining images, the image with the highest sharpness or that meets the preset image quality conditions is selected as the current evaluation image.

[0060] like Figure 3 As shown, the evaluation image first undergoes denoising, white balance, and multi-frame optimization.

[0061] The denoising process is used to reduce random noise generated in low-light environments and camera sensor noise, preventing small noise areas from being identified as minor dirt by subsequent algorithms.

[0062] White balance processing is used to reduce the overall color deviation caused by different light source color temperatures in the tunnel, thereby creating a more stable color comparison relationship between the current image and a historical clean reference image.

[0063] Multi-frame optimization further ensures that subsequent input images have the required edge, texture, and brightness information.

[0064] An image file corresponding to the facility number should be established for each target electromechanical facility. The facility image file should be associated with at least the facility number, facility location, historical assessment images, and clean baseline images.

[0065] When performing initial operations on a target facility, an image archive is created based on the facility number and linked to a pre-stored baseline image of the facility's cleanliness. Subsequent operations on the same facility can then directly retrieve the corresponding baseline data using the facility number.

[0066] Pre-stored cleanliness baseline images can be acquired after facility installation or maintenance, when the surface meets the preset cleanliness requirements, or they can be provided from existing facility maintenance records.

[0067] When the robot completes the actual cleaning and the post-cleaning state is confirmed to meet the target requirements after re-inspection, the corresponding post-cleaning image can be used as the cleanliness benchmark to update the candidate image.

[0068] For cleaned images that meet the preset baseline update conditions, they are written into the facility image archive for subsequent dirt comparison of the same facility.

[0069] therefore, Figure 3 The phrase “Write to facility image archive (update clean baseline)” indicates that the facility baseline is updated using qualified images that have been cleaned and re-verified, rather than directly overwriting the existing clean baseline after obtaining any evaluation image.

[0070] After completing the pre-cleaning image processing, an assessment image of the current target facility and a corresponding clean baseline image are generated, providing a data basis for subsequent quantitative treatment of dirt and grime.

[0071] In this embodiment, S3 includes: like Figure 3 As shown, the preprocessed evaluation image is entered into the frequency domain high-frequency component analysis, block contrast analysis, local gradient / dark spot region detection, and clean benchmark image comparison paths, respectively.

[0072] First, a frequency domain transformation is performed on the evaluation image to extract high-frequency components from different frequency ranges and determine the energy proportion of the high-frequency components.

[0073] Clean facilities typically have clear edges, text, surface textures, and local contours, which can generate a more obvious high-frequency response. When dust, oil film, water stains, or other contaminants cover the surface, they will cause edge diffusion, texture occlusion, or scattering, resulting in a corresponding decrease in high-frequency components.

[0074] Therefore, the proportion of high-frequency component energy can reflect the pollution status of the facility surface from the degree of image detail preservation.

[0075] The evaluation image is further segmented, and the contrast is calculated separately for each local region.

[0076] For uniformly distributed dust or semi-transparent oil stains, the difference in brightness in the area is usually reduced; for unevenly distributed contaminants, there will be obvious differences in contrast between different image blocks.

[0077] By performing block-based calculations, we can avoid missing local contamination areas by using only a single contrast value for the entire image.

[0078] Local gradient and dark spot detection are used to find local anomalies. By analyzing local gradient changes, blurred spots, dark spots, halos, color spots, and scattering areas, local imaging anomalies that may be caused by oil stains, water stains, or fixed contaminants can be identified.

[0079] At the same time, the corresponding clean baseline image is read from the facility image archive, and the current assessment image is compared with the baseline image according to the location of the facility surface.

[0080] When the robot's positioning is relatively stable, the working camera has a relatively fixed spatial relationship with the main area formed by the same facility. The correspondence between two images can be established by using a uniform facility surface cropping area and image size.

[0081] The grayscale, color, and local texture state of the corresponding regions are compared to form a similarity offset.

[0082] The fixed edges, mounting connectors, characters, seams, and structural textures of the facility itself are usually visible in both the clean baseline image and the current image. Therefore, the aforementioned fixed structures will not be directly judged as dirty simply because they have strong edges.

[0083] When new dust, oil film, water stains or other deposits are added to the surface of the facility, the corresponding area in the current image will have a new grayscale, color or texture deviation relative to the clean baseline, thereby further improving the distinction between the dirt judgment and the facility's own structure.

[0084] High-frequency component energy ratio, block contrast, local gradient / dark spot region detection results, and clean baseline similarity shift are all included. Figure 3 The feature fusion process is shown.

[0085] Figure 3 Feature fusion can be achieved using a weighted scoring path or a CNN classifier path.

[0086] When using a weighted scoring approach, corresponding weights are configured according to the degree of influence of different dirt indicators on the pollution status, and the processing results of multiple indicators are combined to obtain the dirt category and degree level.

[0087] If the robot's computing power is sufficient to meet the requirements of real-time deep learning inference, a processing path based on convolutional neural networks can be adopted.

[0088] Figure 3 The CNN classifier in the text represents a deep learning processing unit that uses convolutional neural networks to extract spatial features, analyze dirty areas, and subsequently determine the type and degree of contamination.

[0089] In this embodiment, the deep learning processing unit uses the improved PIDNet dirty region segmentation model as the basis for spatial analysis of dirty regions.

