A laser decontamination window cleaning robot and method

Abstract

The application discloses a laser stain removal window cleaning robot and method, the method comprising: collecting the inside and outside transmission and reflection difference of glass by low-energy scanning to generate the optical density of stains; selecting a target area according to a threshold value and implementing internal and external complementary multi-band excitation, extracting the transmission and reflection curve to obtain the stain type and aging information; transmitting a modulated pulse according to the obtained information and synchronously recording the glass vibration and temperature rise signal, jointly solving the stain thickness and adhesion strength; setting a test cleaning pulse according to the above multi-dimensional parameters and monitoring the transmittance increment to calculate a correction amount; coupling the correction amount with the area, curvature and residual energy of the region to generate a cleaning parameter set including power, pulse width, speed and the like for each target area and outputting the same. The scheme realizes real-time detection and adaptive laser cleaning of composite stains in a limited load scene without increasing a high-power spectrometer.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance and stain detection technology for building facades, and in particular to a laser-based window cleaning robot and method. Background Technology

[0002] Laser-powered window cleaning robots are intelligent equipment designed for the maintenance of high-rise building facades. Two units, one inside and one outside the window, are securely clamped to both sides of the glass via magnetic coupling and negative pressure adsorption. A high-energy-density, narrow-pulse laser beam penetrates the glass to remove dirt instantly and non-contactly. Each unit integrates a laser cleaning module, a composite adsorption device, a multi-modal detection system, and an electronic control unit, achieving integrated positioning, detection, and cleaning operations while maintaining a lightweight design.

[0003] Long-term operation has shown that the fouling on high-rise curtain walls exhibits distinct vertical stratification: the bottom layer is primarily composed of a mixed organic film of hydrocarbons and urban dust; the middle layer consists mainly of silicate and carbonate particles; and the upper layer is mostly a transparent polymer film formed by the condensation of atmospheric VOCs and bioaerosols. At the same height, horizontal heterogeneity is also compounded by differences in orientation. Further complicating matters, ultraviolet radiation causes photochemical aging of the organic fouling over time, forming a highly cross-linked polymer network with spectral absorption shifts. This network, layered with recent deposits, results in a strong nonlinearity and wavelength selectivity in laser absorption. In this scenario of multiple components, thicknesses, and aging levels, fixed wavelength or single-energy strategies often lead to deep-seated fouling residue or secondary surface carbonization, ultimately reducing glass transmittance and increasing rework costs.

[0004] Most existing systems rely on grayscale thresholds or simple morphology analysis for dirt classification, making it difficult to accurately distinguish complex, composite dirt. While higher-resolution spectrometers such as Raman, LIBS, or FT-IR can provide more accurate compositional information, these solutions are bulky and power-intensive, making them unsuitable for window cleaning robot platforms with limited load and real-time requirements. Therefore, achieving real-time, accurate identification of complex dirt within limited hardware resources, and dynamically matching laser cleaning parameters accordingly, has become a critical technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The main objective of this invention is to solve the technical problem that existing technologies cannot achieve real-time and accurate identification of complex dirt within limited hardware resources, and dynamically match laser cleaning parameters accordingly.

[0006] The first aspect of this invention provides a laser-based window cleaning method, the laser-based window cleaning method comprising: A low-energy probe light full-coverage scan is performed on the glass surface to collect the outer reflected signal and the inner transmitted signal and perform differential calculation to obtain the dirt optical density distribution data. Based on the dirt optical density distribution data, target areas with optical density exceeding a preset threshold are selected. Multi-band response measurements are performed on the target areas, and the transmission-reflection response curves are analyzed to obtain dirt type data and aging degree data. Based on the dirt type data and aging degree data, modulated pulse laser is emitted to the target area, and the inner vibration signal and temperature rise signal are collected and jointly calculated to obtain dirt thickness data and adhesion strength data. Based on the data on dirt type, aging degree, dirt thickness, and adhesion strength, the parameters for trial cleaning laser are set, and trial cleaning is performed on the target area.

[0007] Preferably, the step of performing a low-energy probe light full-coverage scan on the glass surface, acquiring the outer reflected signal and the inner transmitted signal, and performing differential calculations to obtain dirt optical density distribution data includes: The probe light scanning path is divided into three-dimensional grids to obtain a scanning coordinate matrix containing grid coordinate data and scanning timestamp data. The outer reflected light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the reflected light intensity matrix and the outer optical path displacement data corresponding to the scanning coordinate matrix. The inner transmitted light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the transmitted light intensity matrix and inner optical path displacement data corresponding to the scanning coordinate matrix. Based on the outer optical path displacement data and the inner optical path displacement data, curvature calculation processing is performed to obtain the glass curvature distribution matrix; Based on the glass curvature distribution matrix, the reflected light intensity matrix and the transmitted light intensity matrix are subjected to thickness-curvature compensation processing to obtain the compensated optical response matrix. Based on the compensated optical response matrix, grid cells with transmitted light intensity located in the preset high transmission percentile range are selected for reference ratio calculation to obtain dirt optical density distribution data.

[0008] Preferably, the step of performing curvature calculation processing based on the outer optical path displacement data and the inner optical path displacement data to obtain the glass curvature distribution matrix includes: The outer optical path displacement data and the inner optical path displacement data are processed by grid-by-grid difference calculation to obtain the optical path difference matrix; Based on the optical path difference matrix, a first-order gradient operation is performed on the difference between adjacent grids to obtain the slope matrix of the glass surface. The glass curvature distribution matrix is ​​obtained by performing second-order gradient calculations on the slopes of adjacent grids based on the glass surface slope matrix.

[0009] Preferably, the step of selecting target areas with optical density exceeding a preset threshold based on the dirt optical density distribution data, performing multi-band response measurements on the target areas, analyzing the transmission-reflection response curves, and obtaining dirt type data and aging degree data includes: Based on the dirt optical density distribution data, connected component clustering is performed on continuous grids with optical density exceeding a preset threshold to obtain a target region set. The target area set is grouped based on its height and azimuth coordinates to obtain a target area group list. For each target region in the target region grouping list, perform multi-band excitation processing of the inner transmission channel and multi-band excitation processing of the outer reflection channel in sequence to obtain the corresponding target region multi-band transmission-reflection raw data; Based on the original multi-band transmission-reflection data, the boundary grid of the target area is locally normalized to obtain multi-band transmission-reflection correction data. Based on the multi-band transmission-reflection correction data, peak-valley morphological features of each band curve are extracted to obtain dirt type data and aging degree data.

[0010] Preferably, the inner transmission channel multi-band excitation processing and the outer reflection channel multi-band excitation processing are sequentially performed on each target region in the target region grouping list to obtain the corresponding target region multi-band transmission-reflection raw data, including: Based on the height and azimuth coordinates of the target area grouping list, the preset band pool is filtered to obtain a complementary band sequence; For each band in the complementary band sequence, perform outer channel excitation processing and acquire the outer reflected signal and the inner transmitted signal to obtain the outer sampling matrix; Based on the timestamp data of the complementary band sequence and the completion of the outer channel excitation, a time offset is set, and inner channel excitation processing is performed on each band of the complementary band sequence to collect the outer reflection signal and the inner transmission signal, thereby obtaining the inner sampling matrix. The outer sampling matrix and the inner sampling matrix are paired at the band level to obtain the channel difference matrix. Data splicing is performed based on the channel difference matrix and complementary band sequence to obtain the original multi-band transmission-reflection data of the corresponding target area.

[0011] Preferably, the step of emitting modulated pulsed laser to the target area based on the dirt type data and aging degree data, collecting the inner vibration signal and temperature rise signal, and performing joint calculation to obtain dirt thickness data and adhesion strength data includes: Based on the dirt type data and aging degree data, the repetition frequency, peak power and pulse width of the modulated pulse laser are set by parameter setting to obtain the pulse parameter set; Perform dual-level sequential excitation processing on the pulse parameter group to obtain the original inner vibration signal and the original inner temperature rise signal; The original inner vibration signal is subjected to frequency domain transformation to obtain vibration mode spectrum data, and the original inner temperature rise signal is subjected to time gradient calculation to obtain temperature rise peak data. By performing joint regression analysis on vibration modal spectrum data and temperature rise peak data, dirt thickness data and adhesion strength data are obtained.

[0012] Preferably, the step of performing dual-level sequential excitation processing on the pulse parameter group to obtain the original inner vibration signal and the original inner temperature rise signal includes: Based on the pulse parameter set, glass structure data and magnetic gap data, the power of the test level, the power of the main excitation level and the time interval between levels are proportionally calculated to obtain the level sequence parameters. According to the energy level sequence parameters, test energy level pulses are emitted to the target area and vibration and temperature rise signals are collected simultaneously to obtain test response data. The glass-air cavity resonant frequency data is calculated based on the test response data, and the parameters of the main excitation level power and the time interval between levels are revised to obtain the revised level sequence parameters. According to the revised energy level sequence parameters, the main excitation energy level pulse is emitted to the target region and the vibration signal and temperature rise signal are collected simultaneously. The test response data and the main excitation response data are merged and processed to obtain the original inner vibration signal and the original inner temperature rise signal.

[0013] Preferably, the step of setting the trial cleaning laser parameters based on the dirt type data, aging degree data, dirt thickness data, and adhesion strength data, and performing trial cleaning on the target area, includes: Based on the dirt type data, aging degree data, dirt thickness data, adhesion strength data, glass curvature distribution matrix and magnetic gap data, power coefficient calculation, pulse width coefficient calculation and scanning speed coefficient calculation are performed to obtain the test cleaning laser parameter set; The target area is subjected to trial cleaning treatment according to the trial cleaning laser parameter set, and real-time transmittance curve data is collected simultaneously to obtain the transmittance change curve. The transmittance increment data is calculated based on the transmittance change curve, and the transmittance increment data is compared with the preset transmittance threshold to obtain the laser parameter difference coefficient. The laser parameter group for the trial cleaning is revised based on the laser parameter difference coefficient to obtain laser parameter correction data.

