Method and system for detecting coating quality of coating machine

Abstract

The invention discloses a method and system for detecting the coating quality of a coating machine, and relates to the technical field of industrial automation and intelligent detection. Spectral reflectivity data and film layer thickness oscillation frequency data are collected and preprocessed, coating quality characteristics are extracted, a quality classification model and a thickness prediction model are constructed, and the coating quality is detected. And outputting a coating quality classification result, a quality confidence coefficient, a thickness prediction value and a prediction error interval, performing classification judgment and thickness judgment, formulating a corresponding compensation mechanism, obtaining a real-time risk index, synchronously performing graded early warning and time-space correlation early warning, and designing a three-dimensional visual board to integrate key indexes. Through collaborative analysis of multi-modal data fusion and an intelligent algorithm, online real-time detection and dynamic regulation and control of the coating quality are realized, the defect recognition sensitivity and the process abnormality response efficiency are remarkably improved while high-precision optical performance and thickness uniformity are ensured, and the method is suitable for large-scale popularization and application. And the production yield is effectively optimized, and the stability and controllability of the coating process are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation and intelligent detection technology, and in particular to a method and system for detecting the coating quality of a coating machine. Background Art

[0002] A coating machine is a precision device that deposits thin films on the surface of materials through physical and chemical methods under a vacuum environment. Its core function is to improve the performance of the substrate and impart new functions. According to the technical principles, it can be divided into two categories: evaporation coating and sputtering coating. Evaporation coating vaporizes the target material by heating it, and atoms and molecules condense into a film on the surface of the substrate. Sputtering coating uses high-energy particles to bombard the target material, and the sputtered atoms are deposited to form a thin film. This equipment is widely used in optics, electronics, automobiles and medical fields. For example, anti-reflective films are coated on optical lenses to increase light transmittance, conductive and insulating layers are deposited for semiconductor devices, and corrosion-resistant coatings are coated for automotive parts. Its technical advantages include the ability to achieve nanoscale film thickness control, support the deposition of multiple materials including metals, non-metals and compounds, and obtain high hardness, low friction and anti-oxidation properties through process optimization. It is a key technical means to achieve material surface functionalization in modern industry.

[0003] To address the issues of a single dimension for evaluating coating quality, delayed real-time dynamic control, and blind spots in identifying hidden defects during the coating process, existing technologies employ offline spot checks combined with single-point threshold determination. However, this approach can result in long delays in responding to process anomalies and an inability to capture the dynamic coupling relationship between film micro-parameters. This can lead to the continuous accumulation of hidden defects, loss of control over coating uniformity, and large fluctuations in batch yield. To address these issues, a method and system for detecting coating quality in a coating machine are proposed. Summary of the Invention

[0004] The object of the present invention is to provide a method and system for detecting the coating quality of a coating machine, so as to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: In the first aspect, a method for detecting the coating quality of a coating machine comprises the following steps:

[0006] S1. Collect and pre-process the spectral reflectance data and coating layer thickness data of the coating surface;

[0007] S2, extracting coating quality features from the pre-processed spectral reflectance data and coating thickness data;

[0008] S3. Based on the coating quality characteristics, a quality classification model and a thickness prediction model are constructed, and the coating quality classification results, quality confidence, thickness prediction value and prediction error range are output respectively. Then, classification judgment and thickness judgment are made, and corresponding compensation mechanisms are formulated;

[0009] S4. Based on the coating quality classification results and thickness prediction values, obtain the real-time risk index, and simultaneously perform graded warnings and spatiotemporal correlation warnings.

[0010] S5. Combine quality confidence, thickness prediction value and real-time risk index to design a three-dimensional visual dashboard.

[0011] A further improvement of the technical solution of the present invention is that in said S1, the collection and preprocessing process of the spectral reflectance data of the coating surface and the coating layer thickness data includes:

[0012] A fiber optic spectrum sensor probe is deployed in the vacuum chamber of the coating machine. A halogen light source emits a continuous spectrum of 400-700nm, which is transmitted to the coating surface via an optical fiber. The reflected light is introduced into the spectrometer's spectroscopic system through a collection optical fiber. After being dispersed by a grating, it is received by a linear array CCD detector. The percentage difference between the reflected light intensity on the coating surface and the reflected light intensity of a reference standard white plate and the dark current noise is calculated as the spectral reflectance data of the coating surface.