[0090] The improved PIDNet includes branches for detail information processing, context information processing, and boundary information processing.

[0091] The detailed information processing branch is used to maintain high spatial resolution and extract small-scale stains, particulate contamination, fine water stain edges, and local texture variations on the facility surface.

[0092] The context information processing branch is used to obtain a wider range of semantic information, enabling large areas of floating dust cover, continuous oil film and sheet-like water stains to form corresponding feature responses in a larger spatial range.

[0093] The boundary information processing branch is used to strengthen the boundary between the contaminated area and the surface of normal facilities, so that the contaminated area with irregular edges can form a more distinct spatial segmentation result.

[0094] In order to enable PIDNet to not only make judgments based on the current visual texture of the facility surface, but also to make full use of the original dirt quantification indicators, the aforementioned high-frequency component energy ratio, block contrast, local gradient / dark spot detection results and reference image similarity offset are converted into dirt quantification prior information.

[0095] Among them, data with clear spatial locations, such as local gradients, dark spot regions, and block contrast, constitute local prior feature distributions according to their regions in the original image.

[0096] Data such as the proportion of high-frequency component energy, which reflects the overall state of the image, are encoded to form global dirt quantification features.

[0097] The aforementioned prior features are jointly fused with the image features output by each processing branch of PIDNet, so that the output result is simultaneously constrained by the current image spatial features of the facility and the quantified dirt index.

[0098] After fusion, the results of the segmentation of dirty areas on the surface of the target facility are obtained.

[0099] For one or more dirty areas, the image features inside the corresponding areas are further extracted, and the regional features, together with the high-frequency component energy ratio, local contrast, local anomaly index and the difference of the reference image, are used to determine the type and degree of dirt.

[0100] Therefore, the PIDNet part mainly solves the spatial segmentation problem of "where dirt exists" on the surface of the target facility, while the type and degree determination part further solves the problem of "what kind of dirt it is and what level of pollution it reaches".

[0101] Before model training, a sample set of dirty images of electromechanical facilities in highway tunnels is established.

[0102] The sample set includes surface images obtained under different facility types, different pollution forms, and different pollution levels. The polluted areas in the images are labeled at the pixel level or at the region level, and corresponding dirt type and degree level labels are set.

[0103] During training, the model learns the differences between the facility's own structure and the real contaminated area, as well as the differences in color, texture, boundary, and spatial distribution characteristics of different contamination states.

[0104] The trained model parameters are deployed in the robot's local processing unit or in an edge processing unit that communicates with the robot.

[0105] When the robot performs actual facility cleaning tasks, it only calls the already trained model for image reasoning, without needing to retrain the model during each operation, thus meeting the real-time processing needs under continuous facility operation conditions.

[0106] The types of dirt in this embodiment may include dust, oil stains, insect glue and bird droppings, water stains and scale, and mixed types.

[0107] The degree of dirtiness can be represented by levels 0 to 4, where level 0 indicates that the surface of the facility is clean or the degree of pollution is negligible, and levels 1 to 4 gradually increase as the coverage, adhesion and impact on the function of the facility increase.

[0108] The resulting dirt type and dirt level are used for subsequent cleaning necessity assessment and parameter database lookup.

[0109] In this embodiment, S4 includes: After completing the quantitative assessment of dirt and grime, the type and level of dirt and grime corresponding to the current target electromechanical facilities are sent to the cleaning decision module.

[0110] The cleaning decision module reads the preset cleaning threshold corresponding to the current facility and compares the detected severity level with the threshold.

[0111] For different types of electromechanical facilities, the same or different cleaning thresholds can be adopted according to their functional requirements.

[0112] For example, facilities with high requirements for surface imaging quality can use relatively strict triggering conditions, while facilities whose main functions are less affected by mild surface contamination can use different cleaning triggering standards.

[0113] When the level of dirt is below the cleaning threshold, it is determined that the current facility does not require actual cleaning.

[0114] At this time, no scrubbing or spraying actions are generated; only the current facility number, assessment image, type of dirt, degree of dirt, and assessment time are saved, and the data is then processed along the designated path. Figure 2 The unclean branch shown enters the job log process.

[0115] This avoids the need for indiscriminate cleaning of all facilities on a fixed schedule, thereby reducing the need for repeated scrubbing of lightly contaminated facilities and the unnecessary consumption of water, cleaning agents, electricity, and the lifespan of the scrubbing equipment.

[0116] When the level of dirt reaches or exceeds the preset cleaning threshold, it is determined that cleaning is required, and the type and level of dirt are used as the basis for querying the parameter library.

[0117] When a facility has two or more forms of pollution, it can be classified as a mixed type of contamination. Alternatively, a corresponding cleaning process can be selected based on the main contaminated area or the type of pollution with higher cleaning requirements.

[0118] Regardless of the pollution category representation method used, whether actual cleaning is triggered is still based on the comparison between the final degree of dirtiness and the cleaning threshold.

[0119] In this embodiment, S5 includes: like Figure 4 As shown, the cleaning decision module queries a pre-established parameter library based on the current type and level of dirt.

[0120] The parameter library stores one or more sets of cleaning parameters according to different types of dirt, and further sets the parameter levels according to the degree of dirt.

