A method, system, and medium for dynamic laser delamination of multilayer heterogeneous coatings based on spectral signatures

Abstract

The application discloses a kind of based on spectral feature multilayer hetero-coating laser layering dynamic removal method, system and medium, belong to the technical field of laser cleaning, the spectrometer data of different coating in laser paint removal is collected, and cross-coating dataset is obtained by division;Based on the spectrometer data of different coating, the Gaussian fitting curve of different coating characteristic element is determined by fitting, to identify the coating information of current processing;Based on cross-coating dataset, a deep learning model is trained to obtain a cross-coating prediction model to predict whether the coating currently being processed will cross the coating and provide early warning information;Based on the coating information, configure the laser processing parameters of the corresponding coating layer;If the early warning information that the coating will cross the coating is predicted, configure the laser processing parameters of the next layer based on the early warning information, to achieve precise layering removal and avoid damaging the substrate or underlying coating.The application efficiently realizes real-time adaptive control of laser process parameters, and has good practicability.

Description

Technical Field

[0001] This invention belongs to the technical field of laser cleaning, specifically relating to a method, system, and medium for dynamic laser-based layered removal of multilayer heterogeneous coatings based on spectral characteristics. Background Technology

[0002] Laser cleaning, as a novel cleaning technology, has become a potential application for paint removal from aircraft aluminum alloy and composite material skins due to its advantages such as non-contact operation, environmental friendliness, high controllability, and automation. To achieve the cleaning objective without damaging the substrate, real-time monitoring and closed-loop control are necessary to monitor the cleaning process and laser parameters, and provide feedback on the paint removal quality and effect. LIBS (Liquid Laser-Based Spatial Analysis) is a technique that uses pulsed lasers to form plasma on the sample surface. By collecting the elemental emission spectra formed during the plasma cooling process, it performs qualitative and quantitative analysis of the elements in the sample. It offers advantages such as short detection time, fast online analysis, good real-time monitoring, and the ability to automatically identify multiple elements. By monitoring the spectral information released by the material being cleaned in real time, online control of various laser parameters can be achieved, thereby reducing damage to the substrate from laser processing and enabling controllable delamination in multi-coat cleaning scenarios.

[0003] To avoid heat accumulation on the treated surface reaching the substrate damage threshold, existing technologies rely on fixing thermocouples to the bottom surface of the aircraft skin to be removed, connecting them to a temperature sensor and controller to collect temperature changes, and immediately stopping laser removal when the temperature shows a sudden increase; or forming an inert gas protective layer on the stealth coating surface area.

[0004] For example, existing Chinese patent CN116748687A discloses a laser removal method for stealth coatings on aircraft surfaces that can identify substrate damage responses. The method includes the following steps: 1) Fixing a thermocouple to the bottom surface of the aircraft skin sample to be removed, and connecting it to a temperature acquisition device and a control terminal to record the temperature changes of the bottom surface in real time; 2) Placing the aircraft skin sample on a worktable with the stealth coating facing the laser; 3) Turning on the laser, adjusting the parameters, and removing the stealth coating; 4) Observing the temperature changes of the bottom surface of the sample in real time at the control terminal. When the temperature curve shows a steep increase, the laser removal is immediately stopped. This method utilizes the temperature response caused by the direct action of the laser on the substrate to link whether the substrate is damaged with the visualized temperature changes.

[0005] Chinese patent CN113634902A discloses a gas-assisted laser processing method for removing stealth coatings from aircraft surfaces, comprising the following steps: 1) Fixing an aircraft surface sample coated with a stealth coating onto a work platform, with the coated side facing the laser cleaning head; 2) A gas-assisted protection device, including a main device and a tail device, introduces inert gas through a protective gas inlet to form an inert gas protective layer on the surface of the stealth coating; 3) Removing the stealth coating using a pulsed laser; 4) Turning off the laser and the gas-assisted protection device, cleaning the residual waste on the sample surface, and removing the sample; 5) If the stealth coating is not completely removed, repeating steps 1) to 4). This method ensures the safety of the laser removal process through the gas-assisted protection device.