[0014] Preferably, after obtaining the laser parameter correction data, the process includes: Based on the laser parameter correction data, target area data, glass curvature distribution matrix and robot remaining energy data, power ratio processing, pulse width ratio processing and scanning speed ratio processing are performed to obtain the regional energy demand matrix. Based on the regional energy demand matrix and the in-window-out-window scanning sequence data, a path sequence optimization process is performed to obtain a set of cleaning parameters. The cleaning parameter set is processed by queue allocation to obtain a queue of cleaning parameters to be executed and the queue of cleaning parameters to be executed is output.

[0015] A second aspect of the present invention provides a laser-based window cleaning robot, wherein the laser-based window cleaning robot employs the laser-based window cleaning method described in any of the above embodiments.

[0016] The technical solution provided in this application initially uses low-energy scanning to collect reflected signals on the outside of the glass and simultaneously collect transmitted signals on the inside. The two types of signals are then converted into optical density distribution data through differential analysis. The reflectance-transmittance ratio is complementary to the incident light energy at the same point, and the differential operation cancels out the influence of external illumination and glass thickness fluctuations, thereby revealing the spatial distribution of pollution intensity with minimal energy and establishing a coordinate reference for subsequent detection.

[0017] Within regions where the optical density exceeds a threshold, the inner and outer bodies sequentially emit complementary wavelength probe light, and the response curves on both sides are recorded. The transmission-reflection morphology of different wavelengths varies with organic and inorganic components and the degree of ultraviolet aging. Peak-valley characteristic analysis can determine the type of dirt and the aging state. This step utilizes the robot's inherent perspective difference to construct a bidirectional spectrum along the same optical path, obtaining diagnostic information related to material composition without the need for a high-resolution spectral module.

[0018] After obtaining information on the type and aging of the fouling, the method emits a modulated pulse sequence towards the target area while simultaneously monitoring vibration and temperature rise on the inside of the glass. The vibration modes are jointly determined by the thickness and adhesion strength of the fouling layer; the peak temperature rise reflects the total absorbed energy. By jointly calculating these two parameters, thickness and adhesion parameters can be derived without the need for a mechanical probe, further expanding the detection scope without increasing the hardware burden.

[0019] Subsequently, based on the aforementioned multidimensional detection results, a trial cleaning pulse with a power slightly lower than the removal threshold was set, and the difference in the instantaneous increment of transmittance was calculated. The transmittance increment directly corresponds to the actual amount of dirt removed, and the difference coefficient is the quantitative index of model error. By writing back this coefficient, the pulse power, pulse width, and scanning speed are revised to achieve parameter self-calibration.

[0020] The terminal stage matches the revised pulse parameters with the area, curvature, and remaining energy of each region to generate a set of cleaning parameters for each zone, which is then output to the execution layer. Since each parameter is traced back to optical, thermal, or mechanical detection data, the generation process maintains a detection-decision closed loop, ensuring subsequent cleaning actions.

[0021] The entire process uses lightweight optical and acoustic-thermal signal measurement to connect the four-level chain of identification, diagnosis, correction and output. Without relying on large-scale spectroscopic equipment, it provides mobile window cleaning platforms with real-time, precise and full-parameter dirt detection and laser parameter matching capabilities, fundamentally solving the technical bottlenecks of difficult identification of composite dirt and energy configuration imbalance on high-rise curtain walls. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of one embodiment of the laser-based window cleaning method according to the present invention.

[0024] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0027] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, provided that they are feasible for those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0028] One embodiment of this application provides a method for cleaning windows using laser-based stain removal. Figure 1 This is a flowchart illustrating a laser-based window cleaning method according to an embodiment of this application. In this embodiment, the method includes: Please see Figure 1 A low-energy probe light full-coverage scan is performed on the glass surface to collect the outer reflected signal and the inner transmitted signal and perform differential calculation to obtain the dirt optical density distribution data. In one embodiment of the present invention, the step of performing a low-energy probe light full-coverage scan on the glass surface, acquiring the outer reflected signal and the inner transmitted signal, and performing differential calculation to obtain dirt optical density distribution data includes: The probe light scanning path is divided into three-dimensional grids to obtain a scanning coordinate matrix containing grid coordinate data and scanning timestamp data. The outer reflected light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the reflected light intensity matrix and the outer optical path displacement data corresponding to the scanning coordinate matrix. The inner transmitted light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the transmitted light intensity matrix and inner optical path displacement data corresponding to the scanning coordinate matrix. Based on the outer optical path displacement data and the inner optical path displacement data, curvature calculation processing is performed to obtain the glass curvature distribution matrix; Based on the glass curvature distribution matrix, the reflected light intensity matrix and the transmitted light intensity matrix are subjected to thickness-curvature compensation processing to obtain the compensated optical response matrix. Based on the compensated optical response matrix, grid cells with transmitted light intensity located in the preset high transmission percentile range are selected for reference ratio calculation to obtain dirt optical density distribution data.

[0029] The following is a detailed description of the steps involved in the above embodiments: When performing spatial three-dimensional meshing on the probe light scanning path, the external working unit uses a low-power collimated laser (typically 5-10mW) to scan the glass surface. The motion controller discretizes each 20mm×20mm sampling point and the scanning trajectory into (x,y,z) three-dimensional nodes in the machine coordinate system, and writes a millisecond-level timestamp to each node, forming a scanning coordinate matrix. The scanning coordinate matrix is ​​a data structure that contains the three-dimensional spatial position and corresponding time information of each sampling point during the scanning process. x and y correspond to the column and row indices in the glass plane, and z records the normal distance from the laser emission point to the outer surface of the glass, reflecting the micro-displacement of the adsorption gap caused by changes in magnetic force. This matrix provides a unified spatial-temporal reference for all optical readings on the outer and inner sides, ensuring a one-to-one correspondence between subsequent multi-source data. This meshing method ensures that the probe light systematically covers the entire glass surface, avoiding sampling blind spots. Furthermore, after meshing, the scanning head can directly call the matrix when repeatedly running on the same trajectory without reconstruction, reducing the accumulation of positioning errors.

[0030] When synchronously acquiring and processing the reflected light signal from the outside based on the scanning coordinate matrix, the laser beam returns to the external working unit and is guided to the linear array photodetector by a beam splitter. The detector writes the reflection amplitude corresponding to each node into the same index position as the scanning coordinate matrix, generating a reflected light intensity matrix. The reflected light intensity matrix is ​​a data set recording the reflected light intensity value at each sampling point, and has the same index structure as the scanning coordinate matrix. Simultaneously, a miniature triangulation ranging module on the external body records the length change of the laser incident and reflection paths, writing this difference into the external optical path displacement data. The external optical path displacement data is a data set representing the difference between the actual and ideal propagation paths of the laser. The optical path displacement is defined as the measured optical path at a node minus the theoretical optical path; positive values ​​represent the outer arch of the glass, and negative values ​​represent the inner concavity. A high-speed data acquisition card (such as a DAQ card with a sampling rate of 50kHz) is used to ensure sampling accuracy, while a PID control system maintains the distance stability between the external working unit and the glass surface. Synchronous triggering ensures that the amplitude and displacement have the same timestamp, facilitating the removal of sampling time delay using a differential algorithm during curvature calculation. This synchronous acquisition method provides accurate spatial distribution information for subsequent analysis.

[0031] When synchronously acquiring and processing the transmitted light signal from the inner side according to the scanning coordinate matrix, the corresponding node, after passing through the glass into the indoor side, is collected by the anti-reflection window of the working machine inside the window and sent to the photodiode array. The array output is written into the transmitted light intensity matrix, which is a data set recording the intensity values ​​of light transmitted through the glass, with the indexing rules consistent with the outer side. The inner body also uses a ranging module to record the length of the transmitted light path, generating inner light path displacement data. The inner light path displacement data records the difference between the actual propagation path and the theoretical path of the transmitted light. The sensitivity of the inner photodetector is set to 0.1 μW / cm. 2The above measures ensure the detection of even the weakest transmitted light signals. The two working machines exchange time reference signals in real time via a wireless communication module (such as Bluetooth 5.0), with time synchronization accuracy controlled within 1ms. Since the internal and external ranging devices are calibrated on the same reference plane, their difference can directly represent the combined change in glass thickness and instantaneous bending, providing a data source for subsequent geometric compensation. This collaborative acquisition method fully utilizes the dual-machine structure of the laser window cleaning robot.

[0032] When calculating curvature based on the displacement data of the outer and inner optical paths, the algorithm first performs point-to-point subtraction of the outer and inner displacement matrices to obtain Δz. Then, it calculates first-order difference matrices in the x and y directions to obtain the local normal tilt angle. Subsequently, it performs second-order difference on the tilt angle matrix to output the glass curvature distribution matrix. The glass curvature distribution matrix is ​​a data set representing the degree of curvature of the glass surface at various points. The values ​​of the curvature matrix elements are equal to the reciprocal of the radius of curvature around the node; positive values ​​represent outward arch curvature, and negative values ​​represent inward concavity curvature. On a typical 6mm tempered glass curtain wall, the curvature threshold between nodes is set to 1×10-1. -3 mm -1 It can identify micro-bends with radii greater than 2m without excessively amplifying measurement noise. This curvature calculation method based on optical path offset does not require contact with the glass surface, avoiding interference from dirt in traditional contact measurements, while providing high-precision curvature data.

[0033] When performing thickness-curvature compensation on the reflected and transmitted light intensity matrices based on the glass curvature distribution matrix, the compensation model first corrects the incident angle at each node using the curvature matrix, thereby correcting the cosine weakening of the reflection-transmission ratio with respect to angle. Simultaneously, it estimates the local optical path increment based on the nominal glass thickness and the change in Δz, and then calculates it according to e... (-α·Δl) A factor is used to adjust the transmission amplitude to compensate for the additional attenuation caused by abrupt changes in thickness, where α is the absorption coefficient of the glass material and Δl is the optical path increment. After two levels of correction, a compensated optical response matrix is ​​obtained. The compensated optical response matrix is ​​a set of optical signal data corrected for geometric factors. Its amplitude has an approximately linear relationship with the dirt absorption capacity, reducing the influence of geometric deformation on the accuracy of discrimination. This compensation process eliminates the interference of the glass's own characteristics on the measurement results.