[0121] The initial parameters of the parameter library can be established based on the verification results of different cleaning methods and parameter combinations, so that a lower cleaning intensity is used for light contamination and a relatively higher cleaning intensity is used for heavy contamination.

[0122] For surface dust and dirt, which is mainly characterized by loose particles covering the surface, dry brushing or air blowing can be used first, reducing the use of liquid media while still meeting cleaning requirements.

[0123] For contaminants with high adhesion, such as oil stains, insect glue, or bird droppings, wet or combined cleaning methods can be used, including atomized spraying of cleaning agents, soaking, brushing, and rinsing with clean water.

[0124] For water stains and scale, a method of first spraying to soften them and then brushing can be used to reduce the adhesion strength of the contaminants before the brushing mechanism acts on them.

[0125] For mixed-type dirt, a combination of composite parameters can be selected based on the components of the dirt and the main type of dirt.

[0126] Figure 4 The parameter library shown can also be switched according to battery status and seasonal status.

[0127] Seasonal status is used to adapt to changes in environmental humidity, dust levels, water seepage, and pollutant adhesion characteristics in tunnels under different seasons.

[0128] Battery status is used as an auxiliary parameter selection condition. When the robot's remaining battery power is within the normal range, the standard parameter combination corresponding to the type and level of dirt is invoked.

[0129] When the robot's remaining battery power is close to the task control threshold, under the condition that the current facility can achieve the basic cleaning goal, the effective parameter combination with relatively low resource consumption can be selected first. At the same time, the current battery status is uploaded to the cloud operation and maintenance platform, and the platform determines whether the robot should continue to receive new facility cleaning tasks.

[0130] Therefore, battery status will not replace the type and degree of dirt as the basis for cleaning decisions, but only serve as an auxiliary execution condition during the continuous operation of the robot.

[0131] Figure 4 The cleaning parameter combinations output by the parameter library include cleaning method, cleaning time, contact pressure, brush speed, spray volume, and lifting speed.

[0132] Cleaning time is used to determine the duration of brushing, air blowing, or other actions.

[0133] Contact pressure is used to control the pressure exerted by the brush plate or brush roller on the surface of the facility.

[0134] Brush speed and rotation speed are used to control the operating speed of the rotating brush components.

[0135] Spray volume is used to control the output of clean water or cleaning agent during a unit operation.

[0136] Lifting speed is mainly used for position adjustment of robots or cleaning actuators during longitudinal movement on facility surfaces.

[0137] For wet and complex cleaning, the cleaning agent ratio, soaking time, and number of passes can be further correlated.

[0138] Among them, the main parameters that directly affect the surface cleaning effect of the target facility are cleaning time, contact pressure, brush roller speed, spray volume, cleaning agent ratio and number of passes, while lifting speed is an auxiliary motion control parameter in the execution process.

[0139] Set corresponding parameter levels for different severity levels.

[0140] When increased cleaning intensity is required, the corresponding cleaning time should be increased first, so that the cleaning execution agency can improve the degree of pollution removal without significantly increasing the mechanical force on the facility surface.

[0141] If the cleaning time still fails to meet the requirements, increase the contact pressure setting.

[0142] This parameter adjustment sequence can reduce the likelihood of surface wear caused by directly applying high contact pressure under conditions of mild or moderate pollution.

[0143] After completing parameter matching, the cleaning method and cleaning parameters are compiled into a dual-arm cleaning operation instruction.

[0144] The work instructions include at least the cleaning actions that the first and second cleaning robotic arms need to perform, the corresponding parameters, and the start-stop sequence between the two robotic arms.

[0145] When a cleaning task arises from a re-inspection after cleaning, it also receives the actual execution parameters and parameter upgrade information from the previous round, adjusts the parameter levels based on the current parameter combination, and then generates a new cleaning operation instruction to avoid reverting to the original default parameters every time a cleaning is performed.

[0146] In this embodiment, S6 includes: After receiving the dual-arm cleaning operation command, the main control and communication module controls the first and second cleaning robotic arms to enter the corresponding operation state.

[0147] Before cleaning is performed, a small-scale pose correction can be made based on the robot's current parking position and the positional relationship between the robot and the surface of the target facility.

[0148] When the longitudinal position needs to be changed, the robot is controlled to make small movements based on the current auxiliary motion parameters; when the circumferential working direction needs to be changed, the robot's orientation is adjusted through the rotary structure.

[0149] When using dry cleaning, a cleaning robotic arm equipped with brushes or rollers is gradually moved to the surface of the facility.

[0150] After the brushing mechanism comes into contact with the surface of the facility, it uses a universal connection structure to adaptively deflect according to the curvature and local direction of the target surface, so that the brush plate or brush roller fits the target area more evenly.

[0151] Once the contact pressure required by the current work instruction is reached, the brushing mechanism is controlled to run at the corresponding brush roller speed and the cleaning time is continuously set.

[0152] For loose dust with low adhesion, an air blowing device can be used to blow air into the target area before, during, or after brushing to remove loose particles from the facility surface.

[0153] When using wet cleaning, the first cleaning robotic arm can move the spraying component to the target contaminated area and output clean water or cleaning agent according to the set spray volume.

[0154] For contaminants that require soaking and softening, maintain a preset soaking time after the cleaning agent spraying is completed to reduce the adhesion between oil, fixed contaminants, or other highly adhesive contaminants and the facility surface.

