High-altitude glass curtain wall cleaning equipment intelligent control system based on internet of things

Abstract

This invention relates to the field of intelligent cleaning control technology and discloses an intelligent control system for high-altitude glass curtain wall cleaning equipment based on the Internet of Things. The edge controller performs the following processes: synchronously acquiring and fusing data from a nine-axis IMU, a negative pressure sensor, an ultrasonic ranging sensor, a vision camera, a wind speed, temperature and humidity sensor, and an infrared frame recognition sensor; generating a full-coverage cleaning path based on curtain wall grid modeling and an improved A* search algorithm; identifying stain levels using a stain classification convolutional neural network and generating adaptive cleaning control parameters; performing closed-loop negative pressure regulation of the vacuum pump based on attitude deviation and wind speed and executing multiple safety protection responses in parallel; and uploading operation data to a cloud server via a wireless communication module to support remote monitoring.

Description

Technical Field

[0001] This invention relates to an intelligent control system for autonomous operation control of high-altitude glass curtain wall cleaning equipment. Background Technology

[0002] For a long time, high-altitude glass curtain wall cleaning operations have relied on manual suspension or semi-automatic window cleaning equipment. In the context of super high-rise buildings, the working environment is affected by a combination of wind disturbances, water stains on the curtain wall surface, and deviations in glass flatness, which places high demands on the equipment's adsorption reliability and motion control precision.

[0003] Existing equipment typically uses a single vacuum adsorption method for high-altitude fixation, with the adsorption force set to a fixed value. It lacks real-time adjustment capabilities to adapt to changes in posture and wind force, and also lacks emergency braking functionality for exceeding posture limits. Regarding path planning, existing equipment largely relies on manual remote control operation and lacks automatic path planning capabilities tailored to curtain wall dimensions, window frame layouts, and obstacle distribution. In terms of cleaning parameter control, existing equipment employs a crude approach with uniform water flow and brushing intensity, applying the same cleaning force indiscriminately to stains of different types and severity.

[0004] The limitations of the existing technology lead to the following technical problems: In ultra-high-rise environments, the equipment may slip and detach due to insufficient negative pressure or loss of posture, posing safety hazards; manual remote control operation results in missed areas, repeated cleaning, and boundary collisions in the cleaning trajectory, leading to low cleaning coverage; and the coarse cleaning parameter mode causes excessive consumption of water and electricity in lightly soiled areas, while the cleaning intensity is insufficient in heavily soiled areas, resulting in poor uniformity of cleaning quality. This invention proposes corresponding technical solutions to address these problems. Summary of the Invention

[0005] This invention proposes measures for intelligent cleaning control of high-altitude glass curtain walls.

[0006] According to the present invention, this is achieved by having an edge controller perform unified timestamp alignment and fusion calibration on the multi-source sensor sensing data, and on this basis complete a complete control process including grid modeling and full-coverage path generation, stain level identification based on convolutional neural network, adaptive cleaning control parameter generation, and closed-loop negative pressure regulation and multiple safety protection responses.

[0007] According to the present invention, there are mutual constraints among the time inconsistencies of multi-sensor data, the dynamic adaptive control requirements for complex stain conditions, and the safety assurance of high-altitude adsorption in high-altitude glass curtain wall cleaning operations. These need to be coordinated and resolved through a unified perception fusion and closed-loop control system.

[0008] This invention discloses a computer-implemented method for intelligent cleaning control of high-altitude glass curtain walls. The high-altitude glass curtain wall cleaning equipment is equipped with an inertial measurement unit, a negative pressure sensor, an ultrasonic ranging sensor, a visual camera, a wind speed, temperature and humidity sensor, and an infrared frame recognition sensor. An edge controller synchronously acquires the sensor data and aligns it with a unified timestamp to form a synchronous sensing dataset. Based on the curtain wall parameter data, grid modeling is performed on the area of ​​the curtain wall to be cleaned, and a path search algorithm oriented towards full coverage is used to generate a full-coverage cleaning path that traverses all passable grid cells. Based on the image frames in the synchronous sensing dataset, a pre-trained stain classification convolutional neural network is used to output stain level labels for the current cleaning area. Based on the stain level labels and environmental parameters, an adaptive cleaning control parameter set including water flow instructions, cleaning brush rotation speed instructions, and travel speed instructions is generated. Based on attitude data, real-time negative pressure values, and wind speed values, closed-loop negative pressure regulation control is performed on the vacuum pump, and protective responses are executed according to multiple safety conditions.

[0009] In a preferred embodiment of the present invention, the formation of the synchronous sensing dataset further includes a fusion calibration process: performing complementary filtering calibration on each component of the attitude data vector, wherein the inputs to the complementary filtering are the low-frequency attitude estimation value output by the accelerometer and the high-frequency attitude estimation value output by the gyroscope, and the output is a calibrated attitude data vector; performing median filtering on the distance dataset within a preset time sliding window to obtain a denoised distance dataset; and uniformly encapsulating the calibrated attitude data vector, real-time negative pressure value, denoised distance dataset, image frame, environmental parameter vector, and window frame boundary detection signal into a fusion sensing data packet, which is then passed to subsequent steps as input for control decisions.

[0010] In a preferred embodiment of the present invention, the cost function of the path search algorithm for full coverage is defined as: the actual movement cost from the path start point to the current grid node, the heuristically estimated cost from the current grid node to complete full coverage, and the number of currently remaining unvisited passable grid cells weighted by the coverage penalty weight coefficient; wherein the heuristically estimated cost is calculated by multiplying the number of currently unvisited grid cells by the unit movement cost; during each step of the path search process, adjacent grid nodes that can maintain a continuous reciprocating motion direction are preferentially selected, so that the generated full coverage sweep path tends to a serpentine reciprocating traversal pattern.