[0006] However, the aforementioned prior art has the following drawbacks:

[0007] (1) Using thermocouples to obtain temperature changes to determine whether the substrate is damaged has problems such as slow response speed and high risk of damaging the substrate;

[0008] (2) In the case of multi-coating laser cleaning, the cleaning process needs to be layered and controllable, which is difficult to achieve by the above methods. Summary of the Invention

[0009] The purpose of this invention is to provide a method, system, and medium for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics, in order to solve the aforementioned problems.

[0010] This invention is mainly achieved through the following technical solutions:

[0011] A method for dynamic laser delamination removal of multilayer heterostructure coatings based on spectral characteristics includes the following steps:

[0012] Step S1: Collect spectrometer data of different coatings during laser paint removal and perform data preprocessing;

[0013] Based on the coating information, find the spectral data of adjacent coatings, filter the datasets with the connection time of adjacent coatings according to the stored data timestamps, and divide the filtered datasets into cross-coating spectral data to form cross-coating datasets;

[0014] Step S2: Based on the spectrometer data of different preprocessed coatings, fit Gaussian fitting curves to determine the characteristic elements of different coatings; identify the coating information of the current process based on the Gaussian fitting curves, and proceed to step S4.

[0015] Step S3: Train a deep learning model based on the cross-coating dataset to obtain a cross-coating prediction model; use the cross-coating prediction model to predict whether the currently processed coating is about to cross the coating.

[0016] Step S4: Based on the coating information in step S2, configure the corresponding laser processing parameters for the number of coating layers. The laser processing parameters include laser power, pulse width, scanning speed, and working distance. If a warning message indicating that the coating is about to cross layers is predicted in step S3, the laser processing parameters for the next layer are dynamically configured based on the warning message in step S3 to achieve precise layer removal and avoid damage to the substrate or the lower coating layer.

[0017] Specifically, based on the coating information determined in S2, the spectral data of adjacent coatings, such as the data of the first and second layers, are found. According to the storage data timestamp, the datasets of the last period of the first layer and the first period of the second layer are filtered. The filtered data is divided into cross-coating spectral data, which is stored in the cross-coating dataset storage format. This data is used as the input of step S3 to train the cross-coating prediction model.

[0018] To better realize the present invention, further, in step S1, data preprocessing is performed: Gaussian filtering and baseline correction are performed on the spectrometer data of different coatings to eliminate background noise;

[0019] The formula for Gaussian filtering is:

[0020] ;

[0021] Where: δ is the standard deviation;

[0022] x represents the intensity;

[0023] l For detection wavelength;

[0024] I Smonth ( l () represents the wavelength after Gaussian filtering. l Corresponding strength;

[0025] The formula for baseline correction is:

[0026] ;

[0027] ;

[0028] in: For local residual baselines;

[0029] n is the total number of wavelength ranges;

[0030] I Corrected ( l () represents the wavelength after baseline correction. l Corresponding strength;

[0031] I( l i ) is the first i wavelength l The corresponding intensity.

[0032] To better realize the present invention, in step S1, the cross-coating sequence number is used as the label of the cross-coating spectral data of the spectrometer, and the cross-coating dataset is stored as a multi-dimensional matrix to store the intensity values ​​corresponding to multiple wavelengths in a continuous time period.

[0033] To better implement this invention, further, in step S2, firstly, the spectral data of different coating characteristic elements are averaged, and then a Gaussian fitting curve is obtained; with I( l The minimum intensity error corresponding to the feature element is used as the criterion to identify the coating information of the current process; where the Gaussian fitting formula is:

[0034] ;

[0035] Where: A is the peak height, approximately equal to the intensity corresponding to the wave crest.

[0036] For local remaining baselines,

[0037] σ is the half-width parameter.

[0038] l 0 center wavelength,

[0039] l The wavelength range to be fitted corresponds to the wave peak.

[0040] I( l ) represents the intensity corresponding to the calculated wavelength range.

[0041] To better realize the present invention, further, in step S3, a deep learning model is trained based on the cross-coating dataset to obtain a cross-coating prediction model; the deep learning model includes a long short-term memory network, an attention mechanism module, and a fully connected layer arranged sequentially from front to back; the attention mechanism module focuses attention on the intensity change data of adjacent coatings in the cross-coating dataset; the long short-term memory network predicts whether a cross-coating warning is imminent after processing the input spectrometer cross-coating time-series data through each LSTM unit.