[0034] When calculating the reference ratio based on the selected grid cells with transmitted light intensity within a preset high transmittance percentile range according to the compensated optical response matrix, nodes with transmittance amplitude above the 95th percentile are selected as the reference clean area for this window surface. For each non-reference node, the ratio of its transmittance-reflectance ratio to the mean of the reference nodes is calculated, and the natural logarithm is taken to obtain the optical density. The optical density of all nodes is written into the dirt optical density distribution data. The dirt optical density distribution data is a set of data representing the dirt concentration at various points on the glass surface, and is related to the dirt layer thickness and absorption coefficient through a logarithmic relationship. A percentile threshold of 95% is preferred to shield the interference of small, uncontaminated patches on the reference while avoiding insufficient samples due to excessively high percentiles. Experiments show that this threshold has good robustness to differences in sunlight and orientation. This yields a regionalized contamination intensity map with an accuracy better than ±4%, providing reliable input for subsequent multi-band analysis. This density calculation method based on the difference between internal and external light signals effectively eliminates interference from ambient light fluctuations and the inherent characteristics of the glass.

[0035] In one embodiment of the present invention, the step of performing curvature calculation processing based on the outer optical path displacement data and the inner optical path displacement data to obtain the glass curvature distribution matrix includes: The outer optical path displacement data and the inner optical path displacement data are processed by grid-by-grid difference calculation to obtain the optical path difference matrix; Based on the optical path difference matrix, a first-order gradient operation is performed on the difference between adjacent grids to obtain the slope matrix of the glass surface. The glass curvature distribution matrix is ​​obtained by performing second-order gradient calculations on the slopes of adjacent grids based on the glass surface slope matrix.

[0036] The following is a detailed description of the steps involved in the above embodiments: When performing grid-by-grid difference calculations on the outer and inner optical path displacement data, the two sets of data are first matched according to their corresponding grid coordinates. The outer optical path displacement data is the difference between the laser reflection path measured by the external working machine and the theoretical path, while the inner optical path displacement data is the difference between the transmitted light path measured by the internal working machine and the theoretical path. The digital signal processor performs difference calculations for each grid position (i,j): ΔD(i,j) = Dout(i,j) - Din(i,j), where Dout(i,j) represents the outer optical path displacement value, and Din(i,j) represents the inner optical path displacement value. The calculation result ΔD(i,j) is stored in the corresponding position in the optical path difference matrix. The optical path difference matrix is ​​a data set describing the difference in optical path displacement on both sides of the glass, recording the combined effect of glass deformation and thickness variation. For example, in the detection scenario of the curved curtain wall of a high-rise building, if the outer optical path displacement at a certain grid position is +2.5mm and the inner optical path displacement is +1.2mm, then the optical path difference at that position is +1.3mm, indicating that the glass has an outward bending at that location. This point-by-point difference calculation can eliminate common-mode interference such as changes in the body's attitude, retaining only information related to the geometric characteristics of the glass itself, thus improving the accuracy of subsequent curvature calculations.

[0037] When performing first-order gradient calculations on the differences between adjacent grids based on the optical path difference matrix, the digital signal processor calculates the rate of change of the optical path difference along the x and y directions respectively, forming the glass surface slope matrix. The glass surface slope matrix is ​​a set of data describing the degree of tilt of the glass surface at each point. Specifically, for grid position (i,j), the slope in the x-direction is calculated as: Sx(i,j)=[ΔD(i+1,j)-ΔD(i,j)] / dx, where dx is the actual physical spacing between adjacent grids in the x-direction, typically 20mm; similarly, the slope in the y-direction is: Sy(i,j)=[ΔD(i,j+1)-ΔD(i,j)] / dy. The calculated Sx(i,j) and Sy(i,j) are stored in the x-direction slope matrix and the y-direction slope matrix respectively, together forming the glass surface slope matrix. In actual building glass inspections, such as in the glass area near the central air conditioning unit of an office building, temperature gradients cause uneven thermal expansion, potentially resulting in a slope value of 0.005 in the x-direction (i.e., a change of 0.1 mm for every 20 mm), indicating a significant tilt in that area. This first-order gradient processing converts the absolute change in optical path difference into a relative rate of change, effectively reducing the impact of measurement starting point deviation and laying the foundation for accurate calculation of glass curvature.

[0038] When performing second-order gradient calculations on the slopes of adjacent grids based on the glass surface slope matrix, the digital signal processor further calculates the rate of change of the slope to obtain the glass curvature distribution matrix. The glass curvature distribution matrix is ​​a two-dimensional data array describing the degree of glass curvature, where each element represents the magnitude of curvature at its corresponding location. For grid position (i,j), the curvature in the x-direction is calculated as: Kx(i,j)=[Sx(i+1,j)-Sx(i,j)] / dx; the curvature in the y-direction is calculated as: Ky(i,j)=[Sy(i,j+1)-Sy(i,j)] / dy. The total curvature value K(i,j)=Kx(i,j)+Ky(i,j) is stored in the corresponding position of the glass curvature distribution matrix. The unit of curvature value is mm. -1 , representing the reciprocal of the radius of curvature, with a positive value indicating outward convexity and a negative value indicating inward concavity. In practical applications, such as the large glass curtain wall in the atrium of a large shopping mall, slight deformation may occur in areas with strong sunlight, with a measured curvature value of 1×10. -3 mm -1 This is equivalent to a bending radius of 1000mm (1 meter). For the 6mm tempered glass commonly used in building curtain walls, a 1×10... -3 mm -1 The curvature threshold setting effectively identifies micro-bends with a bending radius of over 2 meters, while avoiding excessive amplification of measurement noise. This second-order gradient processing directly reflects the bending characteristics of the glass surface, providing key parameters for subsequent optical density data compensation and ensuring the accuracy of dirt detection on irregular curved surfaces.

[0039] Please continue reading. Figure 1 Based on the dirt optical density distribution data, target areas with optical density exceeding a preset threshold are selected, multi-band response measurements are performed on the target areas, and the transmission-reflection response curves are analyzed to obtain dirt type data and aging degree data. In one embodiment of the present invention, the step of selecting a target area with an optical density exceeding a preset threshold based on the dirt optical density distribution data, performing multi-band response measurement on the target area, analyzing the transmission-reflection response curve, and obtaining dirt type data and aging degree data includes: Based on the dirt optical density distribution data, connected component clustering is performed on continuous grids with optical density exceeding a preset threshold to obtain a target region set. The target area set is grouped based on its height and azimuth coordinates to obtain a target area group list. For each target region in the target region grouping list, perform multi-band excitation processing of the inner transmission channel and multi-band excitation processing of the outer reflection channel in sequence to obtain the corresponding target region multi-band transmission-reflection raw data; Based on the original multi-band transmission-reflection data, the boundary grid of the target area is locally normalized to obtain multi-band transmission-reflection correction data. Based on the multi-band transmission-reflection correction data, peak-valley morphological features of each band curve are extracted to obtain dirt type data and aging degree data.

[0040] The following is a detailed description of the steps involved in the above embodiments: When performing connected component clustering on continuous grids with optical densities exceeding a preset threshold based on dirt optical density distribution data, a dirt optical density threshold is first set. This threshold is typically 1.5 to 2 times the baseline optical density of clean glass. Connected component clustering is a data processing method that groups adjacent grids that meet specific conditions into the same group. In practice, an embedded processor executes an 8-connected component algorithm, which checks the horizontal, vertical, and diagonal adjacent grids of each grid. If the optical densities of adjacent grids all exceed the preset threshold and there are direct or indirect connection paths between them, these grids are marked as the same connected region. For example, in the inspection of the facade of a high-rise office building, an oily area of ​​approximately 20cm × 30cm was found on the south-facing window of the 15th floor. Its optical density value was between 0.25 and 0.4, much higher than the 0.05 to 0.08 of the clean area. The system would mark this continuous grid as the same target region. Each connected region is assigned a unique region ID, as well as attribute information such as region boundary coordinates and area. All these connected regions and their attribute information constitute the target region set. This clustering process avoids the inefficient practice of analyzing each contamination point scattered across the entire window surface one by one. Instead, it treats physically continuous contamination areas as a whole, significantly reducing the computational load for subsequent multi-band detection.

[0041] When grouping target areas based on their height and azimuth coordinates, the system categorizes target areas with similar spatial distribution characteristics into the same group. Height coordinates refer to the vertical location of the target area, typically expressed as its height above the ground; azimuth coordinates refer to the distribution of the target area along the building's orientation, such as east, south, west, and north. In practice, a hierarchical clustering algorithm is used. First, the target areas are divided into three categories based on height: low-rise (0-30m), mid-rise (30-100m), and high-rise (above 100m). Then, within each height category, further subdivisions are made based on azimuth angles at 45° intervals (0° / east, 45° / northeast, 90° / north, etc.). For example, if a high-rise commercial building has multiple rain streaks and dirt areas on its southwest-facing exterior wall at a height of 50-60m, the system will group these areas together into the "Mid-rise Southwest Area Group." The grouping results are output as a target area group list, which includes the ID of each group, a list of target area IDs, and representative spatial coordinates. This layered and zoned grouping method fully considers the differences in dirt types at different heights and orientations, such as oil stains on lower floors, dust on higher floors, and rain streaks on windward surfaces, thus preparing for targeted testing in the future.

[0042] When performing multi-band excitation processing of the inner transmission channel and the outer reflection channel for each target area in the target area grouping list, the system first controls the external working unit to stay at the center of the target area and emits probe light of different wavelengths in sequence. Simultaneously, the internal and external working units record the transmission and reflection signals, respectively. Then, the internal working unit emits the same wavelength sequence, repeating the measurement process. Multi-band excitation processing refers to the process of irradiating the target area with light sources of different wavelengths and recording the response. In actual implementation, the system uses a multi-wavelength LED or semiconductor laser array with a wavelength range of 365nm (ultraviolet) to 1064nm (near-infrared), which is excited sequentially according to a preset sequence. For example, when detecting stubborn watermarks on the exterior wall of a high-rise building, the system will emit probe light of six wavelengths: 365nm, 450nm, 532nm, 650nm, 850nm, and 1064nm in sequence. For each band, the photodiode array on the external working unit records the reflected signal intensity, while the photodetector on the internal working unit simultaneously records the transmitted signal intensity. The data acquisition card collects data at a rate of 100 samples per band and performs noise suppression processing. The transmission and reflection signal intensity data for all bands are organized according to time sequence and wavelength information to form multi-band transmission-reflection raw data for the corresponding target area. This dual-channel multi-band detection fully utilizes the differences in absorption and reflection characteristics of different types of dirt to different wavelengths of light, providing rich spectral information for dirt composition analysis.