[0011] In a preferred embodiment of the present invention, the grid modeling includes: dividing the curtain wall area to be cleaned into a two-dimensional grid map according to a preset grid unit size; grid units in the two-dimensional grid map that spatially overlap with the set of window frame position coordinates and the set of obstacle distribution areas are marked as impassable, and the remaining grid units are marked as passable; during the cleaning process of the high-altitude glass curtain wall cleaning equipment along the full-coverage cleaning path, the edge controller continuously writes the index of the grid units that have been cleaned to a non-volatile memory; when the high-altitude glass curtain wall cleaning equipment is interrupted and restarted, the set of completed grid unit indexes is read from the non-volatile memory, and the subsequence of grid units that have not yet been cleaned in the full-coverage cleaning path is used as the continuing path.

[0012] In a preferred embodiment of the present invention, the step of outputting stain level labels using a pre-trained stain classification convolutional neural network includes: sequentially performing adaptive histogram equalization, size normalization, and pixel value normalization on image frames to obtain a standardized image; inputting the standardized image into the stain classification convolutional neural network, extracting features through several convolutional and pooling layers, mapping the extracted features to scores for each stain level category by a fully connected layer, and outputting a probability distribution vector corresponding to each preset stain level using a softmax activation function; and taking the level label corresponding to the category with the highest probability value in the probability distribution vector as the stain level label for the current cleaning area.

[0013] In a preferred embodiment of the present invention, generating the adaptive cleaning control parameter set includes: querying the baseline water spray flow rate, baseline cleaning brush rotation speed, and baseline travel speed corresponding to the stain level label from a cleaning parameter mapping table pre-stored in the edge controller, wherein the higher the stain level, the larger the baseline water spray flow rate and baseline cleaning brush rotation speed, and the smaller the corresponding baseline travel speed; calculating an environmental correction factor based on the ratio of wind speed value to reference wind speed value and the deviation ratio of temperature value to reference temperature value, using preset wind speed correction coefficients and temperature correction coefficients as weights, wherein the environmental correction factor is a dimensionless scalar obtained by superimposing the wind speed ratio term and temperature deviation ratio term on the baseline value; multiplying the baseline water spray flow rate and baseline cleaning brush rotation speed by the environmental correction factor to obtain the water spray flow rate command and cleaning brush rotation speed command, respectively, and dividing the baseline travel speed by the environmental correction factor to obtain the travel speed command.

[0014] In a preferred embodiment of the present invention, the generation of the adaptive cleaning control parameter set further includes obstacle deceleration correction: when any distance value in the denoised distance dataset is less than a preset obstacle distance threshold or when the window frame boundary detection signal is triggered, the travel speed command is multiplied by the smaller of the ratio of the current minimum obstacle distance value to the obstacle distance threshold and the value one to obtain the corrected travel speed command; at the same time, the edge controller queries the pre-stored cleaning solution ratio table according to the stain level label to obtain the mixing volume ratio of clean water and cleaning solution corresponding to the current stain level, and sends the mixing volume ratio as a control command to the automatic mixing valve.

[0015] In a preferred embodiment of the present invention, the closed-loop negative pressure regulation control includes: calculating a comprehensive tilt deviation angle based on calibrated pitch and roll angles; using a pre-stored reference negative pressure value as a basis, superimposing a comprehensive tilt deviation angle weighted by a tilt angle correction coefficient and a wind speed value weighted by a wind speed correction coefficient to obtain a target negative pressure value; calculating the deviation between the target negative pressure value and the real-time negative pressure value; using the deviation as input, generating a vacuum pump speed regulation amount through a PID control algorithm and outputting it to the vacuum pump drive interface.

[0016] In a preferred embodiment of the present invention, the multiple safety conditions are continuously monitored in parallel and have higher priority than the cleaning operation control commands, including: when the real-time negative pressure value is lower than a preset safety negative pressure threshold, an audible and visual alarm trigger signal is output and a walking motor lock command is output; when the comprehensive tilt deviation angle exceeds a preset attitude safety threshold, a shutdown command for all execution components and a walking motor lock command are output; when the wind speed value exceeds a preset operating wind speed threshold, a complete cleaning operation stop command is output to put the high-altitude glass curtain wall cleaning equipment into a standby state, while maintaining continuous operation of the vacuum pump; when the main power supply voltage drops below a preset undervoltage threshold, the system automatically switches to a backup energy storage module that supplies power only to the vacuum pump to maintain the negative pressure adsorption state.

[0017] This invention discloses an intelligent cleaning control system for high-altitude glass curtain walls, used to execute the aforementioned method, comprising: a multi-sensor synchronous acquisition module, designed to synchronously acquire sensing data from an inertial measurement unit, a negative pressure sensor, an ultrasonic ranging sensor, a visual camera, a wind speed, temperature and humidity sensor, and an infrared frame recognition sensor, and align them with a unified timestamp to form a synchronous sensing dataset; a grid modeling and path generation module, designed to perform grid modeling on the curtain wall area to be cleaned and generate a full-coverage cleaning path using a path search algorithm oriented towards full coverage; a stain recognition and classification module, designed to output stain level labels based on image frames using a pre-trained stain classification convolutional neural network; an adaptive cleaning parameter generation module, designed to generate an adaptive cleaning control parameter set based on the stain level labels and environmental parameters and output it to the drive control interface of the execution component; and a closed-loop negative pressure safety protection module, designed to perform closed-loop negative pressure regulation control based on attitude data, real-time negative pressure value, and wind speed value, and perform protective responses according to multiple safety conditions.