[0042] To better realize the present invention, the attention mechanism module further uses spatiotemporal dual attention, simultaneously focusing on changes in the time dimension and the spectral intensity dimension.

[0043] To better realize the present invention, further, training a deep learning model based on a cross-coating dataset includes the following steps:

[0044] Step A1: First, input the [number of samples, time, number of features] data from the cross-coating dataset, and extract the temporal features using a Long Short-Term Memory network.

[0045] Step A2: Then, it enters the attention mechanism module: it goes through time attention weighting to focus on key time steps; then through feature attention weighting to focus on key intensity features; finally, it goes through a global average pooling layer to compress the time dimension.

[0046] Step A3: Finally, after passing through the fully connected layer, the predicted value is obtained, which is also a warning message about whether the layer is about to be crossed.

[0047] This invention is mainly achieved through the following technical solutions:

[0048] A dynamic laser delamination removal system for multilayer heterogeneous coatings based on spectral characteristics is implemented based on the aforementioned dynamic laser delamination removal method for multilayer heterogeneous coatings based on spectral characteristics. The system includes a data acquisition and preprocessing module, a coating identification module, a cross-coating prediction module, and a laser parameter configuration module.

[0049] The data acquisition and preprocessing module is used to acquire spectrometer data of different coatings in laser paint removal and perform preprocessing, as well as to divide the spectrometer cross-coating spectral data.

[0050] The coating identification module is used to fit Gaussian fitting curves to determine the characteristic elements of different coatings based on the preprocessed spectrometer data of different coatings; and to obtain coating information based on the analysis of the Gaussian fitting curves.

[0051] The cross-coating prediction module is used to train a deep learning model based on the cross-coating dataset to obtain a cross-coating prediction model; using the cross-coating prediction model, a warning information is given to predict whether the currently processed coating is about to cross the coating.

[0052] The laser parameter configuration module is used to dynamically adjust laser processing parameters based on coating information and cross-coating early warning information.

[0053] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics.

[0054] The beneficial effects of this invention are as follows:

[0055] To address the challenges of low coating recognition accuracy and lag in parameter adjustment affecting control in existing technologies, this invention achieves efficient real-time adaptive control of laser process parameters through spectral data acquisition, coating recognition, and cross-coating prediction. Specifically, this invention utilizes a spectrometer to acquire characteristic emission lines from atoms and ions released during laser coating processing to determine the location of the coating to be removed. This allows for control of laser intensity to avoid damaging the substrate, solving the problem of interface perception across multi-layer heterogeneous coatings during laser removal and further enabling precise, layered, online control of laser removal. Secondly, this invention combines LSTM and attention mechanisms to design a cross-coating prediction model that predicts in advance whether the laser will affect the next coating layer, thereby updating control parameters beforehand, shortening system response time, and improving laser removal efficiency, demonstrating good practicality. Attached Figure Description

[0056] Figure 1 This is a block diagram illustrating the principle of the laser-based dynamic removal system for multilayer heterogeneous coatings based on spectral characteristics, as described in this invention.

[0057] Figure 2 A schematic diagram of the parameters in the Gaussian fitting formula;

[0058] Figure 3 A schematic diagram of the storage structure for cross-coating datasets;

[0059] Figure 4 This is a schematic diagram of the cross-coating prediction model;

[0060] Figure 5 This is a flowchart of the laser-based dynamic removal method for multilayer heterogeneous coatings based on spectral characteristics, according to the present invention. Detailed Implementation

[0061] Example 1:

[0062] A method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics, such as... Figure 5 As shown, it includes the following steps:

[0063] T1. Data acquisition and preprocessing: Determine the wavelengths of the characteristic elements detected in different coatings, perform data preprocessing, and then divide the spectrometer cross-coating spectral data to form a cross-coating dataset.

[0064] T1.1 Determine the wavelengths of the characteristic elements to be detected. Identify the characteristic elements of different coatings and their corresponding detection wavelengths. For example, if there are 5 coating layers to be removed, each layer's characteristic elements are C, N, Al, Si, and H, corresponding to wavelengths of 193.09 nm, 174.50 nm, 308.22 nm, 251.61 nm, and 410.17 nm, respectively. Simultaneously confirm the parameter configuration for each laser layer. This configuration will be automatically applied to the laser during the processing of the corresponding coating.