[0155] After soaking, the second cleaning robotic arm moves the brush plate or brush roller close to the target area and performs brushing according to the corresponding contact pressure, brush roller speed and cleaning time.

[0156] After scrubbing, the surface of the facility can be rinsed again with clean water through the spray system to remove the cleaning agent and loose contaminants.

[0157] Subsequently, the air blowing or squeegee assembly is activated as needed to reduce the impact of residual liquid on the facility surface on subsequent re-inspection images and normal facility operation.

[0158] For complex cleaning methods that require multiple actions to be coordinated continuously, the two cleaning robotic arms enter the target area sequentially according to the start and stop sequence in the operation instructions.

[0159] For example, the first cleaning robotic arm sprays cleaning agent and exits the corresponding area after the immersion time is reached; the second cleaning robotic arm then enters the same area to perform brushing; after brushing, the first robotic arm or the corresponding execution end performs rinsing.

[0160] For large facility surfaces, the surface can be divided into two or more working areas, with the first cleaning robot arm and the second cleaning robot arm handling different areas respectively.

[0161] During parallel or alternating operation of the two robotic arms, the main control unit continuously maintains the movement sequence and effective working space between them, so that the two cleaning actuators avoid structural interference in the same local space.

[0162] After the two robotic arms complete the cleaning actions and number of passes specified in the current work instruction, they stop brushing, spraying, and other actions, and control the cleaning execution end to move out of the main imaging range of the work camera to avoid the cleaning mechanism from obstructing the subsequent re-inspection images.

[0163] In this embodiment, S7 includes: After the cleaning process is completed, the visual acquisition module acquires images of the same target electromechanical facility surface again.

[0164] To improve the comparability between the pre- and post-evaluation results, the post-cleaning images should preferably use the same or corresponding robot parking position, camera angle, and lighting conditions as the pre-cleaning images.

[0165] After obtaining the cleaned image, the image preprocessing, high-frequency component analysis, block contrast analysis, local gradient / dark spot detection, clean baseline image comparison, and deep learning feature fusion are re-performed as before cleaning.

[0166] The degree of dirtiness after cleaning is obtained through the above process and compared with the target level corresponding to the current facility.

[0167] If the cleaning level is equal to or lower than the target level after cleaning, the cleaning is considered to have met the requirements.

[0168] Write the images before cleaning, the images after cleaning, the degree of dirt before and after cleaning, the actual cleaning method and parameter combination used, and the current re-inspection results into the corresponding facility operation record.

[0169] If the cleaning level is still higher than the target level after cleaning, it indicates that the current parameter combination is insufficient to achieve the intended cleaning effect, and parameter upgrade information needs to be generated based on the completed cleaning process.

[0170] First, determine whether the current facility has reached the set limit for the number of cleaning cycles performed.

[0171] If there is still room for further cleaning, prioritize increasing the current cleaning time setting.

[0172] When the cleaning time has reached the allowable range for the corresponding setting, or historical data shows that continuing to increase the cleaning time will not significantly improve the effect, then increase the contact pressure.

[0173] If the re-examination images show that the original cleaning method cannot effectively deal with the current deposits, the cleaning method can be switched, for example, from dry method to wet or combined method.

[0174] Based on the parameter upgrade results, the dual-arm cleaning operation instructions are regenerated, and cleaning and post-cleaning image acquisition are performed again.

[0175] The new re-inspection results continue to be compared with the target level until the requirements are met or the number of cleaning cycles reaches the set limit.

[0176] If the target requirements are still not met after reaching the maximum number of cleaning cycles, the cleaning intensity will no longer be increased, and the corresponding facility will be marked as an abnormal facility.

[0177] Anomaly records can include the target facility number, facility image, initial dirt level, actual cleaning parameters for each cleaning session, final level, and number of re-cleaning sessions. These records are uploaded to the cloud-based operations and maintenance platform for subsequent manual assessment to determine whether there are any abnormal conditions on the facility surface, such as paint residue, adhesive residue, structural damage, or other defects that are difficult to remove through routine cleaning processes.

[0178] The retest results also serve as Figure 4 The data source for the online iteration of the parameter library shown.

[0179] The system records whether the target level is reached directly after the first cleaning operation, based on the type and degree of dirt and the actual combination of cleaning parameters. It also calculates the first pass rate for the corresponding parameter combination based on the accumulated records.

[0180] When the pass rate of a certain parameter combination is lower than the preset requirement, it indicates that the current parameter is generally too low in terms of cleaning intensity for the corresponding pollution state. The cleaning time and contact pressure level are then increased, and the correction results are rewritten into the parameter library.

[0181] When a certain combination of parameters maintains a high first pass rate and its cleaning results are significantly better than the target requirements in the long term, it can be considered that the current cleaning intensity has a certain redundancy, and the cleaning time, contact pressure or other corresponding parameters can be appropriately reduced.

[0182] Figure 4 The requirement of achieving a 100% pass rate and significantly exceeding the target is used as a strict indicator for parameter reduction. In actual implementation, corresponding high pass rate judgment conditions can be set according to the cumulative number of operations and facility maintenance requirements.

[0183] When the pass rate is within the normal range, and the conditions for adjusting parameters are neither met nor the conditions for adjusting parameters are met, the current parameter combination remains unchanged, and the results of subsequent operations are used for statistical analysis.