[0018] This invention provides an intelligent control system for high-altitude glass curtain wall cleaning equipment based on the Internet of Things, which solves the technical problems of insufficient safety in high-altitude operations, low cleaning coverage, and low utilization of cleaning resources, and achieves the following technical effects: Firstly, through synchronous acquisition and fusion calibration processing of multiple sensors, the data on which subsequent control decisions are based has redundancy and mutual verification. The impact of single sensor data deviation on control accuracy is reduced, providing a reliable data foundation for adaptive control.

[0019] Secondly, by using curtain wall grid modeling and an improved A* search algorithm for full coverage, a full-coverage cleaning path is generated. The cleaning trajectory can deterministically cover all passable areas, overcoming the uncertainty of missed cleaning and repeated cleaning caused by manual operation, and improving the integrity of cleaning coverage and the consistency of operation.

[0020] Third, by identifying stain levels through a stain classification convolutional neural network and combining it with environmental correction factors, differentiated adaptive cleaning control parameters are generated. The water spray flow rate and cleaning brush speed are matched with the actual stain severity and environmental conditions. Unnecessary water and electricity consumption in lightly stained areas is reduced, while the cleaning intensity in heavily stained areas is guaranteed. The utilization rate of cleaning resources and the uniformity of cleaning quality are improved at the same time.

[0021] Fourth, through closed-loop negative pressure regulation based on attitude deviation and wind speed and multiple safety protection responses executed in parallel, the equipment's adsorption reliability can be adaptively adjusted according to changes in external working conditions, and it has the ability to respond instantly when safety conditions are triggered. The risk of slippage and fall caused by insufficient negative pressure or attitude loss in ultra-high-rise environments is reduced. Attached Figure Description

[0022] Figure 1 This is a flowchart of the intelligent cleaning control method for high-altitude glass curtain walls provided in an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the original data of each sensor with the fusion calibration results provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the output probability distribution of the stain classification convolutional neural network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the adaptive cleaning control parameter step-by-step correction process provided in the embodiments of the present invention; Figure 5 This is a schematic diagram illustrating the contribution analysis of each component of the environmental correction factor provided in an embodiment of the present invention; Figure 6 This is a schematic diagram showing the comparison between the target value and the measured value of closed-loop negative pressure regulation and the safety threshold provided in the embodiments of the present invention; Figure 7 This is a schematic diagram showing the comparison between the current value of the security condition and the threshold provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the overall process of multi-sensor data acquisition and fusion control provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the distribution of curtain wall grid modeling parameters provided in an embodiment of the present invention. Detailed Implementation

[0023] The steps of this implementation method are as follows: Step 1: Multi-sensor data acquisition and synchronous fusion processing Within one acquisition cycle, the edge controller synchronously acquires the following multi-source sensing data: the pitch angle of the high-altitude glass curtain wall cleaning equipment obtained from the nine-axis IMU inertial measurement unit. Roll angle and heading angle Constructed attitude data vector The real-time negative pressure value of the vacuum adsorption chamber is obtained from the negative pressure sensor. Data sets of distances between high-altitude glass curtain wall cleaning equipment and surrounding window frames and obstacles are obtained from ultrasonic ranging sensors. ,in The number of ultrasonic ranging sensors; image frames of the current cleaning area are acquired from a high-definition vision camera. High-altitude wind speed values ​​are obtained from wind speed, temperature, and humidity sensors. Ambient temperature value and humidity value The constructed environmental parameter vector ; Obtain window frame boundary detection signals from infrared frame recognition sensors The data from all the aforementioned sensors are aligned using a unified timestamp to form a synchronous sensing dataset for that acquisition period. The sampling frequency of each sensor is uniformly set to a preset system sampling frequency to ensure the consistency of sensing data from different sources in the time dimension.

[0024] In a preferred embodiment of the present invention, in order to reduce the impact of data deviation caused by environmental interference or its own drift of a single sensor on the accuracy of subsequent control commands, the following sub-step of performing fusion calibration on the synchronous sensing dataset is included in addition to step 1: Step 101: Process the attitude data vector Each component in the data undergoes complementary filtering calibration. The inputs to the complementary filter are the low-frequency attitude estimates from the accelerometers and the high-frequency attitude estimates from the gyroscopes in the nine-axis IMU. The output of the complementary filter is the calibrated attitude data vector. .

[0025] Step 102: For the distance dataset Median filtering is performed within a preset time sliding window to obtain a denoised distance dataset. .

[0026] Step 103: Convert the calibrated attitude data vector Real-time negative pressure value Denoising distance dataset Image frames Environmental parameter vector and window frame boundary detection signal Unified encapsulation into fusion-aware data packets Fusion sensing data packets It is passed on to subsequent steps as input for control decisions.

[0027] A super high-rise complex (hereinafter referred to as Building A) features a full glass curtain wall structure, with a total height of 268 meters and 62 floors. The total curtain wall area is approximately 18,400 square meters. In June 20XX, the property management of Building A arranged for a high-altitude glass curtain wall cleaning machine (equipment number DEV-A01) to perform regular cleaning operations on the curtain wall section from the 38th to the 42nd floors. This section of the curtain wall is located at an altitude of approximately 148 to 168 meters above the ground. During the operation period that day, the wind speed at high altitude was relatively high, and the curtain wall surface had obvious water stains and mixed dirt from recent rainfall. The edge controller synchronously collected data from all sensors at a system sampling frequency of 100Hz, completing the data acquisition for a certain collection cycle at 09:07:23 on June 14, 20XX.