[0065] T1.2 Data Preprocessing: Background noise was eliminated from the spectral data of different coatings obtained during laser paint removal by Gaussian filtering and baseline correction.

[0066] The formula for Gaussian filtering is:

[0067] ;

[0068] Where: δ is the standard deviation, with a value of 1;

[0069] x represents the intensity;

[0070] l For detection wavelength;

[0071] I Smonth ( l () represents the wavelength after Gaussian filtering. l Corresponding strength;

[0072] The formula for baseline correction is:

[0073] ;

[0074] ;

[0075] in: For local residual baselines;

[0076] n is the total number of wavelength ranges;

[0077] I Corrected ( l () represents the wavelength after baseline correction. l Corresponding strength;

[0078] I( l i ) is the first i wavelength l The corresponding intensity.

[0079] T1.3. The cross-coating dataset is obtained by partitioning the data; the cross-coating spectral data from the spectrometer is acquired, and the cross-coating sequence number is used as the label for that cross-coating spectral data. For example... Figure 3 As shown, the cross-coating dataset is stored as a multi-dimensional matrix, storing the intensity values ​​corresponding to multiple wavelengths over a continuous time period.

[0080] T2, Coating Identification Module: By fitting Gaussian curves to obtain spectral data of different coating characteristic elements, the module identifies the number of coating layers currently being processed.

[0081] T2.1 Determine the Gaussian fitting curves for different coating characteristic elements. By averaging the spectral data of different coating characteristic elements, calculate the Gaussian fitting curve of the averaged spectral data for each characteristic element. This curve serves as the basis for identifying the processed coating.

[0082] T2.2, such as Figure 2 As shown, the Gaussian fitting formula is:

[0083] ;

[0084] Where: A is the peak height, approximately equal to the intensity corresponding to the wave crest.

[0085] For local remaining baselines,

[0086] σ is the half-width parameter.

[0087] l 0 center wavelength,

[0088] l The wavelength range to be fitted corresponds to the wave peak.

[0089] I( l ) represents the intensity corresponding to the calculated wavelength range. The target is to make I( l The intensity error corresponding to the characteristic element is the smallest.

[0090] T3, the cross-coating prediction module, continuously acquires spectral information from the spectrometer and stores it as such. Figure 3 The input data shown is the intensity corresponding to the wavelength over a period of time. Inputting this data into the cross-coating prediction module will determine whether the laser is about to act on the next layer. If a cross-coating warning is predicted, the module will enter the parameter configuration module and dynamically configure the laser processing parameters for the next layer. Otherwise, the identification will continue.

[0091] T3.1, Coating layer recognition model, such as Figure 4 As shown, the model combines LSTM and AM. AM is used to make the model focus on the intensity change data of adjacent coatings in the cross-coating dataset, thereby improving the model's ability to capture key change points; LSTM is used to calculate whether the input cross-coating time series data is about to cross the coating after each LSTM unit.

[0092] T3.1.1 uses spatiotemporal dual attention in the attention mechanism, simultaneously focusing on the time dimension and the spectral intensity dimension.

[0093] T3.2 Network Structure: The original input data [number of samples, time, number of features] is processed through an LSTM layer to extract temporal features, then through temporal attention weighting to focus on key time steps, then through feature attention weighting to focus on key intensity features, then through a global average pooling layer to compress the time dimension, and finally through a fully connected output layer to obtain the predicted value.

[0094] T3.3 LSTM model adjustment parameters include, but are not limited to, the number of LSTM units, batch size, number of iterations, learning rate, optimizer, etc.

[0095] T3.3.1, Number of LSTM Units: The number of units determines the complexity and learning performance of the model. Too few units may fail to capture key changes, while too many units may capture too much information and lead to overfitting. The optimal number of units is selected based on cross-validation. The number of LSTM units is set to 32 or 64.

[0096] T3.3.2 Batch Size: Batch size refers to the number of subsets into which the dataset is divided. Larger batch sizes may lead to excessive computation and impact the server's parallel computing capabilities, while smaller batch sizes can better reduce fluctuations in gradient descent but may increase noise. Batch sizes can be set to 32, 64, 128, etc.

[0097] T3.3.3, Number of iterations: The number of iterations is called epoch. A larger number of this parameter may lead to overfitting of the model, while a smaller number of the parameter may lead to underfitting. Set the number of iterations to 100-300 times.