[0013] A quartz crystal sensor is embedded in the coating substrate fixture. The surface of the quartz crystal sensor is coplanar with the coating surface of the substrate. The oscillation circuit is connected to a frequency counter via a coaxial cable to obtain the change in the quartz crystal resonant frequency and the coating layer thickness data.

[0014] The included angle between the axis of the optical fiber spectrum sensor probe and the normal of the coated substrate surface shall not exceed 5 degrees, the distance between the optical fiber spectrum sensor probe and the coated substrate shall be maintained at 50±2mm, the optical fiber spectrum sensor shall be connected to the external spectrum analyzer via an optical fiber jumper, a collimating lens group shall be integrated in the optical path to ensure that the incident beam diameter does not exceed 3mm, and the installation flatness error between the quartz crystal sensor surface and the substrate coating surface shall not exceed 0.01mm;

[0015] The wavelet threshold denoising algorithm is used to perform multi-scale decomposition and baseline correction on the continuous spectrum emitted by the halogen light source. The coating thickness data is filtered by sliding window weighted average, and the acquisition delay of the spectral reflectance data and the coating thickness data is compensated by the time series alignment model.

[0016] A further improvement of the technical solution of the present invention is that in S2, the process of extracting the coating quality characteristics from the pre-processed spectral reflectance data and coating thickness data includes:

[0017] Coating quality characteristics include dominant wavelength characteristics, color uniformity, deposition rate and thickness fluctuation coefficient;

[0018] The dominant wavelength characteristic λ d and color uniformity U C From the pre-processed spectral reflectance data R(λ i ), the calculation process is as follows:

[0019]

[0020] Among them, λ i is the wavelength value of the spectrum sampling point in the continuous spectrum range of 400-700nm, R max With R min The maximum and minimum values ​​extracted from the spectral reflectance data;

[0021] The deposition rate and thickness fluctuation coefficient F h It is calculated from the pre-processed coating thickness time series data h(t), and the calculation process is as follows:

[0022]

[0023] Where Δt is the sampling time interval, h i is a single thickness sampling value, is the mean thickness of the current batch, and N is the total number of samples.

[0024] A further improvement of the technical solution of the present invention is that in S3, the process of building a quality classification model based on the coating quality characteristics and outputting the coating quality classification results and quality confidence includes:

[0025] Taking the two-dimensional vector composed of the dominant wavelength feature and chromatic uniformity as input, a quality classification model framework is constructed. The input feature space is divided into qualified and unqualified areas through nonlinear mapping. At the same time, a confidence quantification mechanism is established.

[0026] The particle swarm algorithm is used to iteratively optimize the kernel function parameters and penalty coefficients of the support vector machine. Each particle position represents a set of candidate parameters. The classification accuracy of the validation set is used as the fitness evaluation indicator. After position updates and global optimal tracking, the optimal parameter configuration is obtained, and the nonlinear relationship between input features and quality categories is modeled.

[0027] The signed distance between the sample output by the optimized support vector machine decision function and the classification hyperplane is input into the Sigmoid probability function and converted into a quality confidence in the interval [0,1]. Positive distances correspond to qualified classes and the quality confidence is greater than 0.5. The closer to 1, the higher the confidence. Negative distances correspond to unqualified classes and the confidence is less than 0.5. The closer to 0, the lower the confidence.

[0028] A further improvement of the technical solution of the present invention is that in said S3, the process of building a thickness prediction model in combination with the coating quality characteristics and outputting the thickness prediction value and the prediction error range includes:

[0029] Input a two-dimensional matrix consisting of deposition rate and thickness fluctuation coefficient, extract orthogonal principal components through covariance matrix decomposition, and output the eigenvalues ​​and eigenvectors after dimensionality reduction;

[0030] The eigenvalues ​​and eigenvectors are input into the input layer of the back-propagation neural network architecture. The hidden layer nodes capture the complex relationship between the input features through nonlinear activation functions. The output layer generates thickness prediction values ​​through linear combination and uses a hybrid loss function to balance the absolute error and relative error. This establishes a thickness prediction model that nonlinearly maps the input deposition rate, thickness fluctuation coefficient, and coating layer thickness.