[0028] During this acquisition period, the original attitude data vector output by the nine-axis IMU inertial measurement unit is: These correspond to pitch, roll, and yaw angles. After complementary filtering calibration, the output is a calibrated attitude data vector. The low-frequency component of the accelerometer corrects for the cumulative drift bias of the gyroscope. The device is equipped with a 4-channel ultrasonic ranging sensor ( The original distance dataset is After sliding window midpoint filtering, the denoised distance dataset is The environmental parameter vector collected by the wind speed, temperature, and humidity sensors is: The window frame boundary detection signal output by the infrared frame recognition sensor (Not triggered). The above data is uniformly encapsulated into a fusion-aware data package. The real-time negative pressure value acquisition result is .

[0029] Table 1. Raw data and fusion calibration results of each sensor during the acquisition period.

[0030] Step 2: Curtain wall grid modeling and full-coverage cleaning path generation Based on the curtain wall parameter data, a grid modeling process is performed on the area of ​​the curtain wall to be cleaned to generate a full-coverage cleaning path. The curtain wall parameter data includes the total width of the curtain wall. Total height Window frame position coordinate set ( (Number of window frames) and the set of obstacle distribution areas ( (The number of obstacles) The above curtain wall parameter data was obtained from the previous building scan or pre-stored in the edge controller through manual input.

[0031] Step 201: Arrange the curtain wall area to be cleaned according to the preset grid unit size. Divided into two-dimensional grid maps 2D raster map The number of columns is The number of rows is ,in This is a rounding up operation. Two-dimensional raster map. Each grid cell in ( For row index, The column index is assigned a passable or impassable label. Grid cells that spatially overlap with the set of window frame position coordinates and the set of obstacle distribution areas are marked as impassable, while the remaining grid cells are marked as passable.

[0032] Step 202: In the 2D raster map Above, an improved A* search algorithm oriented towards full coverage is used to generate a full-coverage sweep path that traverses all passable grid cells. ,in For full coverage cleaning path The total number of raster cells included. The improved A* search algorithm for full coverage aims to traverse all accessible raster cells, and its cost function is defined as: ; in, This refers to the currently evaluated raster node; For grid nodes The comprehensive value of the product; From the starting point of the path to the current grid node The actual movement cost, from the path origin to the grid node. The cumulative number of grid cells traversed is calculated. To start from the current grid node The heuristic estimate of the cost required to achieve full coverage is calculated by multiplying the number of currently unvisited raster cells by the cost per unit movement. To extend to grid nodes The number of unvisited passable grid cells remaining at that time; The coverage penalty weight coefficient is used to control the degree of preference for coverage completeness during path search. The larger the value, the more the improved A* search algorithm, aimed at full coverage, tends to prioritize accessing uncovered raster cells. During the path search process, at each step of node expansion, it prioritizes adjacent raster nodes that can maintain a continuous reciprocating motion direction, thereby ensuring a more complete coverage sweep path. It tends towards a serpentine, repetitive traversal pattern. Full coverage cleaning path. The starting grid unit is determined by the grid coordinates corresponding to the current position of the high-altitude glass curtain wall cleaning equipment, and the full-coverage cleaning path is determined by this. The terminating raster cell is the last passable raster cell that has been traversed.

[0033] Furthermore, to prevent the high-altitude glass curtain wall cleaning equipment from repeatedly cleaning already completed areas due to insufficient power, malfunction, or external command interruption, the following breakpoint memory processing is also performed in addition to step 202: along the full-coverage cleaning path of the high-altitude glass curtain wall cleaning equipment... During the cleaning process, the edge controller continuously writes the indexes of the grid cells that have been cleaned to non-volatile memory. When the high-altitude glass curtain wall cleaning equipment restarts after an interruption, it reads the set of completed grid cell indexes from the non-volatile memory and completes the full-coverage cleaning path. The grid unit subsequence that has not yet been cleaned serves as the continuation path, and the high-altitude glass curtain wall cleaning equipment continues the cleaning operation from the interruption point.

[0034] The total width of the curtain wall section from the 38th to the 42nd floor of Building A is The total height is The curtain wall parameter data was pre-stored in the edge controller during the initial building scan. This section has a total of window frames. There are 10 windows, evenly distributed in 4 rows and 10 columns, with each window frame occupying an area approximately 0.15m wide and 0.12m high along the edge of the curtain wall; there is one external equipment box obstruction within the section. Located on the right side of the 39th floor, the coordinates of the obstacle area have been entered into the system.

[0035] The grid cell size is preset to Therefore, a two-dimensional raster map is calculated. The number of columns is The number of rows is The total number of cells in the raster map is After spatial overlap marking of the window frame position coordinate set and the obstacle distribution area set, a total of 187 grid cells were marked as impassable, and the total number of passable grid cells was 3653. The grid coordinates corresponding to the current device position are used as the basis for this classification. (i.e., the lower left corner of the curtain wall section) is taken as the starting point, and the coverage penalty weight coefficient is taken as... An improved A* search algorithm for full coverage aims to generate a serpentine traversal pattern for full-coverage cleaning paths. Total number of raster cells along the path .

[0036] Table 2. Curtain Wall Grid Modeling Parameters and Path Generation Results

[0037] Step 3: Stain Identification and Grading Based on fusion sensing data packets Image frames in The pre-trained stain classification convolutional neural network is used to identify and classify the type and severity of stains in the current cleaning area, and output stain level labels. .

[0038] Step 301: For image frames Preprocessing operations are performed, including adaptive histogram equalization, size normalization, and pixel value normalization, to obtain a standardized image. Among them, adaptive histogram equalization is used to improve image contrast under varying lighting conditions and curtain wall reflections, thereby enhancing the stability of subsequent stain identification; size normalization processing adjusts image frames... Scale to a fixed size required by the input layer of the stain classification convolutional neural network; pixel value normalization uses range-based mean normalization to scale the pixel values ​​of each channel to... The range was adjusted to eliminate the influence of differences in the absolute value of image brightness under different lighting conditions on the input of the convolutional neural network for stain classification.