[0098] T3.3.4 Learning Rate: The learning rate is a hyperparameter that represents the speed at which the model weights and biases are updated each time. An excessively large learning rate may prevent the model from finding the global maximum, while an excessively small learning rate may lead to slow model convergence and the finding of local maxima. The initial learning rate is set in the range of 0.0001-0.01, and then dynamically adjusted during training, such as according to the size of the epoch and the change in loss.

[0099] Preferably, in this example, the hyperparameters are set as follows: learning rate lr = 0.001, batch = 30, number of iterations epoch = 150, and number of LSTM units = 32.

[0100] T4, parameter configuration module, configures laser processing power, pulse width, scanning speed and working distance, etc.

[0101] Based on the coating identification module in step T2, the number of processing layers is determined by confirming the current coating feature elements. Then, laser processing parameters such as laser power, pulse width, scanning speed, and working distance are configured. At the same time, based on the prediction results of the cross-coating prediction module in step T3, the optimal laser processing parameters are calculated and set in real time to achieve precise layer removal and avoid damage to the substrate or the underlying coating.

[0102] For example, when the coating identification module in step T2 confirms that the laser is acting on the i-th layer, it configures the relevant parameters of the i-th layer to the laser. During the processing, the cross-coating prediction module in T3 detects in real time whether the laser will act on the next layer. If so, it configures the parameter information of the (i+1)-th layer to the laser in a timely manner to avoid the laser damaging the next layer coating. The precise layered online control of laser processing is achieved through the modules in T2 and T3.

[0103] Example 2:

[0104] A dynamic laser-based delamination removal system for multilayer heterogeneous coatings based on spectral characteristics, such as... Figure 1 As shown, the implementation based on the above method includes a data acquisition and preprocessing module, a coating identification module, a cross-coating prediction module, and a laser parameter configuration module.

[0105] The data acquisition and preprocessing module is used to acquire spectrometer data of different coatings in laser paint removal and perform preprocessing, as well as to divide the spectrometer cross-coating spectral data.

[0106] The coating identification module is used to fit Gaussian fitting curves to determine the characteristic elements of different coatings based on the preprocessed spectrometer data of different coatings; and to obtain coating information based on the analysis of the Gaussian fitting curves.

[0107] The cross-coating prediction module is used to train a deep learning model based on the cross-coating dataset to obtain a cross-coating prediction model; using the cross-coating prediction model, a warning information is given to predict whether the currently processed coating is about to cross the coating.

[0108] The laser parameter configuration module is used to dynamically adjust laser processing parameters based on coating information and cross-coating early warning information.

[0109] like Figure 4 As shown, the deep learning model includes a Long Short-Term Memory network, an attention mechanism module, and a fully connected layer arranged sequentially from front to back. The attention mechanism module uses spatiotemporal dual attention, simultaneously focusing on changes in both the time dimension and the spectral intensity dimension. Specifically, in the attention mechanism module, temporal attention weighting focuses on key time steps; then feature attention weighting focuses on key intensity features; finally, a global average pooling layer compresses the time dimension.

[0110] Training a deep learning model based on a cross-coating dataset includes the following steps:

[0111] Step A1: First, input the [number of samples, time, number of features] data from the cross-coating dataset, and extract the temporal features using a Long Short-Term Memory network.

[0112] Step A2: Then, it enters the attention mechanism module: it goes through time attention weighting to focus on key time steps; then through feature attention weighting to focus on key intensity features; finally, it goes through a global average pooling layer to compress the time dimension.

[0113] Step A3: Finally, after passing through the fully connected layer, the predicted value is obtained, which is also a warning message about whether the layer is about to be crossed.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics, characterized in that, Includes the following steps: Step S1: Collect spectrometer data of different coatings during laser paint removal and perform data preprocessing; Step S2: Based on the spectrometer data of different coatings after preprocessing, fit Gaussian fitting curves to determine the characteristic elements of different coatings; identify the coating information of the current process based on the Gaussian fitting curves; Based on the coating information, find the spectral data of adjacent coatings, filter the datasets with the connection time of adjacent coatings according to the stored data timestamps, and divide the filtered datasets into cross-coating spectral data to form cross-coating datasets; Step S3: Train a deep learning model based on the cross-coating dataset to obtain a cross-coating prediction model; use the cross-coating prediction model to predict whether the currently processed coating is about to cross the coating. Step S4: Based on the coating information in step S2, configure the laser processing parameters for the corresponding number of coating layers. The laser processing parameters include laser power, pulse width, scanning speed, and working distance. If a warning message indicating that the coating is about to cross a layer is obtained in step S3, the laser processing parameters for the next layer are dynamically configured based on the warning message in step S3.

2. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 1, characterized in that, In step S1, data preprocessing involves Gaussian filtering and baseline correction of the spectrometer data for different coatings. The formula for Gaussian filtering is: ； Where: δ is the standard deviation; x represents the intensity; λ For detection wavelength; I Smonth ( λ () represents the wavelength after Gaussian filtering. λ Corresponding strength; The formula for baseline correction is: ； ； in: For local residual baselines; n is the total number of wavelength ranges; I Corrected ( λ () represents the wavelength after baseline correction. λ Corresponding strength; I( λ i ) is the first i wavelength λ The corresponding intensity.

3. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 1 or 2, characterized in that, In step S1, the cross-coating sequence number is used as the label for the cross-coating spectral data of the spectrometer, and the cross-coating dataset is stored as a multi-dimensional matrix.

4. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 2, characterized in that, In step S2, firstly, the spectral data of different coating characteristic elements are averaged, and then a Gaussian fitting curve is obtained; with I( λ The minimum intensity error corresponding to the feature element is used as the criterion to identify the coating information of the current process; where the Gaussian fitting formula is: ； Where: A is the peak height, approximately equal to the intensity corresponding to the wave crest. For local remaining baselines, σ is the half-width parameter. λ 0 center wavelength, λ The wavelength range to be fitted corresponds to the wave peak. I( λ ) represents the intensity corresponding to the calculated wavelength range.

5. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 1, characterized in that, In step S3, the deep learning model includes a long short-term memory network, an attention mechanism module, and a fully connected layer arranged sequentially from front to back. The attention mechanism module focuses its attention on the intensity change data of adjacent coatings in the cross-coating dataset. The long short-term memory network calculates the input spectrometer cross-coating time-series data through each LSTM unit and predicts whether a cross-coating is imminent.

6. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 5, characterized in that, The attention mechanism module uses spatiotemporal dual attention, simultaneously focusing on changes in both the time dimension and the spectral intensity dimension.

7. The method for dynamic laser delamination removal of multilayer heterogeneous coatings based on spectral characteristics according to claim 6, characterized in that, Training a deep learning model based on a cross-coating dataset includes the following steps: Step A1: First, input the [number of samples, time, number of features] data from the cross-coating dataset, and extract the temporal features using a Long Short-Term Memory network. Step A2: Then, it enters the attention mechanism module: it goes through time attention weighting to focus on key time steps; then through feature attention weighting to focus on key intensity features; finally, it goes through a global average pooling layer to compress the time dimension. Step A3: Finally, after passing through the fully connected layer, the predicted value is obtained, which is also a warning information about whether the layer is about to be crossed.

8. A laser-based dynamic removal system for multilayer heterogeneous coatings based on spectral characteristics, implemented based on the laser-based dynamic removal method for multilayer heterogeneous coatings based on spectral characteristics according to any one of claims 1-7, characterized in that, It includes a data acquisition and preprocessing module, a coating identification module, a cross-coating prediction module, and a laser parameter configuration module; The data acquisition and preprocessing module is used to acquire spectrometer data of different coatings in laser paint removal and perform preprocessing, as well as to divide the spectrometer cross-coating spectral data. The coating identification module is used to fit Gaussian fitting curves to determine the characteristic elements of different coatings based on the preprocessed spectrometer data of different coatings. Coating information was obtained based on Gaussian fitting curve analysis; The cross-coating prediction module is used to train a deep learning model based on the cross-coating dataset to obtain a cross-coating prediction model; using the cross-coating prediction model, a warning information is given to predict whether the currently processed coating is about to cross the coating. The laser parameter configuration module is used to dynamically adjust laser processing parameters based on coating information and cross-coating early warning information.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the laser-based dynamic removal method for multilayer heterogeneous coatings based on spectral characteristics, as described in any one of claims 1-7.

Images