[0031] Based on the residual distribution of historical thickness prediction values, the standard deviation σ of the statistical thickness prediction value h , combined with the confidence coefficient to determine the symmetric prediction error interval, and output the thickness prediction value and the prediction error interval

[0032] 80% of the data input to the thickness prediction model is divided into a training set and 20% of the data is divided into a validation set. The training set is used to iteratively update the back-propagation neural network weights, and the validation set is used to monitor the generalization performance of the thickness prediction model. When the validation error does not improve for five consecutive iterations, the early stopping mechanism is triggered.

[0033] A further improvement of the technical solution of the present invention is that in S3, the process of performing classification determination and thickness determination and formulating a corresponding compensation mechanism includes:

[0034] The classification judgment includes: when the quality classification result is qualified and the confidence level C q ≥0.9, passed the optical performance test, if there is quality classification as unqualified and confidence level C q <0.9, it is judged as an optical defect;

[0035] The thickness determination includes that the thickness prediction value must meet Among them, μ h is the mean value of the thickness prediction of historical batches, σ h is the standard deviation of the thickness prediction value, and the prediction error interval must be within the process standard range [h min ,h max ], the thickness test is passed; if the thickness prediction value and the prediction error interval do not meet the conditions, the thickness test is failed;

[0036] When the thickness prediction value deviates from the target value h targetWhen , calculate the evaporation source power adjustment ΔP, the calculation process is as follows:

[0037]

[0038] Among them, K p With K i is the compensation coefficient, t is the length of the integration time window;

[0039] If the optical judgment is qualified but the thickness judgment fails, the evaporation source temperature is adjusted Prioritize adjusting the deposition rate. If the thickness passes but the quality confidence level is in the range [0.7, 0.9), trigger the plasma excitation frequency compensation.

[0040] A further improvement of the technical solution of the present invention is that in said S4, based on the coating quality classification result and the thickness prediction value, a real-time risk index is obtained, and the process of simultaneously performing graded warning and spatiotemporal correlation warning includes:

[0041] Define the real-time risk index RI as the quality confidence attenuation term (1-C q ) and thickness offset The weighted sum of and the extreme value normalization of the thickness offset term;

[0042] When RI ≥ 0.8 and When RI<0.8, it is judged as a first-level red warning, the coating machine will be shut down immediately and the vacuum chamber pressure relief protection will be activated. When it is judged as a Level 2 orange warning, the coating machine will slow down and activate the auxiliary evaporation source. When 0.4≤RI<0.6 and C q When <0.7, it is judged as a level 3 yellow warning, the coating machine keeps running but records the parameter deviation trend;

[0043] Detect the moving average slope of the thickness prediction value. When the absolute value of the moving average slope is greater than 0.5σ h / min for 3 minutes, the deposition rate instability warning is triggered;

[0044] If there are C q If the standard deviation exceeds 0.15, an early warning of film uniformity degradation is triggered;

[0045] Obtain the correlation coefficient between the quality confidence and the thickness prediction value. When the correlation coefficient is lower than -0.5 and persists for 5 sampling cycles, it is determined that the decoupling of optical and mechanical properties is abnormal.

[0046] After each batch of production, the real-time risk index RI is updated based on historical data thres =μ RI +2σRI wherein, μ RI and σ RI are the mean and standard deviation of the real-time risk index of this batch;

[0047] A delay confirmation mechanism is introduced, and when the early warning condition is triggered for 3 times continuously and the interval is less than or equal to 10 seconds, a response action is performed.