[0043] When performing local normalization on the boundary grid of the target area based on the original multi-band transmission-reflection data, the system selects relatively clean grid points at the edge of the target area as reference points and performs amplitude calibration on the data in the same band. Local normalization refers to the process of proportionally adjusting the original data according to a local reference standard. In specific implementation, for a ring-shaped area 2-3 grid widths around the target area, grid points with optical density 30% lower than the average of the target area are selected as local reference points. For each band λ, the average transmittance T of the reference points is calculated. ref(λ) and reflectivity R ref(λ) Then, the raw data T of all measurement points within the target area... raw(λ) and R raw(λ) Normalize: T norm(λ) =T raw(λ) / T ref(λ) R norm(λ) =R raw(λ) / R ref(λ) For example, when inspecting areas of grime on the exterior wall of an office building, a relatively clean point is selected at the edge of the area as a reference. If the transmittance at this point is 0.85 at a wavelength of 532 nm, the transmittance of 0.51 measured within the area becomes 0.6 after normalization. The normalized transmittance and reflectance data for each band constitute multi-band transmittance-reflectance correction data. This local normalization method eliminates systematic errors between different areas of the same window surface, such as differences in glass background transmittance and detector sensitivity drift, making the spectral characteristics more prominent and facilitating subsequent feature extraction and analysis.

[0044] When extracting peak-valley morphology features from multi-band transmission-reflection correction data, the system analyzes the characteristic point positions and amplitude variation trends of the spectral curves to extract the spectral fingerprint information of the dirt. Peak-valley morphology feature extraction refers to a data processing method that identifies characteristic absorption peaks, reflection peaks, and their relative intensity relationships from spectral curves. In practice, the data of each band is first arranged in wavelength order to construct spectral curves, and peak detection algorithms are applied to determine the wavelength positions of the minimum transmittance points (absorption peaks) and the maximum reflectance points (reflection peaks). Then, characteristic ratios are calculated, such as the ratio of ultraviolet absorption to visible light absorption (365nm / 532nm transmittance ratio) and the ratio of near-infrared to red light reflectance (850nm / 650nm reflectance ratio). For example, tests on the exterior walls of a residential building revealed that fresh organic dirt has a strong absorption peak at 365nm, while aged organic dirt exhibits a wider absorption band in the 450-532nm region; inorganic silicate dirt shows a significant reflection peak in the near-infrared band. The system matches these feature values ​​with preset spectral library templates to determine whether the dirt is organic (oil stains, biofilm), inorganic (dust, silicates), or a mixture. Simultaneously, it estimates the photochemical aging degree of organic dirt based on changes in the UV-Vis absorption ratio. The final output includes dirt type labels and quantitative indicators of aging degree data. This spectral feature-based dirt analysis method eliminates the need for complex high-resolution spectroscopic instruments, instead utilizing the differential responses across a limited wavelength band to extract key features, achieving accurate identification of common curtain wall dirt types and aging states.

[0045] In one embodiment of the present invention, the process of sequentially performing inner transmission channel multi-band excitation processing and outer reflection channel multi-band excitation processing on each target region in the target region grouping list to obtain the corresponding target region multi-band transmission-reflection raw data includes: Based on the height and azimuth coordinates of the target area grouping list, the preset band pool is filtered to obtain a complementary band sequence; For each band in the complementary band sequence, perform outer channel excitation processing and acquire the outer reflected signal and the inner transmitted signal to obtain the outer sampling matrix; Based on the timestamp data of the complementary band sequence and the completion of the outer channel excitation, a time offset is set, and inner channel excitation processing is performed on each band of the complementary band sequence to collect the outer reflection signal and the inner transmission signal, thereby obtaining the inner sampling matrix. The outer sampling matrix and the inner sampling matrix are paired at the band level to obtain the channel difference matrix. Data splicing is performed based on the channel difference matrix and complementary band sequence to obtain the original multi-band transmission-reflection data of the corresponding target area.

[0046] The following is a detailed description of the steps involved in the above embodiments: When filtering the preset band pool based on the elevation and azimuth coordinates of the target area group list, the system first reads the spatial location information of the target area. The preset band pool refers to the set of all available spectral bands and their parameters predefined by the system, including light source configuration information for multiple discrete wavelengths, covering the ultraviolet (365nm), visible (400-760nm), and near-infrared (760-1100nm) ranges. The system selects specific band combinations based on the height characteristics of the target area: for lower-level areas (0-30m), visible light (450nm, 532nm, 650nm) and near-infrared bands (850nm, 940nm) are prioritized, as these areas often contain oil and dust mixtures; for middle-level areas (30-100m), ultraviolet (365nm, 405nm) detection is added, as these areas often contain carbonate crystals and silicates; for higher-level areas (above 100m), long-wave near-infrared (1064nm) detection is incorporated, as these areas often contain VOCs photochemical products and aerosol condensation. Simultaneously, band adjustments are made based on azimuth information: near-infrared bands are prioritized for north-facing facades due to higher levels of biological contamination; while ultraviolet bands are enhanced for south-facing facades due to significant photoaging of organic contaminants. For example, for a target area 50 meters south of a commercial building, the system selects five characteristic bands—365nm, 450nm, 532nm, 650nm, and 850nm—for detection. The selected band set is organized into a complementary band sequence, i.e., a list of discrete wavelengths arranged in order from shortest to longest wavelength, with the intervals between bands satisfying the Nyquist theorem requirement for spectral feature sampling. This spatial location-based spectral band optimization method significantly improves the accuracy of dirt type identification while reducing unnecessary band measurements, saving energy and time.

[0047] When performing outer channel excitation processing and acquiring outer reflected and inner transmitted signals for each band in the complementary band sequence, the external working unit activates the corresponding wavelength light source sequentially according to the band sequence to illuminate the target area, while the two sides of the unit simultaneously acquire optical signals. Outer channel excitation processing refers to the process by which the external working unit emits light of a specific wavelength and triggers data acquisition. In practice, the multi-wavelength light source array (LED or semiconductor laser) on the external working unit is activated one by one through a control circuit. The emission power of each band light source is between 5-20mW, the pulse width is 50-200μs, and the repetition frequency is 500Hz. Taking the detection of rain streaks on the glass curtain wall of an office building as an example, when performing 532nm green light band detection, the external working unit first emits a 532nm laser beam to illuminate the target area, and simultaneously activates the high-speed sampling circuit. The photodetector array outside the window acquires the intensity of reflected light, while the photodiode inside the window simultaneously acquires the intensity of transmitted light. The system collects 100-200 sampling points for each band, averages the data to improve the signal-to-noise ratio, and records parameters such as wavelength, power, and sampling timestamp in the data header. Sampling data from all bands is stored in the outer sampling matrix according to wavelength index and spatial location. The outer sampling matrix is ​​a three-dimensional data structure containing information in three dimensions: wavelength, grid position, and light intensity. This outside-in transmission-reflection dual-channel synchronous measurement method fully utilizes the cooperative structure of the robot's internal and external bodies to acquire complete optical interaction information, providing a rich data foundation for dirt characteristic analysis.

[0048] Based on the complementary band sequence and the timestamp data of the outer channel excitation completion, a timing offset is set. When performing inner channel excitation processing and acquiring outer reflected and inner transmitted signals for each band in the complementary band sequence, the system switches to the inner working unit to emit the light source after the outer measurement is completed, repeating a similar measurement process. The timing offset refers to the time interval between the completion of the outer channel excitation and the start of the inner channel excitation, typically set to 100-200ms to ensure that the signals do not interfere with each other. The inner channel excitation processing refers to the process of the inner working unit emitting light of a specific wavelength and triggering data acquisition. In practice, the inner working unit emits beams of each wavelength sequentially according to the same band sequence as the outer measurement, but with a time difference. Taking the detection of dirt on the exterior wall of a hotel as an example, after the outer side completes the 450nm blue light measurement and records the timestamp as T1, the system calculates the inner start time T2 = T1 + 150ms, triggers the 450nm light source inside the window at time T2, illuminating the same target area, and simultaneously starts the signal acquisition circuits on both the inner and outer sides. The power of the inner light source is typically 10-15% higher than that of the outer light source to compensate for the initial reflection loss of incident light by the glass. The collected data includes reflected signals (received by the machine inside the window) and transmitted signals (received by the machine outside the window) generated by the inner light source, stored in the same data structure as the outer sampling as an inner sampling matrix. This alternating inner and outer bidirectional excitation measurement strategy provides optical characteristic data for the same contaminated area from two perspectives. By comparing the difference in response between the inner and outer excitations, the location of the contaminant (inner or outer side of the glass) and its optical characteristics can be determined more accurately.

[0049] When performing band-level pairing processing based on the outer and inner sampling matrices, the system matches the two sets of data according to wavelength and position information and calculates the difference. Band-level pairing processing refers to a data processing method that compares the signals generated by outer and inner excitations at the same wavelength in a one-to-one correspondence. In specific implementation, for each wavelength λ in the complementary band sequence, the corresponding data subset is extracted from the outer and inner sampling matrices, denoted as Out(λ) and In(λ) respectively. Each subset contains two parts: the reflected signal R and the transmitted signal T, thus there are four sets of data: the reflected signal R from the outer excitation... out(λ) and transmission T out(λ) The reflection R of the inner excitation in(λ) and transmission T in(λ) The system calculates the difference matrix of these four sets of data, such as the transmittance ΔT(λ) = T out(λ) / T in(λ) and reflectance ΔR(λ)=R out(λ) / R in(λ)For example, in a stubborn scale area on the curtain wall of a shopping mall, the transmittance measured at the 650nm red light band was 0.62 for the outer side and 0.78 for the inner side, resulting in a calculated transmittance of 0.79. This indicates that the scale on the outer side significantly blocks light at this wavelength. The difference data for all wavelengths were organized into a channel difference matrix, which contains all information about the wavelength, location, and the differences in the four signal values. This differential analysis method, comparing the inner and outer sides, effectively eliminates the interference of the optical properties of the glass itself, highlighting the characteristic response of the scale layer, and has a unique advantage, especially in distinguishing between inner and outer scale on double-sided glass.