[0184] Through the above processing, the parameter library has three corresponding states: parameter increase, parameter retention, and parameter decrease, thus avoiding the problem of the re-examination result being in an intermediate state without a subsequent processing path.

[0185] In this embodiment, S8 includes: For target electromechanical facilities that are determined not to require cleaning during the cleaning necessity assessment process, only facility assessment records are generated.

[0186] The assessment record should include at least the facility number, facility location, assessment images, type of soiling, level of soiling, and assessment time.

[0187] Since these facilities do not actually perform cleaning, it is not required to generate post-cleaning images, actual cleaning parameters, or re-cleaning data.

[0188] For target electromechanical facilities that have already undergone cleaning processes, generate cleaning operation records.

[0189] Cleaning operation records should include at least the facility number, facility location, pre-cleaning image, post-cleaning image, type of dirt, initial level of dirt, post-cleaning level of dirt, actual cleaning parameters used, number of re-cleaning operations, and final re-inspection results.

[0190] Both the aforementioned assessment records and cleaning operation records can be uploaded to the main control and communication module. Figure 1 The cloud-based operations and maintenance platform shown.

[0191] The cloud-based operation and maintenance platform establishes facility-level status files based on facility number and spatial location information, forming data on the dirt and grime status of facilities along the route by recording the degree of dirt and grime and historical maintenance information for different facilities.

[0192] like Figure 5 As shown, the cloud scheduling function within the cloud-based operations and maintenance platform generates a dirt map and task queue based on the facility status.

[0193] Figure 5 #1 to #7 in the diagram represent multiple facility nodes distributed sequentially along the work area, and different markings are used to indicate different levels of dirtiness.

[0194] In this embodiment, the degree of complete dirtiness is divided into 0 to 4 levels, where level 0 indicates that the facility is in a clean state or the degree of pollution is negligible.

[0195] Figure 5 It is mainly used to indicate the different facility statuses that need to participate in the current inspection and cleaning task sequence. Therefore, the diagram uses levels 1 to 4 for illustration. Level 0 facilities can only be retained in the facility image file and historical evaluation record, and do not enter the current cleaning task queue separately.

[0196] For Level 1 facilities, only assessment records can be generated based on the actual cleaning thresholds set.

[0197] Facilities that reach the cleaning threshold at levels 2, 3, and 4 are then placed in the subsequent task queue, with the work sequence determined based on the degree of soiling, facility functional priority, and robot position.

[0198] The cloud-based operations and maintenance platform also obtains the robot's current location, remaining battery power, and current task status.

[0199] When multiple robots are present, tasks can be assigned based on the distance between each robot and the target facility, the robot's current remaining battery power, and the priority of the target facility.

[0200] Figure 5 The robots A and B shown represent multiple robots performing facility cleaning and task succession respectively.

[0201] When robot A is close to a facility that needs priority processing and has enough remaining power to meet the operational requirements, robot A can perform the corresponding task first; robot B will continue to process its area or the next priority facility.

[0202] When one of the robots reaches the return-to-home threshold with its remaining battery power, its unfinished cleaning tasks are re-entered into the task queue and then taken over by another robot that meets the distance and battery power requirements.

[0203] Once the current facility's record is completed and uploaded, the robot obtains the task information for the next target facility from the cloud-based operations and maintenance platform.

[0204] When there are subsequent tasks, the robot moves to the next target facility and performs facility approach, visual data acquisition, dirt assessment, cleanliness judgment, and subsequent processing again.

[0205] When the current task queue is completed, or when the robot's remaining battery power reaches the return-to-home condition, it will no longer accept new cleaning tasks and will return to the standby position.

[0206] Therefore, this implementation method forms a continuous processing process consisting of target facility proximity positioning, pre-cleaning visual acquisition, dirty area segmentation and quantitative assessment, cleaning necessity judgment, cleaning method and parameter matching, dual robotic arm autonomous execution, post-cleaning closed-loop re-inspection, online parameter correction, and multi-facility task scheduling.

[0207] In the aforementioned continuous processing, the data generated in the previous process serves as explicit input for the subsequent process: facility positioning results are used to determine the visual acquisition location; pre-cleaning images and clean baseline images are used to generate the type and degree of dirt; dirt assessment results are used to determine whether cleaning is required and to query the parameter library; parameter query results are used to generate dual-arm operation instructions; cleaning execution results are used to form a re-inspection result through post-cleaning images; the re-inspection result is further used for supplementary cleaning parameter upgrades and parameter library corrections; after the current facility is processed, the task queue formed in the cloud is used to determine the next target facility, thereby ensuring the continuity of data acquisition, decision-making, execution, feedback, and scheduling relationships in the robot's autonomous cleaning process.

[0208] Example 1: To verify the feasibility of this invention in practice, it was applied to the autonomous cleaning operation of electromechanical facilities in a highway tunnel. Lighting fixtures, surveillance cameras, and electromechanical facilities with varying degrees of surface contamination within the tunnel were selected as the targets. A dual-arm robot approached the target electromechanical facilities using a climbing and walking module. After reaching the preset working position, it used a pole-mounted environmental camera to identify the location of the target facilities and adjusted the robot's posture so that the working camera faced the light-receiving / imaging surface of the target electromechanical facilities. The working end of the first cleaning robot arm was equipped with a brush plate or brush roller, and the working end of the second cleaning robot arm was equipped with a spray component and an air blowing component. The two robots executed corresponding cleaning actions according to the work instructions generated by the cleaning decision module.