[0048] The further improvement of the technical scheme of the present application is that, in S5, the process of designing a three-dimensional visual board in combination with the quality confidence, the thickness prediction value and the real-time risk index comprises:

[0049] The standard deviation multiple of the thickness prediction value relative to the historical mean value is taken as the standardized thickness deviation degree, X axis is defined as the standardized thickness deviation degree, Y axis is defined as the quality confidence, and Z axis is defined as the normalized real-time risk index, the parameter dimension is unified through the 4σ principle and linear mapping, and a three-dimensional data space base is constructed;

[0050] Real-time data points are colored according to early warning levels, and the quality determination state is distinguished by spheres, cubes and cones, historical data trajectories are displayed in a semi-transparent band, the band width is dynamically adjusted according to the thickness fluctuation coefficient, and the process stability change trend is reflected;

[0051] The green qualified area is demarcated by an ellipsoid equation, the half-axis length is set based on the film thickness tolerance, the quality confidence threshold and the upper limit of the real-time risk index, the process parameter safety range is represented by the space boundary, and the visual alarm is triggered when the boundary is penetrated;

[0052] The data point coordinates are refreshed every 100 milliseconds, the 5-minute sliding window historical trajectory is retained, and when the real-time data point breaks through the qualified area and the associated early warning rule, the boundary flickering and space labeling are activated;

[0053] Three-dimensional space free rotation and 0.5-5 times scaling are supported, and data is displayed according to time and parameter interval, providing a multi-dimensional process analysis perspective.

[0054] In a second aspect, a detection system for film coating quality of a film coating machine is used to realize a detection method for film coating quality of a film coating machine, comprising a film coating data acquisition module, a film coating feature extraction module, a film coating quality detection module and a double-parameter coupling early warning module, wherein the modules are connected by electrical signals.

[0055] The film coating data acquisition module acquires and pre-processes the spectral reflectance data and the film thickness oscillation frequency data of the film coating surface;

[0056] The film coating feature extraction module extracts the film coating quality features from the pre-processed spectral reflectance data and the film thickness oscillation frequency data;

[0057] The coating quality detection module combines coating quality characteristics to construct a quality classification model and a thickness prediction model, respectively outputs coating quality classification results, quality confidence, thickness prediction values and prediction error intervals, and then performs classification determination and thickness determination, and formulates a corresponding compensation mechanism.

[0058] The double-parameter coupling early warning module obtains a real-time risk index based on the coating quality classification results and the thickness prediction values, and simultaneously performs hierarchical early warning and spatio-temporal correlation early warning.

[0059] The quality detection visualization module combines the quality confidence, the thickness prediction values and the real-time risk index to design a three-dimensional visualization board.

[0060] Due to the adoption of the above technical solutions, the present application has the following technical progress compared with the prior art:

[0061] 1. The present application provides a coating quality detection method and system for a coating machine, which breaks through the limitations of traditional single parameter detection through multi-modal data collaborative analysis of spectral reflectance and film thickness oscillation frequency, realizes comprehensive quality evaluation of coating optical performance and mechanical characteristics, and significantly improves defect recognition sensitivity and process abnormality tracing ability.

[0062] 2. The present application provides a coating quality detection method and system for a coating machine, which integrates quality classification confidence and thickness prediction error interval for fusion determination based on the linkage design of dynamic compensation mechanism and hierarchical early warning, forms a real-time closed-loop regulation system, effectively solves the problem of hidden defect response lag, and guarantees the stability and controllability of the coating process.

[0063] 3. The present application provides a coating quality detection method and system for a coating machine, which integrates the spatio-temporal evolution relationship of quality confidence, thickness prediction values and risk index through a three-dimensional visualization board, establishes an intelligent monitoring interface for human-machine collaboration, realizes global perception of coating quality multidimensional indicators and process parameter optimization decision support. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0065] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0067] Example 1, as Figure 1 As shown, the present invention provides a method for detecting the coating quality of a coating machine, comprising the following steps:

[0068] S1. Collect and pre-process the spectral reflectance data and coating thickness data of the coating surface. Deploy the fiber optic spectrum sensor probe in the vacuum chamber of the coating machine. The halogen light source emits a continuous spectrum of 400-700nm, which is transmitted to the coating surface through the optical fiber. The reflected light is introduced into the spectrometer spectrometer system through the collection optical fiber, and is received by the linear array CCD detector after grating dispersion. The percentage difference between the reflected light intensity of the coating surface and the reflected light intensity of the reference standard white board and the dark current noise is counted as the spectral reflectance data of the coating surface. A quartz crystal sensor is embedded in the coating substrate fixture. The surface of the quartz crystal sensor is coplanar with the coating surface of the substrate. The oscillation circuit is connected to the frequency counter through a coaxial cable to obtain the resonant frequency of the quartz crystal. The optical fiber spectrum sensor is connected to an external spectrum analyzer through an optical fiber jumper. A collimating lens group is integrated in the optical path to ensure that the diameter of the incident beam does not exceed 3 mm. The installation flatness error between the quartz crystal sensor surface and the substrate coating surface does not exceed 0.01 mm. A wavelet threshold noise reduction algorithm is used to perform multi-scale decomposition and baseline correction on the continuous spectrum emitted by the halogen light source. A sliding window weighted average filter is used for the coating thickness data. The collection delay of the spectral reflectance data and the coating thickness data is compensated by a time series alignment model.

[0069] S2. Extracting coating quality features from the pre-processed spectral reflectance data and coating thickness data. The coating quality features include dominant wavelength features, chromaticity uniformity, deposition rate, and thickness fluctuation coefficient. The dominant wavelength feature λ d and color uniformity U C From the pre-processed spectral reflectance data R(λ i ), the calculation process is as follows:

[0070]

[0071] Among them, λ iis the wavelength value of the spectrum sampling point in the continuous spectrum range of 400-700nm, R max With R min The maximum and minimum values ​​extracted from the spectral reflectance data are traversed, and the deposition rate and thickness fluctuation coefficient F h It is calculated from the pre-processed coating thickness time series data h(t), and the calculation process is as follows:

[0072]

[0073] Where Δt is the sampling time interval, h i is a single thickness sampling value, is the mean thickness of the current batch, and N is the total number of samples;

[0074] S3. Combined with the coating quality characteristics, a quality classification model and a thickness prediction model are constructed to output the coating quality classification results, quality confidence, thickness prediction value and prediction error range respectively, and then make classification judgments and thickness judgments, and formulate corresponding compensation mechanisms. The two-dimensional vector composed of the main wavelength characteristics and chromatic uniformity is used as input to construct a quality classification model framework. The input feature space is divided into qualified areas and unqualified areas through nonlinear mapping. At the same time, a confidence quantification mechanism is established. The kernel function parameters and penalty coefficients of the support vector machine are iteratively optimized through the particle swarm algorithm. Each particle position represents a set of candidate parameters. The classification accuracy of the verification set is used as the fitness evaluation index. After position update and global optimal tracking, the optimal parameter configuration is obtained. The nonlinear relationship between the input features and the quality category is modeled, and the signed distance from the sample output of the optimized support vector machine decision function to the classification hyperplane is input to The Sigmoid probability function is converted into a quality confidence in the interval [0,1], where the positive distance corresponds to the qualified class and the quality confidence is >0.5. The closer it is to 1, the higher the confidence. The negative distance corresponds to the unqualified class and the confidence is <0.5. The closer it is to 0, the lower the confidence. The two-dimensional matrix composed of the input deposition rate and thickness fluctuation coefficient is extracted by covariance matrix decomposition, and the eigenvalues ​​and eigenvectors after dimensionality reduction are output. The eigenvalues ​​and eigenvectors are input into the input layer of the back-propagation neural network architecture. The hidden layer nodes capture the complex relationship between the input features through nonlinear activation functions. The output layer generates thickness prediction values ​​through linear combination, and uses a mixed loss function to balance the absolute error and relative error. A thickness prediction model is established that nonlinearly maps the input deposition rate, thickness fluctuation coefficient and coating layer thickness. Based on the residual distribution of historical thickness prediction values, the standard deviation σ of the thickness prediction value is statistically analyzed. h , combined with the confidence coefficient to determine the symmetric prediction error interval, and output the thickness prediction value and the prediction error interval 80% of the data of the input thickness prediction model is divided into a training set, and 20% of the data is divided into a validation set, the training set is used to iteratively update the weights of the back propagation neural network, and the validation set is used to monitor the generalization performance of the thickness prediction model. When the validation error does not improve for 5 consecutive iterations, trigger the early stopping mechanism, the classification judgment includes when the quality classification result is qualified and the confidence C q ≥0.9, pass the optical performance test, if the quality classification is unqualified and the confidence C q <0.9, determine that it is an optical defect, the thickness judgment includes that the thickness prediction value needs to meet Wherein, μ h is the mean of the historical batch thickness prediction value, σ h is the standard deviation of the thickness prediction value, and the prediction error interval needs to be contained in the process standard range [h min , h max ], then pass the thickness test, if the thickness prediction value and the prediction error interval do not meet the conditions, then fail the thickness test, when the thickness prediction value deviates from the target value h target , calculate the evaporation source power adjustment amount ΔP, the calculation process is as follows:

[0075]

[0076] Wherein, K p and K i are compensation coefficients, t is the length of the integral time window, if the optical judgment is qualified but the thickness judgment fails, adjust the evaporation source temperature preferentially adjust the deposition rate, if the thickness judgment passes but the quality confidence is in the interval [0.7, 0.9), trigger the plasma excitation frequency compensation

[0077] S4, based on the coating quality classification result and the thickness prediction value, obtain a real-time risk index, and simultaneously perform hierarchical early warning and spatio-temporal correlation early warning, define the real-time risk index RI as the weighted sum of the quality confidence decay term (1-C q ) and the thickness deviation term , when RI≥0.8 and , determine that it is a first-level red early warning, the coating machine immediately stops and starts vacuum chamber pressure relief protection, when 0.6≤RI<0.8 and , determine that it is a second-level orange early warning, the coating machine runs at a reduced speed and activates the auxiliary evaporation source, when 0.4≤RI<0.6 and C q <0.7, determine that it is a third-level yellow early warning, the coating machine keeps running but records the parameter deviation trend, detects the moving average slope of the thickness prediction value, when the absolute value of the moving average slope is greater than 0.5σ h / min for 3 minutes, the deposition rate instability warning is triggered. If there are C q If the standard deviation exceeds 0.15, it triggers the warning of film uniformity degradation and obtains the correlation coefficient between the quality confidence and the thickness prediction value. When the correlation coefficient is lower than -0.5 and lasts for 5 sampling cycles, it is judged as an abnormal decoupling of optical and mechanical properties. After each batch of production, the real-time risk index R is updated according to historical data. Ithres =μ RI +2σ RI , where μ RI and σ RI For the mean and standard deviation of the real-time risk index of this batch, a delayed confirmation mechanism is introduced. When the warning condition is triggered three times in a row with an interval of ≤10 seconds, the response action will be executed.

[0078] S5. Combining quality confidence, thickness prediction value and real-time risk index, a three-dimensional visualization dashboard is designed. The standard deviation of the thickness prediction value relative to the historical mean is used as the standardized thickness deviation. The X-axis is defined as the standardized thickness deviation, the Y-axis is the quality confidence, and the Z-axis is the normalized real-time risk index. The parameter dimensions are unified through the 4σ principle and linear mapping, and a three-dimensional data space base is constructed. The real-time data points are colored according to the warning level. The quality judgment status is distinguished by spheres, cubes and cones. The historical data trajectory is displayed in a semi-transparent band. The bandwidth is dynamically adjusted with the thickness fluctuation coefficient to reflect the process. The green qualified area is delineated by the ellipsoid equation, and the semi-axis length is set based on the film thickness tolerance, quality confidence threshold and real-time risk index upper limit. The spatial boundary represents the safe range of process parameters, and a visual alarm is triggered when the boundary is penetrated. The data point coordinates are refreshed every 100 milliseconds, and the historical trajectory of the 5-minute sliding window is retained. When the real-time data point breaks through the qualified area and the associated warning rules, the boundary flashing and spatial annotation are activated. It supports free rotation in three-dimensional space and 0.5-5 times zoom, as well as filtering and displaying data by time and parameter range, providing a multi-dimensional process analysis perspective.