[0050] When performing data stitching based on the channel difference matrix and complementary band sequences, the system organizes the data from each band into a unified data structure according to a specific format. Data stitching refers to a data processing method that integrates the measurement results of multiple discrete bands into a continuous spectral representation. In specific implementation, the system first constructs a structured data table by using the complementary band sequence as the horizontal axis (wavelength dimension) and the various indicators in the channel difference matrix as the vertical axis (response dimension). For each wavelength λ, six key values ​​are recorded: outer excitation reflection R... out(λ) , External excitation transmission T out(λ) Inner excitation reflection R in(λ) Inner excitation transmission T in(λ) Transmittance ΔT(λ) and reflectance ΔR(λ) are measured. For example, in analyzing dirt on the exterior walls of a hotel, the system selected five wavelength bands: 365nm, 450nm, 532nm, 650nm, and 850nm. For each band, the system recorded the aforementioned six values, forming a 5×6 data matrix. Furthermore, the system used spline interpolation to connect the discrete wavelength points to form a continuous spectral curve, obtaining a complete response spectrum covering the 365-850nm range. This complete spectral data, along with the original sampling points, constitutes the multi-band transmission-reflectance raw data. This data stitching method integrates the results of multiple dispersed measurements into a unified and intuitive spectral representation, facilitating subsequent peak-valley analysis and feature extraction, while preserving the accurate values ​​of the original measurements, ensuring the scientific validity and traceability of the data.

[0051] Please continue reading. Figure 1 Based on the dirt type data and aging degree data, modulated pulse laser is emitted to the target area, and the inner vibration signal and temperature rise signal are collected and jointly calculated to obtain dirt thickness data and adhesion strength data. In one embodiment of the present invention, the step of emitting modulated pulsed laser to the target area based on the dirt type data and aging degree data, collecting the inner vibration signal and temperature rise signal and performing joint calculation to obtain dirt thickness data and adhesion strength data includes: Based on the dirt type data and aging degree data, the repetition frequency, peak power and pulse width of the modulated pulse laser are set by parameter setting to obtain the pulse parameter set; Perform dual-level sequential excitation processing on the pulse parameter group to obtain the original inner vibration signal and the original inner temperature rise signal; The original inner vibration signal is subjected to frequency domain transformation to obtain vibration mode spectrum data, and the original inner temperature rise signal is subjected to time gradient calculation to obtain temperature rise peak data. By performing joint regression analysis on vibration modal spectrum data and temperature rise peak data, dirt thickness data and adhesion strength data are obtained.

[0052] The following is a detailed description of the steps involved in the above embodiments: When setting parameters for the repetition frequency, peak power, and pulse width of the modulated pulsed laser based on dirt type and aging data, the system automatically adjusts the laser parameters according to the dirt characteristics identified in the preceding steps. Parameter setting refers to the process of determining the most suitable laser pulse parameters for detection based on dirt characteristic data. In practice, the system first queries a preset parameter mapping table, which stores the correspondence between different dirt types and laser parameters. For organic dirt (such as oil stains and biofilms), a higher repetition frequency (800-1200Hz) and a lower peak power (10-15W) are set; for inorganic dirt (such as dust and silicates), a lower repetition frequency (300-600Hz) and a higher peak power (20-30W) are set; for mixed dirt, a medium frequency (600-800Hz) and power (15-20W) are used. Simultaneously, the pulse width is adjusted based on the aging level data: short pulse widths (50-100μs) are used for fresh dirt, medium pulse widths (100-200μs) for moderately aged dirt, and long pulse widths (200-300μs) for highly aged dirt. For example, for highly aged organic dirt detected on the exterior wall of an office building, the system pulse parameters are set as follows: repetition frequency 1000Hz, peak power 12W, and pulse width 250μs. These parameters together constitute the pulse parameter set, which is a set of laser control parameters used for subsequent excitation processing. Setting laser parameters suitable for the characteristics of various types of dirt ensures the acquisition of optimal vibration and thermal response signals, generating sufficiently strong excitation while avoiding damage to the glass or producing a cleaning effect, thus improving the accuracy and safety of detection.

[0053] When performing dual-level sequential excitation processing on the pulse parameter set, the system emits two-stage pulse sequences towards the target area in a low-energy-high-energy order and measures the response signal on the inside of the glass. Dual-level sequential excitation processing refers to a step-by-step measurement method where a low-energy probe pulse is emitted first to measure the reference response, and then a main test pulse of the designed energy is emitted to obtain the complete response. Specifically, the external working unit first sets the laser power to 30% of the peak power in the pulse parameter set and emits a low-energy pulse sequence (probe level) lasting 100ms. Simultaneously, the piezoelectric accelerometer (sensitivity 10mV / g) of the internal working unit records the vibration signal, and the thermocouple (response time <10ms) records the temperature change. After waiting 200-300ms, the external working unit increases the power to the set value and emits a main test pulse sequence (main excitation level) lasting 200ms, recording the vibration and temperature signals again. For example, when detecting water stains and dirt on the curtain wall of a commercial building, the system first emits a 5W test pulse to obtain a glass vibration reference, and then emits a 18W main test pulse to record the complete vibration and temperature rise response. The signals generated by the test pulse and the main test pulse are preprocessed and then combined into the original inner vibration signal and the original inner temperature rise signal. This dual-level sequential measurement method avoids the nonlinear effects and thermal saturation problems that may arise from direct high-power excitation, while providing a clear system reference response. This helps to separate the characteristics of the dirt from the inherent system response, improving the signal-to-noise ratio and accuracy of the measurement.

[0054] The system performs frequency domain transformation on the original vibration signal to obtain vibration modal spectrum data. When performing time gradient calculation on the original temperature rise signal, the system performs different mathematical transformations on the two types of signals to extract their characteristic information. Frequency domain transformation refers to a data processing method that converts time-series signals into data with frequency distribution characteristics. In specific implementation, the system applies the Fast Fourier Transform (FFT) algorithm to the original vibration signal, decomposing the vibration signal in the 0-2000Hz range into amplitude distributions of different frequency components. It mainly focuses on three characteristic frequency bands: low frequency (50-200Hz) corresponding to the overall vibration mode, mid-frequency (200-800Hz) corresponding to the local vibration mode, and high frequency (800-2000Hz) corresponding to the micro-vibration mode. The energy distribution of these frequency bands constitutes the vibration modal spectrum data. Time gradient calculation refers to a data processing method that calculates the rate of temperature change over time. For the raw temperature rise signal, the system calculates the temperature derivative (ΔT / Δt) at each time point to obtain the temperature change rate curve. From this curve, characteristic values ​​such as the maximum rate of change, the time to reach the peak, and the decay time constant are extracted to form the temperature rise peak data. For example, in the detection of dirt on the exterior wall of a high-rise residential building, vibration analysis revealed a significant resonance peak at 350Hz, and temperature rise analysis showed a maximum heating rate of 0.12℃ / s. These characteristic values ​​are directly related to the physical properties of the dirt. Frequency domain analysis can separate vibration characteristics at different scales, while time gradient analysis can reveal heat conduction characteristics. The combination of the two provides a multi-dimensional characterization of the dirt's physical parameters, significantly improving the accuracy of subsequent thickness and adhesion strength calculations.

[0055] When performing joint regression analysis based on vibration modal spectrum data and temperature rise peak data, the system inputs these two sets of characteristic data into a statistical algorithm to derive the physical parameters of the fouling. Joint regression analysis refers to a data processing method that uses multi-source characteristic data to jointly solve for target parameters. In specific implementation, the system, based on a pre-calibrated physical model, uses a multiple linear regression algorithm to combine the characteristic quantities of the vibration modal spectrum and temperature rise peak. Fouling thickness is mainly inverted through three indicators: mid-frequency vibration energy ratio, maximum temperature rise, and temperature rise peak time; adhesion strength is mainly solved through high-frequency vibration attenuation rate, low-frequency resonance peak half-width at half-maximum, and temperature decay time constant. The system substitutes these characteristic quantities into the regression equation to calculate the fouling thickness value in micrometers and the adhesion strength value in Pascals. For example, the inspection results of the exterior wall of an office building show that the vibration mode has energy concentration at 450Hz, with a temperature rise peak of 0.8℃. Regression calculations show that the fouling thickness in this area is 125μm, and the adhesion strength is 8.5kPa, indicating a medium-thickness and relatively firmly adhered fouling layer. Thickness ranges are typically categorized as thin (<50μm), medium (50-200μm), and thick (>200μm), while adhesion strength is classified into three levels: loose (<5kPa), medium (5-15kPa), and strong (>15kPa). This vibration-thermal coupling calculation method overcomes the limitations of traditional single measurement methods, significantly improving the accuracy and reliability of non-contact dirt parameter measurement through complementary verification of different physical quantities.

[0056] In one embodiment of the present invention, the step of performing dual-level sequential excitation processing on the pulse parameter group to obtain the original inner vibration signal and the original inner temperature rise signal includes: Based on the pulse parameter set, glass structure data and magnetic gap data, the power of the test level, the power of the main excitation level and the time interval between levels are proportionally calculated to obtain the level sequence parameters. According to the energy level sequence parameters, test energy level pulses are emitted to the target area and vibration and temperature rise signals are collected simultaneously to obtain test response data. The glass-air cavity resonant frequency data is calculated based on the test response data, and the parameters of the main excitation level power and the time interval between levels are revised to obtain the revised level sequence parameters. According to the revised energy level sequence parameters, the main excitation energy level pulse is emitted to the target region and the vibration signal and temperature rise signal are collected simultaneously. The test response data and the main excitation response data are merged and processed to obtain the original inner vibration signal and the original inner temperature rise signal.