[0039] Step 302: Standardize the image The input is fed into a pre-trained convolutional neural network for stain classification. The input layer of the stain classification convolutional neural network receives standardized images. After feature extraction through several convolutional and pooling layers, a fully connected layer maps the extracted features to scores for each stain level category. The output layer uses... The activation function outputs the probability distribution vector of the current cleaning area corresponding to each preset stain level. ,in The total number of preset stain severity categories, The current cleaning area belongs to the first The probability of each stain severity category This is an index for the stain severity category. The probability distribution vector is taken. The stain level label corresponding to the category with the highest probability value is used as the stain level label for the current cleaning area. ,Right now The stain classification convolutional neural network uses a set of curtain wall images labeled with stain levels as training data, with cross-entropy loss function as the optimization objective, and is trained using the Adam optimization algorithm.

[0040] In some embodiments, The corresponding level categories are as follows: Level 1 represents ordinary dust, Level 2 represents rain stains, Level 3 represents limescale, and Level 4 represents stubborn stains. The higher the level number, the more difficult it is to clean.

[0041] DEV-A01 along the full-coverage cleaning path Proceeding to a certain grid cell in the middle section of the 38th layer At that time, the high-definition vision camera captures image frames of the currently being cleaned area. The original resolution was 1920×1080 pixels. In the image preprocessing stage, adaptive histogram equalization eliminated localized overexposure areas caused by sidelighting that morning. Size normalization scaled the image to 224×224 pixels, and pixel value normalization mapped the pixel values ​​of each channel to... Range, to obtain a standardized image .Will Input stain classification convolutional neural network ( The output probability distribution vector is .

[0042] Therefore, the stain grade label is determined as follows: This means the current area is identified as a rain stain type, corresponding to level 2. This result is consistent with the actual situation of residual water marks on the curtain wall surface after rainfall that day.

[0043] Table 3. Output probability distribution of the convolutional neural network for stain classification

[0044] Step 4: Generation of Adaptive Cleaning Control Parameters Based on stain level labels and fusion sensing data packets Environmental parameter vector Generate an adaptive cleaning control parameter set corresponding to the current cleaning area. The adaptive cleaning control parameter set includes water pump spray flow instructions. Cleaning brush speed command Speed ​​command for high-altitude glass curtain wall cleaning equipment .

[0045] Step 401: Query the stain level label from the cleaning parameter mapping table pre-stored in the edge controller. Corresponding benchmark spray flow rate Reference cleaning brush speed and base speed The cleaning parameter mapping table is a pre-calibrated lookup table, where each stain level corresponds to a set of baseline parameter values. The higher the stain level, the greater the corresponding baseline water flow rate and baseline cleaning brush speed, and the lower the corresponding baseline travel speed. The baseline parameter values ​​are obtained by performing multiple cleaning tests on stain samples of each level under standard environmental conditions and selecting parameter combinations that meet the preset cleanliness standards. Standard environmental conditions refer to wind speed as a reference value. The temperature is a reference temperature value. The operating conditions.

[0046] Meanwhile, the edge controller determines the stain level label. Query the pre-stored cleaning solution mixing table to obtain the corresponding water to cleaning solution volume ratio for the current stain level. and mix volume ratio The control command is sent to the automatic dispensing valve to adjust the amount of cleaning solution as needed according to the severity of the stains.

[0047] Step 402: Based on environmental parameter vectors wind speed value and temperature value Calculate the environmental correction factor Before calculation, for and Each with its own reference value and Dimensionless processing is performed to ensure that wind speed (unit: m / s) and temperature (unit: °C) participate in the calculation as a ratio, eliminating the dimensional difference between them. The formula for calculating the environmental correction factor is as follows: ; in, This represents the current upper-level wind speed. For reference wind speed values; This is the current ambient temperature value; This is a reference temperature value; This is the wind speed correction factor; This is the temperature correction factor; both the reference value and the correction factor mentioned above are pre-stored in the edge controller. Because... and All are dimensionless ratios, environmental correction factors. It is a dimensionless scalar and can be directly multiplied by a reference cleaning parameter with different dimensions.

[0048] It should be noted that, In the above formula, the normalized reference for temperature deviation is physically represented as the reference temperature value under standard environmental conditions, expressed in °C. Positive values ​​in the absolute temperature sense are pre-stored in the edge controller to ensure It is a bounded positive value within the actual operating temperature range.

[0049] Step 403: Based on baseline parameters and environmental correction factors Calculate the adaptive cleaning control parameters: ; in, This is the command for the water pump's spray flow rate; The baseline spray flow rate; This is a command to adjust the cleaning brush rotation speed. The reference cleaning brush speed; This is the command for travel speed; The base speed of travel; This is an environmental correction factor. The greater the ambient wind speed or the more the temperature deviates from the reference value, the greater the environmental correction factor becomes. The larger the area, the greater the water flow rate and the faster the cleaning brush rotates to enhance cleaning power, while the slower the travel speed to extend the cleaning time per unit area.

[0050] Step 404: Based on fused sensing data packets Distance dataset after denoising and window frame boundary detection signal Perform near-obstacle deceleration correction: when any distance value in the distance dataset is less than a preset near-obstacle distance threshold. or window frame boundary detection signal When triggered, the travel speed command will be executed. Revised to: ; in, This is the revised travel speed command; The original travel speed command; This represents the current minimum obstacle distance value. This is a preset near-obstacle distance threshold; This is for calculating the minimum value. Because... and Since they have the same dimensions, their ratio is a dimensionless ratio and can be directly correlated with the travel speed command. Multiplication. Therefore, the closer the high-altitude glass curtain wall cleaning equipment is to obstacles, the greater the compression of its travel speed, achieving smooth deceleration and avoiding collisions caused by excessive travel speed when the high-altitude glass curtain wall cleaning equipment is close to window frames or protruding structures of the curtain wall.