[0209] During the cleaning of lighting fixtures, the supplementary lighting unit illuminates the light-transmitting surface of the fixtures according to a preset brightness curve. The working camera continuously acquires multiple frames of surface images, and selects the image with sufficient clarity as the evaluation image. The dirt assessment module calculates that the high-frequency component energy ratio of the evaluation image is approximately 62% of that of the clean state baseline image, the RMS contrast decreases by 35% compared to the clean state, and the proportion of local distortion areas is 8%. After fusing the above dirt quantification indicators with the detailed features, spatial semantic features, and boundary features extracted by the improved PIDNet dirt region segmentation model, the surface of the lighting fixture is determined to be dust-type dirt, with a dirt level of 2, reaching the preset cleaning threshold.

[0210] The cleaning decision module queries a pre-established parameter library based on the type of dust and level 2 dirt, matches it with a dry brushing method, and determines the brush roller contact pressure to be 80 N, the brush roller speed to be 200 r / min, the single-pass cleaning time to be 20 s, and the number of passes to be 2. During the cleaning process, the first cleaning robotic arm drives the brush roller to complete the main brushing action, while the second cleaning robotic arm maintains a avoidance position and provides auxiliary air blowing according to the work instructions. After completing the set number of passes, the work camera re-captures images of the same lamp surface, and the re-inspection shows that the dirt level has dropped to level 0, reaching the target level. The system generates a cleaning operation record that includes before and after cleaning images, dirt assessment data, actual cleaning parameters, facility number, and operation time.

[0211] During the cleaning of the surveillance camera, the robot approaches the camera lens to provide supplemental lighting and acquire images. A quantitative assessment of dirt indicates the presence of shellac-like deposits on the lens, classifying the dirt level as Level 3. The cleaning decision module matches a wet composite cleaning method and generates a dual-arm cleaning command: the second cleaning arm first sprays and soaks the lens with a 3% surfactant-based atomized cleaning agent for 10 seconds; then, the first cleaning arm controls a soft brush roller to brush for 30 seconds at a contact pressure of 120 N and a rotation speed of 150 r / min; finally, the second cleaning arm performs atomized water spraying and air blowing to remove water. A post-cleaning inspection shows the dirt level has decreased to Level 1, reaching the target level. The system then terminates the current facility cleaning and records the parameter combinations and inspection results.

[0212] For heavily oil-contaminated facilities, the dirt assessment module determines the dirt level to be Level 4. The cleaning decision module executes cleaning agent immersion, brushing, and spraying according to the high-level parameters in the parameter library. If the re-inspection result after the first cleaning is still Level 2, failing to meet the target level, the re-inspection module generates parameter upgrade information based on the current parameter combination, prioritizing extending the cleaning time by 50% and increasing the contact pressure by one parameter level before feeding it back to the cleaning decision module. The system then generates a re-cleaning operation instruction and executes cleaning again. If the second re-inspection is still Level 2, because the number of re-cleaning attempts has reached the preset limit, the facility is marked as an abnormal facility and reported along with before-and-after images and re-inspection data for subsequent manual verification of abnormal adhesions such as surface scratches, paint spots, or adhesive residue.

[0213] After the robot completes the processing of each target facility, it writes the assessment images and dirt assessment data of facilities that do not require cleaning, as well as the before-and-after images, actual operation parameters, and re-inspection results of the cleaned facilities, into the corresponding operation records. The main control and communication modules upload the operation records to the cloud scheduling platform. Subsequently, based on the facility's dirt level, facility priority, robot's current location, and remaining battery power, the system can generate a task queue and operation sequence for the next target facility. At the same time, the system calculates the first-pass rate of re-inspection for different dirt types and parameter combinations based on the cumulative cleaning operation records, corrects the cleaning time and contact pressure parameter levels, and writes the corrected parameter combinations back to the parameter library for subsequent cleaning of similar facilities.

[0214] Table 1 Verification Results of Self-Cleaning of Electromechanical Facilities in Highway Tunnels

[0215] Table 1 shows that the 10 samples of electromechanical facilities in highway tunnels cover dust, oil stains, water stains, solidified dirt, and mixed types of dirt, and include processing results such as no cleaning required, one-time cleaning meeting the standard, re-cleaning meeting the standard, and abnormal reporting. Among them, the dirt level of lamp B is only level 1, which is below the cleaning threshold. The system directly records and skips cleaning when it is below the cleaning threshold. For the other facilities that reach the threshold, the corresponding cleaning method is matched according to the type and degree of dirt.

[0216] For Level 2 floating dust, dry brushing or dry brushing with air blowing can reduce the re-inspection level to Level 0. For Level 2 to 3 water stains, oil stains, and solidified dirt, spray brushing or a wet combined method can reduce the re-inspection level to Level 0 or Level 1. Level 4 oil stains can reach Level 1 after parameter upgrades and touch-up cleaning, indicating that the re-inspection results can continue to be used as input for adjusting cleaning time, contact pressure, and touch-up cleaning operations.