[0057] The following is a detailed description of the steps involved in the above embodiments: When performing proportional calculations on the test level power, main excitation level power, and inter-level time intervals based on pulse parameter sets, glass structure data, and magnetic gap data, the system comprehensively considers multiple factors to precisely adjust the laser parameters. Glass structure data refers to physical parameters such as glass type, thickness, and laminated structure; magnetic gap data refers to the actual distance between the inner and outer bodies of the window cleaning robot. Proportional calculation is a data processing method that adjusts laser parameters proportionally based on physical characteristics. In practice, the system first sets the test level power to 30-40% of the peak power in the pulse parameter set. For example, for a peak power setting of 20W, the test level power is set to 7W. For different types of glass, the system adjusts according to the thickness ratio: if the standard condition is 6mm tempered glass, the test power is increased by 33% when facing 8mm thick glass; when facing insulated or laminated glass, it is increased by another 20-25% to compensate for additional attenuation. The magnetic gap data also affects the power setting: when the gap increases to 1.5 times the standard value (usually 3-5 mm), the power is increased by 15% to compensate for the attenuation of sound waves and heat conduction. The main excitation level power is maintained at the original peak power in the pulse parameter set, and the time interval between energy levels is set to 150-250 ms, with the specific value determined according to the glass thickness: for every 1 mm increase in thickness, the time interval increases by 30 ms. For example, a shopping mall uses 10 mm tempered glass with a magnetic gap of 7 mm. For a pulse parameter set with an original peak power of 25 W, the system calculates the test level power to be 12 W, the main excitation level power to be maintained at 25 W, and the time interval between energy levels to be set to 240 ms. These parameters together constitute the energy level sequence parameters, i.e., the complete set of execution parameters for dual-level excitation. This parameter adjustment based on physical characteristics ensures consistent and reliable measurement signals under different glass structures and machine operating conditions, effectively solving the measurement deviation problem caused by changes in the field environment.

[0058] When the system emits probe energy level pulses to the target area according to the energy level sequence parameters and simultaneously acquires vibration and temperature rise signals, it first performs low-power detection to obtain the system response baseline. The probe energy level pulse refers to a low-power preliminary excitation laser pulse sequence used to measure the system's fundamental response without producing significant thermal effects. In specific implementation, the laser on the external working machine emits a pulse sequence with a duration of 50-100 ms according to the probe energy level power (usually 30-40% of the main excitation energy level) and predetermined pulse parameters (repetition frequency, pulse width) to irradiate the target contaminated area. Simultaneously, the sensor system on the internal working machine starts data acquisition: a piezoelectric accelerometer (typical sensitivity 10mV / g) records the micro-vibration signal of the glass at a sampling rate of 10kHz; a miniature thermocouple or infrared temperature sensor (resolution 0.1℃) records the temperature change of the inner surface at a sampling rate of 100Hz. For example, in the detection of oil stains on the exterior wall of an office building, the system emitted a sequence of probe pulses with a power of 5W, a repetition frequency of 500Hz, and a pulse width of 150μs. It recorded a vibration signal with a peak value of approximately 0.05g and a temperature rise signal of less than 0.2℃. The acquired raw vibration waveform, spectral data, and temperature time series together constitute the probe response data. The key to probe energy level excitation is that the power is low enough to avoid a cleaning effect or significant heat accumulation on the dirt, while simultaneously generating a detectable weak response to provide a reference benchmark for the subsequent main excitation level, distinguishing the system's inherent characteristics from the characteristics of the dirt.

[0059] Based on the test response data, the glass-air cavity resonant frequency data is calculated. When revising the parameters of the main excitation level power and the time interval between levels, the system analyzes the system response characteristics under low-power excitation and optimizes the main measured parameters. The glass-air cavity resonant frequency data refers to the natural vibration frequency information of the air gap system formed between the glass and the inner and outer casings. In specific implementation, the system performs spectral analysis on the vibration signal in the test response, extracting the main resonant peak frequencies f1, f2, ..., f... within the 0-1000Hz range. nThese resonant frequencies, along with their corresponding amplitudes, reflect the inherent characteristics of the glass-air cavity system, not the characteristics of the fouling. Based on this frequency data, the system revises the main excitation level parameters: if a strong resonant frequency close to the pulse repetition frequency (e.g., within ±10%) exists, the main excitation level repetition frequency is adjusted to offset this resonant region; if a strong low-frequency resonance is detected (e.g., in the 50-100Hz range), the time interval between energy levels is increased by 50ms to ensure sufficient system decay. For example, in a test of a hotel's double-glazed window, the probe response showed a significant resonance peak near 350Hz, and the system adjusted the original 360Hz main excitation level repetition frequency to 400Hz. The temperature rise signal is also used for parameter revision: if the probe energy level has generated a temperature rise exceeding 0.3℃, the main excitation level power is reduced by 10-15% to avoid overheating. These adjusted parameters together constitute the revised energy level sequence parameters. This parameter adaptive adjustment based on measured response greatly improves the accuracy and safety of measurements, avoiding the risks of system resonance and overheating that may result from blindly using preset parameters. It is particularly suitable for measurement scenarios of modern building curtain walls with complex and varied structures.

[0060] The system emits master excitation level pulses to the target area according to the revised energy level sequence parameters, and simultaneously collects vibration and temperature rise signals. When merging the trial response data and master excitation response data, the system performs excitation measurement of the formal power and integrates the data from before and after. The master excitation level pulse refers to a high-power formal measurement laser pulse sequence used to generate a significant response signal related to fouling characteristics. In practice, after completing the trial energy level measurement and waiting for the energy level interval (usually 150-250ms), the laser on the external working machine irradiates the same target area with a pulse sequence of 100-200ms duration, according to the revised master excitation level power (usually the full design power) and optimized pulse parameters. The sensor system on the internal working machine restarts data acquisition, recording the vibration and temperature rise responses generated by the master excitation level. Merging processing refers to the method of splicing the two sets of measurement data in a time sequence and performing signal processing. The system performs time alignment, amplitude normalization, and splicing on the two sets of signals, while simultaneously removing invalid data during the energy level interval. For example, in a test of the exterior wall of an office building, the test level (6W) generated a vibration signal of 0.06g, and the main excitation level (20W) generated a vibration signal of 0.26g and a temperature rise of 0.8℃. The system merged these data in time sequence and marked the energy level transition points. The merged complete data sequence constitutes the original vibration signal and the original temperature rise signal of the inner side. These two sets of signals contain the full range of response characteristics of the system under low-energy to high-energy excitation. This dual-level sequential measurement method obtains a more complete dynamic response process of the system, including both the linear response characteristics at low energy and the nonlinear response characteristics closely related to the physical parameters of the fouling at high energy, significantly improving the accuracy and reliability of subsequent parameter inversion.

[0061] Please continue reading. Figure 1 Based on the dirt type data, aging degree data, dirt thickness data, and adhesion strength data, the laser parameters for trial cleaning are set, and trial cleaning is carried out on the target area. In one embodiment of the present invention, the step of setting trial cleaning laser parameters based on the dirt type data, aging degree data, dirt thickness data, and adhesion strength data, and performing trial cleaning on the target area, includes: Based on the dirt type data, aging degree data, dirt thickness data, adhesion strength data, glass curvature distribution matrix and magnetic gap data, power coefficient calculation, pulse width coefficient calculation and scanning speed coefficient calculation are performed to obtain the test cleaning laser parameter set; The target area is subjected to trial cleaning treatment according to the trial cleaning laser parameter set, and real-time transmittance curve data is collected simultaneously to obtain the transmittance change curve. The transmittance increment data is calculated based on the transmittance change curve, and the transmittance increment data is compared with the preset transmittance threshold to obtain the laser parameter difference coefficient. The laser parameter group for the trial cleaning is revised based on the laser parameter difference coefficient to obtain laser parameter correction data.

[0062] The following is a detailed description of the steps involved in the above embodiments: When calculating power coefficient, pulse width coefficient, and scanning speed coefficient based on data such as dirt type, aging degree, dirt thickness, adhesion strength, glass curvature distribution matrix, and magnetic gap, the system comprehensively analyzes the aforementioned test results to calculate the most suitable laser parameters for the trial cleaning. Power coefficient calculation refers to a data processing method that determines the laser output power adjustment ratio based on dirt characteristics; pulse width coefficient calculation refers to a data processing method that determines the laser pulse duration adjustment ratio based on dirt characteristics; and scanning speed coefficient calculation refers to a data processing method that determines the laser beam movement rate adjustment ratio based on dirt characteristics. In specific implementation, the system first selects a reference power based on dirt type data: 15-25W for organic dirt, 25-35W for inorganic dirt, and 20-30W for mixed dirt. Then, several correction factors are applied: Aging level data affects the power factor, with highly aged dirt having a power factor of 1.2-1.3 times, moderately aged dirt 1.1 times, and fresh dirt 1.0 times; dirt thickness data affects the pulse width factor, increasing by 0.1 times for every 50μm increase in thickness; adhesion strength data affects both power and pulse width factor, increasing the power factor by 0.1 times and the pulse width factor by 0.05 times for every 5kPa increase in adhesion strength. A glass curvature distribution matrix is ​​used for local power adjustment: for every 0.001mm change in curvature value... -1The power coefficient is adjusted by ±5% to compensate for focal length variations. The magnetic gap data affects the scanning speed coefficient: for every 1mm increase in gap, the speed coefficient decreases by 10% to ensure energy transfer. For example, the inspection area on the exterior wall of a high-rise building contains moderately aged organic dirt, 150μm thick, with an adhesion strength of 12kPa and a local curvature of 0.002mm. -1 With a magnetic gap of 6mm, the calculated parameters are: base power 20W × aging coefficient 1.1 × adhesion coefficient 1.24 × curvature correction 1.1 = 30W; pulse width 200μs × thickness coefficient 1.3 × adhesion coefficient 1.12 = 292μs; scanning speed 10mm / s × gap coefficient 0.7 = 7mm / s. These three parameters together constitute the parameter set for the test cleaning laser. This parameter optimization calculation method based on multi-dimensional dirt characteristics ensures that the test cleaning laser can accurately act on the target dirt, producing a detectable cleaning effect while avoiding the risk of glass damage caused by excessive energy.