[0051] The calculated adaptive cleaning control parameter set The output is sent to the drive control interface of the actuator. This includes the water pump spray flow rate command. and cleaning brush speed command During the obstacle proximity deceleration correction phase, the value obtained in step 403 remains unchanged, and the travel speed command is output after obstacle proximity deceleration correction. .

[0052] Based on the stain grade label obtained in step 3 (Rain stains), the edge controller queries the cleaning parameter mapping table to obtain the baseline parameters for level 2: baseline spray flow rate. Reference cleaning brush speed Baseline travel speed At the same time, consult the cleaning solution mixing table to obtain the mixing volume ratio corresponding to Grade 2. (The ratio of clean water to cleaning solution), and send the instruction to the automatic mixing valve.

[0053] The reference values ​​and correction coefficients pre-stored in the edge controller are as follows: Reference wind speed value Reference temperature value Wind speed correction factor Temperature correction factor The data collected in step 1 , Substitute into the environmental correction factor calculation formula: ; Therefore, the adaptive cleaning control parameters are calculated: ; Further perform near-obstacle deceleration correction. This is based on the denoised distance dataset from step 1. Current minimum obstacle distance value Preset near-obstacle distance threshold .because Triggering obstacle proximity deceleration correction: ; The final set of adaptive cleaning control parameters output to the actuator is as follows: .

[0054] Table 4 Summary of Adaptive Cleaning Control Parameter Calculation Process

[0055] Step 5: Closed-loop negative pressure adsorption and safety protection control Based on fusion sensing data packets The calibrated attitude data vector Real-time negative pressure value and environmental parameter vector wind speed value The vacuum pump is subjected to closed-loop negative pressure regulation control, and protective responses are implemented according to multiple preset safety conditions.

[0056] Step 501: Calculate the target negative pressure value under the current operating conditions. Before calculation, the overall tilt deviation angle is... and wind speed value Each with its own correction factor The dimensions of absorption are adjusted to make both equal to the reference negative pressure value. The addition operation remains consistent under the pressure dimension. The formula for calculating the target negative pressure value is: ; in, The target negative pressure value is expressed in Pa. The baseline negative pressure value is the minimum negative pressure required to maintain reliable adsorption under windless conditions and when the high-altitude glass curtain wall cleaning equipment is in a horizontal position. The unit is Pa. For the calibrated pitch angle and roll angle The calculated overall tilt deviation angle is expressed in degrees (°). This is the correction factor for negative pressure based on the tilt angle, expressed in Pa / °, used to convert angular quantities into pressure quantities. This is the wind speed value, in m / s; The wind speed is the correction factor for negative pressure, with the unit being Pa·s / m, used to convert wind speed into pressure. The above-mentioned reference negative pressure value and correction factor are pre-stored in the edge controller, thereby unifying the dimensions of each item in the formula to Pa.

[0057] Step 502: Calculate the target negative pressure value With real-time negative pressure value Deviation between With deviation As input, the speed adjustment of the vacuum pump is generated by the PID control algorithm and output to the vacuum pump drive interface, thereby realizing the real-time adaptive adjustment of the negative pressure value according to the attitude change of the high-altitude glass curtain wall cleaning equipment and the wind force change.

[0058] Step 503: While performing the closed-loop negative pressure regulation in steps 501 to 502, the following multiple safety condition judgments and protection responses are executed in parallel: (a) Pressure threshold warning: When the real-time negative pressure value Below the preset safe negative pressure threshold When the time comes, the edge controller outputs an audible and visual alarm trigger signal and outputs a walking motor lock command to prevent the high-altitude glass curtain wall cleaning equipment from continuing to move.

[0059] (b) Emergency braking due to attitude over-limit: when the comprehensive tilt deviation angle Exceeding the preset attitude safety threshold At that time, the edge controller outputs stop commands for all executing components and lock commands for the walking motor.

[0060] (c) Wind speed-linked start / stop: When the wind speed value Exceeding the preset operating wind speed threshold At this time, the edge controller outputs a stop command for all cleaning operations, and the high-altitude glass curtain wall cleaning equipment enters standby mode, while maintaining the continuous operation of the vacuum pump to ensure adsorption safety.

[0061] Among them, the safe negative pressure threshold Attitude safety threshold and operating wind speed threshold The equipment is calibrated based on its rated adsorption capacity, structural stability limits, and industry safety standards. These three safety conditions are continuously monitored in parallel. If any condition is triggered, the corresponding protective response is executed immediately, with priority higher than the cleaning operation control commands.

[0062] Furthermore, to avoid the risk of detachment caused by the loss of adsorption force due to the vacuum pump losing power in the event of a sudden power outage, the following power outage emergency protection control is also implemented in addition to step 503: The high-altitude glass curtain wall cleaning equipment is equipped with a backup energy storage module. When the edge controller detects that the main power supply voltage drops below the preset undervoltage threshold, it automatically switches to the backup energy storage module for power supply. The backup energy storage module only supplies power to the vacuum pump to maintain the negative pressure adsorption state, while the other execution components are in a power-off and shutdown state. This ensures that the high-altitude glass curtain wall cleaning equipment can still maintain the high-altitude adhesion state within the preset emergency adsorption duration after a sudden power outage.