[0217] The hybrid-type Level 4 protective cover remains at Level 2 after composite cleaning. Once the maximum number of cleaning cycles is reached, it is marked as abnormal and reported to prevent further increases in mechanical stress. Overall results show that this invention can form a continuous data chain encompassing quantitative assessment of dirt and grime, determination of cleaning necessity, parameter matching, dual-arm execution, re-inspection and cleaning, and abnormal handling. Furthermore, the parameter library can be corrected using accumulated operational parameters and re-inspection results.

[0218] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for autonomous cleaning of electromechanical facilities in highway tunnels using a dual-arm robot, characterized in that, Includes the following steps: S1. The dual-arm robot approaches the target electromechanical facility, identifies the location of the target electromechanical facility through the environmental camera, and adjusts the robot's posture so that the working camera is aimed at the light-receiving / imaging surface of the target electromechanical facility. S2. The target electromechanical facilities are illuminated by the supplementary lighting unit. One or more frames of surface images of the facilities are captured by the working camera. Images that meet the clarity requirements are selected as evaluation images and associated with the facility image archive. S3. Extract the high-frequency component energy ratio, contrast index, local distortion index and color / grayscale shift index of the evaluation image, and perform fusion processing through the improved PIDNet dirty area segmentation model to obtain the dirty type and dirty degree level. S4. Compare the level of dirtiness with the preset cleaning threshold. If the level is lower than the preset cleaning threshold, record the evaluation result and execute S8. If the level reaches or exceeds the preset cleaning threshold, execute S5. S5. Query the pre-established parameter library according to the type and level of dirt, match the corresponding cleaning method and cleaning parameters, and generate dual-arm cleaning operation instructions based on the matching results. S6. Adjust the robot's position so that the cleaning execution module approaches the target electromechanical facility, and the first and second cleaning robotic arms execute the corresponding cleaning actions according to the dual-arm cleaning operation instructions. S7. Collect images of the target electromechanical facility surface again and conduct a quantitative assessment of dirtiness. Compare the level of dirtiness after cleaning with the target level. If the target level is not reached, generate parameter upgrade information and perform re-cleaning. S8. Generate and upload corresponding work records based on whether the target electromechanical facilities have been cleaned. After completing the processing of the current target electromechanical facilities, move to the next target electromechanical facilities or return to the standby position.

2. The autonomous cleaning method for dual-arm robots of electromechanical facilities in highway tunnels according to claim 1, characterized in that, S1 includes the following steps: S11. Determine the location of the target electromechanical facility based on the task information, and move along the corresponding pole of the target electromechanical facility to the preset working height using the climbing and walking module; S12. Collect images of the target electromechanical facilities and their surrounding environment using the pole-mounted environmental camera, and identify the spatial orientation of the target electromechanical facilities based on the environmental images. S13. Control the climbing and walking module to rise and fall according to the spatial orientation of the target electromechanical facilities, and adjust the robot's circumferential working position through the pole-turning mechanism; S14. Based on the adjusted robot pose, orient the working camera toward the target electromechanical facility's light-receiving / imaging surface and maintain a working position that meets the requirements for image acquisition and cleaning execution.

3. The autonomous cleaning method for dual-arm robots of electromechanical facilities in highway tunnels according to claim 1, characterized in that, S2 includes the following steps: S21. Control the supplementary lighting unit to supplement light according to the preset brightness curve based on the current brightness of the target electromechanical facility's light-receiving / imaging surface; S22. Control the operation camera to continuously acquire one or more frames of facility surface images, and evaluate the sharpness of each acquired frame of facility surface image. S23. Select the facility surface image with the highest clarity or that meets the preset clarity requirement as the evaluation image, and associate it with the corresponding facility number and acquisition time. S24. During the initial operation, establish a facility image file and associate it with a pre-stored clean status baseline image. Subsequently, retrieve the corresponding clean status baseline image based on the facility number.

4. The autonomous cleaning method for dual-arm robot of electromechanical facilities in highway tunnels according to claim 1, characterized in that, S3 includes the following steps: S31. Preprocess and frequency domain transform the evaluation image, calculate the proportion of high-frequency sub-band energy to the total frequency band energy, and obtain the high-frequency component energy ratio. S32. Calculate the contrast and local distortion index of the evaluation image, and compare the evaluation image with the clean state reference image to obtain the color / grayscale shift index. S33. Input the evaluation image into the improved PIDNet dirty region segmentation model, and convert each dirty quantification index into dirty prior features, and fuse them with the multi-branch image features extracted by the model. S34. Obtain the dirty area segmentation result based on the fused features, and determine the dirty type and corresponding dirty degree level based on the dirty area features.

5. A dual-arm robot autonomous cleaning method for electromechanical facilities in highway tunnels according to claim 1, characterized in that, S4 includes the following steps: S41. Receive the type and level of dirt obtained by S3, and read the preset cleaning threshold corresponding to the target electromechanical facility. S42. Compare the level of dirtiness with the preset cleaning threshold. If the level of dirtiness is lower than the preset cleaning threshold, it is determined that cleaning is not required. S43. For target electromechanical facilities that do not require cleaning, save the assessment images, dirt type, dirt level and assessment time, and directly execute S8; S44. When the level of dirt reaches or exceeds the preset cleaning threshold, it is determined that cleaning is required, and the type of dirt and the level of dirt are sent to the cleaning decision module.