[0063] When performing trial cleaning on the target area according to the trial cleaning laser parameter set and simultaneously collecting real-time transmittance curve data, the system applies laser energy lower than that for complete cleaning within a small area at the edge of the soiled area, while monitoring changes in light transmittance. Trial cleaning refers to a small-area test cleaning operation performed on the edge or a localized area of ​​the target area before formal cleaning. Specifically, the external workpiece sets the laser parameters to the power, pulse width, and scanning speed specified in the trial cleaning laser parameter set, selecting a 1-2cm edge of the target area. 2 The system performs scanning and cleaning on small areas. Importantly, the trial cleaning power is set to 75-85% of the theoretical optimal power to ensure partial cleaning rather than complete removal of dirt. Simultaneously, the photodetector (typically a silicon photodiode) on the indoor workpiece records the transmitted light intensity in real time at a sampling rate of 50-100Hz, while the reference light source (typically a low-power LED) on the outdoor workpiece provides stable incident light. For example, for a limescale area on the exterior wall of an office building, the system uses parameters of 28W power, 260μs pulse width, and 6mm / s scanning speed at a 1.5cm edge. 2 A trial cleaning was performed on the area for 5 seconds, while the transmittance was recorded as it gradually increased from an initial value of 0.58 to 0.72. The collected time-transmittance data points formed a transmittance change curve. This small-area trial cleaning method cleverly transforms traditional parameter prediction into practical verification, directly observing the actual response of dirt to the laser, rather than relying solely on theoretical estimates. This significantly improves the accuracy of parameters and the predictability of results for subsequent large-area cleaning.

[0064] The system calculates the incremental transmittance data based on the transmittance change curve and performs a difference operation between the incremental transmittance data and a preset transmittance threshold. The system then analyzes the trial cleaning effect and compares it with the ideal target. The incremental transmittance data refers to the change in transmittance before and after the trial cleaning; the preset transmittance threshold refers to the target transmittance value that should be achieved under the ideal cleaning effect; the difference operation refers to the data processing method that compares the actual increment with the target increment. In specific implementation, the system first calculates the initial transmittance value T. initial (Transmittance before cleaning) and final value T final (Transmittance at the end of the cleaning test) yields the transmittance increment ΔT = T final -T initial Then, based on the type of dirt and its location, the preset transmittance threshold T is queried. target This threshold is typically set to 95% of the transmittance of the dirt-free area. The ratio of the actual increment to the target increment is calculated as ratio = ΔT / (T target -T initial This ratio reflects the deviation between the trial cleaning energy and the ideal cleaning energy. A ratio less than 0.5 indicates severely insufficient cleaning energy; a ratio between 0.5 and 0.8 indicates moderate cleaning energy but needs enhancement; a ratio between 0.8 and 1.0 indicates near-optimal cleaning energy; and a ratio exceeding 1.0 indicates excessive cleaning energy. The system uses the calculated ratio as the laser parameter difference coefficient. For example, in a trial cleaning test of a shopping mall's curtain wall, the initial transmittance was 0.55, rising to 0.68 after the trial cleaning, while the preset transmittance threshold was 0.85. The calculated difference coefficient was 0.43, indicating insufficient cleaning energy. This step establishes a quantitative relationship between laser parameters and cleaning effect through the intuitive physical quantity of transmittance, transforming abstract dirt characteristics into specific energy requirements. This provides a reliable experimental basis for parameter optimization, avoiding the inaccuracies of purely empirical or theoretical estimations.

[0065] When revising the parameters of the test cleaning laser parameter group based on the laser parameter difference coefficient, the system precisely adjusts the initial parameters according to the actual cleaning effect. Parameter revision processing refers to a data processing method that proportionally adjusts the initial parameters based on the test results. In practice, the system modifies parameters according to the following rules: If the difference coefficient is less than 0.5 (insufficient cleaning), the power increases by (1 / difference coefficient) × 0.5 times, the pulse width increases by (1 / difference coefficient) × 0.3 times, and the scanning speed decreases by (1 / difference coefficient) × 0.2 times; if the difference coefficient is between 0.5 and 0.8 (slightly insufficient cleaning), the power increases by (0.9 / difference coefficient) × 0.3 times, the pulse width increases by (0.9 / difference coefficient) × 0.2 times, and the scanning speed decreases by (0.9 / difference coefficient) × 0.1 times; if the difference coefficient is between 0.8 and 1.0 (close to optimal), the power increases by (1 / difference coefficient - 1) × 0.2 times, and the remaining parameters remain unchanged; if the difference coefficient is greater than 1.0 (over-cleaning), the power decreases by (difference coefficient - 1) × 0.3 times, and the scanning speed increases by the difference coefficient × 0.2 times. For example, in the aforementioned case of a shopping mall facade with a difference coefficient of 0.43, the system adjusted the trial cleaning laser power from 28W to 28W×(1+0.5×(1 / 0.43))=60.5W, the pulse width from 260μs to 260μs×(1+0.3×(1 / 0.43))=441μs, and the scanning speed from 6mm / s to 6mm / s×(1-0.2×(1 / 0.43))=3.2mm / s. The revised power, pulse width, and scanning speed values ​​constitute the laser parameter correction data. This parameter closed-loop optimization method based on actual measurement results breaks through the limitations of traditional fixed parameter strategies, achieving adaptive cleaning capabilities for various complex dirt types, while avoiding energy waste and the risk of over-cleaning. It is particularly suitable for handling the complex and varied dirt conditions on the exterior walls of high-rise buildings.

[0066] In one embodiment of the present invention, after obtaining the laser parameter correction data, the laser cleaning window cleaning method further includes: Based on the laser parameter correction data, target area data, glass curvature distribution matrix and robot remaining energy data, power ratio processing, pulse width ratio processing and scanning speed ratio processing are performed to obtain the regional energy demand matrix. Based on the regional energy demand matrix and the in-window-out-window scanning sequence data, a path sequence optimization process is performed to obtain a set of cleaning parameters. The cleaning parameter set is processed by queue allocation to obtain a queue of cleaning parameters to be executed and the queue of cleaning parameters to be executed is output.

[0067] The following is a detailed description of the steps involved in the above embodiments: When performing power allocation, pulse width allocation, and scanning speed allocation based on laser parameter correction data, target area data, glass curvature distribution matrix, and robot remaining energy data, the system performs global optimization configuration of laser cleaning parameters. Target area data refers to the actual surface area of ​​each soiled area; robot remaining energy data refers to the remaining battery power and supported working time of the window cleaning robot. Power allocation is the process of rationally allocating laser power according to the characteristics of each area; pulse width allocation is the process of rationally allocating the laser pulse duration according to the characteristics of each area; scanning speed allocation is the process of rationally allocating the laser movement speed according to the characteristics of each area. In specific implementation, the system first calculates the energy requirement per unit area: for each target area, the total energy required for cleaning is calculated based on laser parameter correction data and area data. For example, if an office building exterior wall has three soiled areas, the first area requires 25W × 300μs × 5mm. 2 =37.5mJ / mm 2 The calculations are similar for other areas. Then, a glass curvature correction is applied to each area: for every 0.001mm change in curvature value... -1 The energy demand is adjusted by 5%. Then, a global adjustment is made based on the robot's remaining energy limit: if the total energy demand exceeds 80% of the remaining energy, priority is given to ensuring parameters in heavily polluted areas, while power is reduced by 3-10% in lightly polluted areas. The results of these three ratios are organized into a regional energy demand matrix, which includes the final laser power, pulse width, and scanning speed information for each region. This multi-parameter global ratio processing maximizes the cleaning effect under limited energy conditions, avoiding the energy efficiency problems that may result from adjusting a single parameter.

[0068] When optimizing the path sequence based on the regional energy demand matrix and the in-window-outside scanning sequence data, the system calculates the most efficient cleaning execution path. The in-window-outside scanning sequence data refers to the movement path planning information when the internal and external cleaning units work together; path sequence optimization refers to the process of calculating the optimal movement trajectory based on energy demand and the cleaning unit's position. In specific implementation, the system uses an improved greedy algorithm to calculate the cleaning unit's movement path: first, the target area is layered vertically, and the shortest total path is selected from top to bottom or bottom to top; within the same layer, the units are sorted according to the principle of shortest horizontal distance; for transitions between adjacent areas, the most energy-efficient speed change curve is calculated. The internal and external cleaning unit positions remain synchronized but offset by a fixed distance (usually 3-5 times the glass thickness) to ensure stable magnetic coupling. For example, when cleaning a hotel's glass curtain wall, the system divides 25 soiled areas into 5 layers, executing each layer from left to right, with a zigzag path transition between layers, reducing unnecessary back-and-forth movements and lowering energy consumption by 25%. The optimized cleaning parameter set is an ordered data structure containing parameters such as the start and end positions, movement speed, laser power, and pulse width for each cleaning path segment. This path optimization process based on physical layout significantly improves cleaning efficiency, reduces ineffective movement and energy consumption of the machine, and extends the cleaning area per charge.

[0069] When performing queue allocation processing on the cleaning parameter set, the system converts the optimized parameters into a sequence of instructions that the execution devices can directly execute. Queue allocation processing refers to the process of reorganizing the cleaning parameters according to the hardware execution logic and timing requirements. In specific implementation, the system classifies the cleaning parameter set according to the subsystems of the working body: the motion control queue contains motor speed, position, and acceleration instructions, discretized at 10ms intervals; the laser control queue contains power switch, modulation frequency, and pulse width instructions, discretized at 5ms intervals; and the sensor feedback queue contains the sampling start / stop time and frequency settings of each detection point, discretized at 20ms intervals. The three types of queues are synchronized through timestamps to form a queue of cleaning parameters to be executed. For example, before the cleaning of the curtain wall of a business center begins, the system generates an execution queue containing 2,500 motion instructions, 3,600 laser control instructions, and 1,800 sensor instructions, with a total execution time of approximately 15 minutes, covering 32 soiled areas. The parameter queue is output to each execution unit, including the stepper motor driver, laser control board, and data acquisition card, through the industrial control computer. This queue allocation process enables the serialized execution of complex cleaning tasks, allowing high-level cleaning strategies to be translated into specific instructions executable by the underlying hardware, ensuring that the entire cleaning process proceeds in an orderly manner according to precisely planned parameters and paths.