[0063] Based on the calibrated attitude data vector acquired in step 1 and wind speed value First, calculate the overall tilt deviation angle: ; The safety protection parameters pre-stored in the edge controller are as follows: reference negative pressure value Tilt angle correction factor Wind speed correction factor Safety negative pressure threshold Attitude safety threshold Operating wind speed threshold Substitute into the target negative pressure value calculation formula: ; Calculate negative pressure deviation The PID control algorithm generates a positive adjustment amount for the vacuum pump speed, which drives the vacuum pump to increase its speed so that the real-time negative pressure value converges to the target negative pressure value.

[0064] Three safety condition checks are performed in parallel: real-time negative pressure value. The pressure threshold warning was not triggered; the overall tilt deviation angle... Attitude over-limit emergency braking not triggered; current wind speed value The wind speed linkage start / stop function was not triggered. All three safety conditions are within the normal range, and the cleaning operation will continue.

[0065] Table 5 Results of Closed-Loop Negative Pressure Regulation and Safety Condition Judgment

[0066] Step 6: Cloud transmission and remote monitoring of job data In a preferred embodiment of the present invention, in order to enable managers to monitor the status and progress of the high-altitude glass curtain wall cleaning equipment in real time, the following cloud transmission and remote monitoring steps are added in addition to the local autonomous cleaning control achieved in steps 1 to 5: Step 601: After each acquisition cycle, the edge controller uploads the following operational data to the cloud server via the wireless communication module: real-time negative pressure value. calibrated attitude data vector The number of grid cells that have been cleaned and the full-coverage cleaning path. Total number of grid cells The ratio represents the work progress, the remaining power of the high-altitude glass curtain wall cleaning equipment, and the currently effective set of adaptive cleaning control parameters. Environmental parameter vector The data includes the current triggered security response status indicator. This operational data is sent to the cloud server for storage at a preset reporting cycle.

[0067] Step 602: After receiving and storing the above-mentioned operation data, the cloud server pushes the current status information and operation progress information of the high-altitude glass curtain wall cleaning equipment to the mobile application or web-based management interface connected to the cloud server. Managers can view the real-time operation status of the high-altitude glass curtain wall cleaning equipment through the mobile application or web-based management interface, and can submit remote control commands to the cloud server through the same interface. These remote control commands include remote start / stop commands, parameter modification commands, and emergency stop commands. The cloud server then transmits the received remote control commands to the edge controller via the wireless communication module, and the edge controller executes the corresponding operations.

[0068] Step 603: The edge controller performs periodic self-checks on the operating status of each sensor. When the output data of any sensor continuously deviates from the preset normal range or sensor communication is interrupted, a corresponding fault code is generated and uploaded to the cloud server along with the work data. The cloud server pushes fault warning information to the mobile application based on the received fault code.

[0069] Based on the local autonomous cleaning control completed in steps 1 to 5, the DEV-A01 edge controller continuously uploads work data to the Building A property management cloud server via wireless communication module at a reporting cycle of 5 seconds. At 09:08:45, the edge controller encapsulated and uploaded the following work data: the number of grid cells that have been cleaned is 486, and the work progress is... The remaining battery power is 91%; the currently active adaptive cleaning control parameter set is... All three security protection response status indicators are "normal". After receiving and storing the information, the cloud server pushes the above status information to the property management personnel's mobile application.

[0070] The edge controller performs self-tests on all sensors within the same cycle. The output data of ultrasonic ranging sensors from channel 1 to channel 4 are all within the normal range. The communication status of the nine-axis IMU, negative pressure sensor, wind speed, temperature and humidity sensor and infrared frame recognition sensor are all normal, and no fault codes are generated.

[0071] Table 6 Contents of job data packets uploaded to the cloud during a certain reporting period

[0072] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An intelligent control system for high-altitude glass curtain wall cleaning equipment based on the Internet of Things, wherein the high-altitude glass curtain wall cleaning equipment is equipped with an inertial measurement unit, a negative pressure sensor, an ultrasonic ranging sensor, a vision camera, a wind speed and temperature / humidity sensor, and an infrared frame recognition sensor, characterized in that, The edge controller synchronously acquires the sensing data from each sensor and aligns them with a unified timestamp to form a synchronous sensing dataset; Based on the curtain wall parameter data, a grid model is performed on the area of ​​the curtain wall to be cleaned, and a path search algorithm oriented towards full coverage is used to generate a full-coverage cleaning path that traverses all passable grid cells. Based on the image frames in the synchronous sensing dataset, a pre-trained stain classification convolutional neural network is used to output the stain level label of the current cleaning area. Based on the stain level label and environmental parameters, an adaptive cleaning control parameter set is generated, including water spray flow command, cleaning brush rotation speed command and travel speed command. The vacuum pump is subjected to closed-loop negative pressure regulation control based on attitude data, real-time negative pressure value and wind speed value, and protective response is performed according to multiple safety conditions.

2. The control system according to claim 1, characterized in that, The formation of the synchronous sensing dataset also includes fusion calibration processing: Complementary filtering calibration is performed on each component of the attitude data vector. The inputs to the complementary filter are the low-frequency attitude estimate from the accelerometer and the high-frequency attitude estimate from the gyroscope, and the output is the calibrated attitude data vector. Median filtering is performed on the distance dataset within a preset time sliding window to obtain a denoised distance dataset. The calibrated attitude data vector, real-time negative pressure value, denoised distance dataset, image frame, environmental parameter vector, and window frame boundary detection signal are uniformly encapsulated into a fusion sensing data packet, which is then passed to subsequent steps as input for control decisions.