6. A dual-arm robot autonomous cleaning method for electromechanical facilities in highway tunnels according to claim 1, characterized in that, S5 includes the following steps: S51. Based on the type of dirt, query the pre-established parameter database to determine the dry, wet, or combined cleaning method corresponding to the type of dirt. S52. Match cleaning time, contact pressure, brush roller speed, spray volume, cleaning agent ratio, and number of passes according to the degree of dirtiness; S53. Adjust the cleaning parameters according to the corresponding parameter level based on the degree of dirtiness, with priority given to adjusting the cleaning time, and then adjusting the contact pressure between the brush roller and the facility surface. S54. Generate a work instruction that includes double-arm cleaning actions, corresponding parameters, and start / stop sequence based on the cleaning method and cleaning parameters; when receiving parameter upgrade information, adjust the current parameter level according to the parameter upgrade information and generate a re-cleaning work instruction.

7. A method for autonomous cleaning of electromechanical facilities in highway tunnels using a dual-arm robot according to claim 1, characterized in that, S6 includes the following steps: S61. Adjust the robot's working posture through the climbing and walking module so that the cleaning execution ends of the first and second cleaning robotic arms approach the surface of the target electromechanical facility. S62. Control the first and second cleaning robotic arms to call one or more of the following components according to the work instructions: brush plate or brush roller, spray component, air blowing component and squeegee component. S63. Control the first cleaning robotic arm and the second cleaning robotic arm to operate in coordination according to the corresponding cleaning actions and start-stop sequence, and execute the matching cleaning parameters; S64. The brush plate or brush roller is made to adhere to the surface of the target electromechanical facility through the universal connection structure, and cleaning is stopped after the set cleaning time and number of operations are completed.

8. A method for autonomous cleaning of electromechanical facilities in highway tunnels using a dual-arm robot according to claim 1, characterized in that, S7 includes the following steps: S71. After cleaning, collect images of the same target electromechanical facility surface again, and obtain the degree of dirtiness after cleaning according to the dirt quantification assessment method in S3. S72. Compare the level of dirtiness after cleaning with the preset target level. If it is equal to or lower than the target level, the cleaning is deemed qualified. Record the actual parameter combination and the re-inspection results. S73. When the level of dirtiness after cleaning is higher than the target level, parameter upgrade information is generated based on the current parameter combination and fed back to the cleaning decision module to generate a re-cleaning operation instruction. S74. Repeat cleaning and re-inspection according to the cleaning operation instructions. End the cleaning operation when the target level is reached. If the target level is not met after reaching the maximum number of cleaning operations, mark the abnormality and report it.

9. A method for autonomous cleaning of electromechanical facilities in highway tunnels using a dual-arm robot according to claim 1, characterized in that, S8 includes the following steps: S81. For target electromechanical facilities that do not require cleaning, write the facility number, facility location, assessment image, dirt assessment data and assessment time into the corresponding work record; S82. For the target electromechanical facilities to be cleaned, write the before and after cleaning images, actual cleaning parameters, dirt assessment data and re-inspection results into the corresponding work record. S83. Calculate the first pass rate of different parameter combinations based on the cumulative operation records, adjust the cleaning time and contact pressure level according to the first pass rate and update the parameter library; S84. Upload the corresponding job record, and the cloud-based operation and maintenance platform will generate a subsequent task queue and job sequence, and control the robot to move to the next target electromechanical facility or return to the standby position.

10. A dual-arm robot autonomous cleaning system for electromechanical facilities in highway tunnels, characterized in that, A dual-arm robot autonomous cleaning method for electromechanical facilities in highway tunnels, as described in any one of claims 1 to 9, is characterized by comprising: The climbing and walking module is used to lift, stop, and adjust the working position of the dual-arm robot relative to the target electromechanical facility along the pole. The visual acquisition module includes a working camera and a supplementary lighting unit, used to acquire images of the target electromechanical facility's illuminated / imaging surface before and after cleaning; The dirt assessment module is used to extract the dirt quantification index of facility surface images and determine the dirt type and dirt level through the improved PIDNet dirt area segmentation model. The parameter library is used to store corresponding cleaning methods and cleaning parameters according to the type and degree of dirt, and to adjust the corresponding parameter levels based on the cumulative re-inspection results; The cleaning decision module is used to determine the necessity of cleaning based on the level of dirt, and to match cleaning methods and cleaning parameters from the parameter library according to the type and level of dirt to generate dual-arm cleaning operation instructions; The cleaning execution module includes a first cleaning robotic arm and a second cleaning robotic arm, as well as various components such as a brushing mechanism, a spraying component, an air blowing component, and a squeegee component disposed at the working ends of the two cleaning robotic arms. The re-inspection module is used to conduct a quantitative assessment of the dirt and grime of the target electromechanical facilities after cleaning, trigger additional cleaning based on the re-inspection results, and generate abnormal information for target electromechanical facilities that still fail to meet the standards after reaching the maximum number of additional cleaning cycles. The main control and communication module is connected to the climbing and walking module, vision acquisition module, dirt assessment module, cleaning decision module, parameter library, cleaning execution module and re-inspection module respectively, and is used to coordinate robot positioning, image acquisition, cleaning decision, dual-arm cleaning, closed-loop re-inspection and operation record uploading.

Images