[0070] Another embodiment of this application proposes a laser-based window cleaning robot, which employs the laser-based window cleaning method of any of the above embodiments, and therefore has the beneficial effects brought about by any of the above embodiments, which will not be repeated here.

[0071] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method of laser decontamination of a window, characterized in that, The laser-based window cleaning method is applied to a laser-based window cleaning robot, which includes an inner body and an outer body located on both sides of the glass. The window cleaning method includes: A low-energy probe light is used to perform a full-coverage scan on the glass surface, and the reflected signal from the outside and the transmitted signal from the inside are collected and differentially calculated to obtain the optical density distribution data of the dirt; wherein, the low-energy probe light is a low-power collimated laser with a power of 5mW to 10mW; Based on the dirt optical density distribution data, target areas with optical density exceeding a preset threshold are selected. Multi-band response measurements are performed on the target areas, and the transmission-reflection response curves are analyzed to obtain dirt type data and aging degree data. The multi-band response measurement of the target areas includes: irradiating the target areas with multiple probe lights of different wavelengths and collecting the transmission and reflection signals of the target areas in each band. Based on the dirt type data and aging degree data, modulated pulse laser is emitted to the target area, and the inner vibration signal and temperature rise signal are collected and jointly calculated to obtain dirt thickness data and adhesion strength data. The dirt type data includes dirt type labels used to characterize whether the dirt is organic, inorganic or mixed. The aging degree data includes quantitative indicators of aging degree used to characterize the photochemical aging degree of organic dirt. Based on data on dirt type, aging degree, dirt thickness, adhesion strength, glass curvature distribution matrix, and magnetic gap, power coefficient, pulse width coefficient, and scanning speed coefficient are calculated to obtain a set of laser parameters for the test cleaning process. The glass curvature distribution matrix is ​​a data matrix obtained by curvature calculation based on outer and inner optical path displacement data. The magnetic gap data refers to the actual distance between the inner and outer parts of the unit. The target area is subjected to trial cleaning treatment according to the trial cleaning laser parameter set, and real-time transmittance curve data is collected simultaneously to obtain the transmittance change curve. The transmittance increment data is calculated based on the transmittance change curve, and the transmittance increment data is compared with the preset transmittance threshold to obtain the laser parameter difference coefficient. The laser parameter group for the trial cleaning is revised based on the laser parameter difference coefficient to obtain laser parameter correction data.

2. The laser-based window cleaning method according to claim 1, characterized in that, The process involves performing a low-energy probe light full-coverage scan on the glass surface, acquiring the outer reflected signal and the inner transmitted signal, and performing differential calculations to obtain dirt optical density distribution data, including: The probe light scanning path is divided into three-dimensional grids to obtain a scanning coordinate matrix containing grid coordinate data and scanning timestamp data. The outer reflected light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the reflected light intensity matrix and the outer optical path displacement data corresponding to the scanning coordinate matrix. The inner transmitted light signal is synchronously acquired and processed according to the scanning coordinate matrix to obtain the transmitted light intensity matrix and inner optical path displacement data corresponding to the scanning coordinate matrix. Based on the outer optical path displacement data and the inner optical path displacement data, curvature calculation processing is performed to obtain the glass curvature distribution matrix; Based on the glass curvature distribution matrix, the reflected light intensity matrix and the transmitted light intensity matrix are subjected to thickness-curvature compensation processing to obtain the compensated optical response matrix. Based on the compensated optical response matrix, grid cells with transmitted light intensity located in the preset high transmission percentile range are selected for reference ratio calculation to obtain dirt optical density distribution data.

3. The laser-based window cleaning method according to claim 2, characterized in that, The curvature calculation process based on the outer optical path displacement data and the inner optical path displacement data to obtain the glass curvature distribution matrix includes: The outer optical path displacement data and the inner optical path displacement data are subjected to grid-by-grid difference calculation to obtain an optical path difference matrix; wherein, the grid-by-grid difference calculation of the outer optical path displacement data and the inner optical path displacement data includes: matching the outer optical path displacement data and the inner optical path displacement data according to the corresponding grid coordinates, and calculating the difference between the outer optical path displacement value and the inner optical path displacement value corresponding to each grid position; Based on the optical path difference matrix, a first-order gradient operation is performed on the difference between adjacent grids to obtain the slope matrix of the glass surface. The glass curvature distribution matrix is ​​obtained by performing second-order gradient calculations on the slopes of adjacent grids based on the glass surface slope matrix.

4. The laser-based window cleaning method according to claim 1, characterized in that, The step involves selecting target areas with optical density exceeding a preset threshold based on the dirt optical density distribution data, performing multi-band response measurements on the target areas, analyzing the transmission-reflection response curves, and obtaining dirt type data and aging degree data, including: Based on the dirt optical density distribution data, connected component clustering is performed on continuous grids with optical density exceeding a preset threshold to obtain a target region set. The target area set is grouped based on its height and azimuth coordinates to obtain a target area group list. For each target region in the target region grouping list, perform multi-band excitation processing of the inner transmission channel and multi-band excitation processing of the outer reflection channel in sequence to obtain the corresponding target region multi-band transmission-reflection raw data; Based on the original multi-band transmission-reflection data, the boundary grid of the target area is locally normalized to obtain multi-band transmission-reflection correction data. Based on the multi-band transmission-reflection correction data, peak-valley morphological features of each band curve are extracted to obtain dirt type data and aging degree data.

5. The laser-based window cleaning method according to claim 4, characterized in that, The inner transmission channel multi-band excitation processing and the outer reflection channel multi-band excitation processing are sequentially performed on each target region in the target region grouping list to obtain the corresponding target region multi-band transmission-reflection raw data, including: Based on the height and azimuth coordinates of the target area grouping list, the preset band pool is filtered to obtain a complementary band sequence; For each band in the complementary band sequence, perform outer channel excitation processing and acquire the outer reflected signal and the inner transmitted signal to obtain the outer sampling matrix; Based on the timestamp data of the complementary band sequence and the completion of the outer channel excitation, a time offset is set, and inner channel excitation processing is performed on each band of the complementary band sequence to collect the outer reflection signal and the inner transmission signal, thereby obtaining the inner sampling matrix. The outer sampling matrix and the inner sampling matrix are paired at the band level to obtain the channel difference matrix; wherein, the band-level pairing process based on the outer sampling matrix and the inner sampling matrix includes: extracting corresponding data subsets from the outer sampling matrix and the inner sampling matrix according to the same wavelength in the complementary band sequence, and comparing the signals generated by the outer excitation and the inner excitation under the same wavelength. Data splicing is performed based on the channel difference matrix and complementary band sequence to obtain the original multi-band transmission-reflection data of the corresponding target area.

6. The laser-based window cleaning method according to claim 1, characterized in that, The process involves emitting modulated pulsed laser light towards the target area based on the dirt type and aging data, collecting internal vibration and temperature rise signals, and performing joint calculations to obtain dirt thickness and adhesion strength data, including: Based on the dirt type data and aging degree data, the repetition frequency, peak power and pulse width of the modulated pulse laser are set by parameter setting to obtain the pulse parameter set; The pulse parameter set is subjected to dual-level sequential excitation processing to obtain the original inner vibration signal and the original inner temperature rise signal; wherein, the dual-level sequential excitation processing refers to, based on the pulse parameter set, first emitting a test level pulse and collecting the vibration signal and temperature rise signal, and then emitting a main excitation level pulse and collecting the vibration signal and temperature rise signal. The original inner vibration signal is subjected to frequency domain transformation to obtain vibration mode spectrum data, and the original inner temperature rise signal is subjected to time gradient calculation to obtain temperature rise peak data. By performing joint regression analysis on vibration modal spectrum data and temperature rise peak data, dirt thickness data and adhesion strength data are obtained.

7. The laser-based window cleaning method according to claim 6, characterized in that, The process of performing dual-level sequential excitation on the pulse parameter group to obtain the original inner vibration signal and the original inner temperature rise signal includes: Based on the pulse parameter set, glass structure data and magnetic gap data, the power of the test level, the power of the main excitation level and the time interval between levels are proportionally calculated to obtain the level sequence parameters. According to the energy level sequence parameters, test energy level pulses are emitted to the target area and vibration and temperature rise signals are collected simultaneously to obtain test response data. The glass-air cavity resonant frequency data is calculated based on the test response data, and the parameters of the main excitation level power and the time interval between levels are revised to obtain the revised level sequence parameters. According to the revised energy level sequence parameters, the main excitation energy level pulse is emitted to the target region and the vibration signal and temperature rise signal are collected simultaneously. The test response data and the main excitation response data are merged and processed to obtain the original inner vibration signal and the original inner temperature rise signal.

8. The laser-based window cleaning method according to claim 1, characterized in that, After obtaining the laser parameter correction data, the laser-based window cleaning method further includes: Based on laser parameter correction data, target area area data, glass curvature distribution matrix, and robot remaining energy data, power matching, pulse width matching, and scanning speed matching are performed to obtain the regional energy demand matrix; wherein, the robot remaining energy data refers to the remaining battery power and the working time that the laser cleaning and window cleaning robot can support. Based on the regional energy demand matrix and the in-window-out-window scanning sequence data, a path sequence optimization process is performed to obtain a set of cleaning parameters. The cleaning parameter set is processed by queue allocation to obtain a queue of cleaning parameters to be executed and the queue of cleaning parameters to be executed is output.

9. A laser-based window cleaning robot, characterized in that, The laser-cleaning window cleaning robot uses the laser-cleaning window cleaning method as described in any one of claims 1 to 8.

Images