3. The control system according to claim 1, characterized in that, The cost function of the path search algorithm for full coverage is defined as: the actual movement cost from the path start point to the current grid node, the heuristically estimated cost from the current grid node to complete full coverage, and the number of currently remaining unvisited passable grid cells weighted by the coverage penalty weight coefficient; wherein the heuristically estimated cost is calculated by multiplying the number of currently unvisited grid cells by the unit movement cost; during each step of the path search process, when expanding nodes, adjacent grid nodes that can maintain a continuous reciprocating movement direction are preferentially selected, so that the generated full coverage sweep path tends to a serpentine reciprocating traversal mode.

4. The control system according to claim 1, characterized in that, The grid modeling includes: dividing the curtain wall area to be cleaned into a two-dimensional grid map according to a preset grid unit size; grid units in the two-dimensional grid map that spatially overlap with the set of window frame position coordinates and the set of obstacle distribution areas are marked as impassable, and the remaining grid units are marked as passable; during the cleaning process of the high-altitude glass curtain wall cleaning equipment along the full-coverage cleaning path, the edge controller continuously writes the index of the grid units that have been cleaned to a non-volatile memory; when the high-altitude glass curtain wall cleaning equipment is interrupted and restarted, it reads the set of completed grid unit indexes from the non-volatile memory and uses the grid unit subsequences that have not yet been cleaned in the full-coverage cleaning path as the continuation path.

5. The control system according to claim 1, characterized in that, The stain level labels output by the pre-trained stain classification convolutional neural network include: Adaptive histogram equalization, size normalization, and pixel value normalization are performed sequentially on the image frame to obtain a standardized image; The standardized image is input into a stain classification convolutional neural network. After feature extraction through several convolutional and pooling layers, the fully connected layer maps the extracted features to the scores of each stain level category. The output layer uses the softmax activation function to output the probability distribution vector corresponding to each preset stain level. The category with the highest probability value in the probability distribution vector is taken as the stain level label of the current cleaning area.

6. The control system according to claim 1, characterized in that, The generated adaptive cleaning control parameter set includes: The reference water flow rate, reference cleaning brush speed and reference travel speed corresponding to the stain level label are queried from the cleaning parameter mapping table pre-stored in the edge controller. The higher the stain level, the greater the reference water flow rate and reference cleaning brush speed, and the smaller the reference travel speed. Based on the ratio of wind speed value to reference wind speed value and the deviation ratio of temperature value to reference temperature value, an environmental correction factor is calculated with preset wind speed correction coefficient and temperature correction coefficient as weights. The environmental correction factor is a dimensionless scalar obtained by superimposing the wind speed ratio term and the temperature deviation ratio term on the reference value. The reference water flow rate and reference cleaning brush speed are multiplied by the environmental correction factor to obtain the water flow rate command and the cleaning brush speed command, respectively. The reference travel speed is divided by the environmental correction factor to obtain the travel speed command.

7. The control system according to claim 6, characterized in that, The generation of the adaptive cleaning control parameter set also includes near-obstacle deceleration correction: when any distance value in the denoised distance dataset is less than the preset near-obstacle distance threshold or the window frame boundary detection signal is triggered, the travel speed command is multiplied by the smaller of the ratio of the current minimum obstacle distance value to the near-obstacle distance threshold and the value one to obtain the corrected travel speed command; at the same time, the edge controller queries the pre-stored cleaning solution ratio table according to the stain level label to obtain the mixing volume ratio of clean water and cleaning solution corresponding to the current stain level, and sends the mixing volume ratio as a control command to the automatic mixing valve.

8. The control system according to claim 1, characterized in that, The closed-loop negative pressure regulation control includes: calculating the comprehensive tilt deviation angle based on the calibrated pitch and roll angles; using a pre-stored reference negative pressure value as a basis, superimposing the comprehensive tilt deviation angle weighted by the tilt angle correction coefficient and the wind speed value weighted by the wind speed correction coefficient to obtain the target negative pressure value; calculating the deviation between the target negative pressure value and the real-time negative pressure value; using the deviation as input, generating the speed adjustment of the vacuum pump through a PID control algorithm and outputting it to the vacuum pump drive interface.

9. The control system according to claim 8, characterized in that, The multiple safety conditions are continuously monitored in parallel and have a higher priority than the cleaning operation control commands, including: When the real-time negative pressure value is lower than the preset safe negative pressure threshold, an audible and visual alarm trigger signal is output and a walking motor lock command is output. When the overall tilt deviation angle exceeds the preset attitude safety threshold, output all execution components stop command and travel motor lock command; When the wind speed exceeds the preset operating wind speed threshold, a command to stop all cleaning operations is output to put the high-altitude glass curtain wall cleaning equipment into standby mode, while maintaining the continuous operation of the vacuum pump. When the main power supply voltage drops below the preset undervoltage threshold, it automatically switches to the backup energy storage module to supply power only to the vacuum pump to maintain the negative pressure adsorption state.

10. The control system according to claim 9, characterized in that, The multi-sensor synchronous acquisition module is designed to synchronously acquire the sensing data from the inertial measurement unit, negative pressure sensor, ultrasonic ranging sensor, visual camera, wind speed, temperature and humidity sensor and infrared frame recognition sensor, and align them with a unified timestamp to form a synchronous sensing dataset. The grid modeling and path generation module is designed to perform grid modeling on the curtain wall area to be cleaned and generate a full-coverage cleaning path using a path search algorithm oriented towards full coverage. The stain recognition and classification module is designed to output stain level labels based on image frames using a pre-trained stain classification convolutional neural network. The adaptive cleaning parameter generation module is designed to generate an adaptive cleaning control parameter set based on the stain level label and environmental parameters and output it to the drive control interface of the execution component. The closed-loop negative pressure safety protection module is designed to perform closed-loop negative pressure regulation control based on attitude data, real-time negative pressure value and wind speed value, and to perform protective response according to multiple safety conditions.

Images