[0079] Example 2, as Figure 1 As shown, based on Example 1, the present invention provides a technical solution: a coating quality detection system for a coating machine, which is used to implement a coating quality detection method for a coating machine, including a coating data acquisition module, a coating feature extraction module, a coating quality detection module and a dual-parameter coupling early warning module, wherein the electrical signals between the modules are connected.

[0080] The coating data acquisition module collects and pre-processes the spectral reflectance data and film thickness oscillation frequency data of the coating surface;

[0081] The coating feature extraction module extracts coating quality features from the pre-processed spectral reflectance data and film thickness oscillation frequency data;

[0082] The coating quality detection module, in combination with the coating quality characteristics, constructs a quality classification model and a thickness prediction model, and outputs the coating quality classification result, quality confidence, thickness prediction value and prediction error range respectively, and then performs classification judgment and thickness judgment, and formulates a corresponding compensation mechanism;

[0083] The dual-parameter coupling early warning module obtains a real-time risk index based on the coating quality classification results and thickness prediction values, and simultaneously performs graded early warning and spatiotemporal correlation early warning.

[0084] The quality inspection visualization module combines quality confidence, thickness prediction value and real-time risk index to design a three-dimensional visualization dashboard.

[0085] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0086] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0087] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0088] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0089] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0090] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0091] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting the coating quality of a coating machine, characterized in that: The following steps are involved: S1. Collect and pre-process the spectral reflectance data and coating layer thickness data of the coating surface; S2, extracting coating quality features from the pre-processed spectral reflectance data and coating thickness data; S3. Based on the coating quality characteristics, a quality classification model and a thickness prediction model are constructed, and the coating quality classification results, quality confidence, thickness prediction value and prediction error range are output respectively. Then, classification judgment and thickness judgment are made, and corresponding compensation mechanisms are formulated; S4. Based on the coating quality classification results and thickness prediction values, obtain the real-time risk index, and simultaneously perform graded warnings and spatiotemporal correlation warnings. S5. Combine quality confidence, thickness prediction value and real-time risk index to design a three-dimensional visual dashboard.

2. The method for detecting the coating quality of a coating machine according to claim 1, wherein: In S1, the collection and preprocessing of the spectral reflectance data and the coating layer thickness data of the coating surface includes: A fiber optic spectrum sensor probe is deployed in the vacuum chamber of the coating machine. A halogen light source emits a continuous spectrum of 400-700nm, which is transmitted to the coating surface via an optical fiber. The reflected light is introduced into the spectrometer's spectroscopic system through a collection optical fiber. After being dispersed by a grating, it is received by a linear array CCD detector. The percentage difference between the reflected light intensity on the coating surface and the reflected light intensity of a reference standard white plate and the dark current noise is calculated as the spectral reflectance data of the coating surface. A quartz crystal sensor is embedded in the coating substrate fixture. The surface of the quartz crystal sensor is coplanar with the coating surface of the substrate. The oscillation circuit is connected to a frequency counter via a coaxial cable to obtain the change in the quartz crystal resonant frequency and the coating layer thickness data. The wavelet threshold denoising algorithm is used to perform multi-scale decomposition and baseline correction on the continuous spectrum emitted by the halogen light source. The coating thickness data is filtered by sliding window weighted average, and the acquisition delay of the spectral reflectance data and the coating thickness data is compensated by the time series alignment model.

3. The method for detecting the coating quality of a coating machine according to claim 2, wherein: In S2, the process of extracting the coating quality characteristics includes: Coating quality characteristics include dominant wavelength characteristics, color uniformity, deposition rate and thickness fluctuation coefficient; The dominant wavelength characteristic λ d and color uniformity U C From the pre-processed spectral reflectance data R(λ i ), the calculation process is as follows: Among them, λ i is the wavelength value of the spectrum sampling point in the continuous spectrum range of 400-700nm, R max With R min The maximum and minimum values ​​extracted from the spectral reflectance data; The deposition rate and thickness fluctuation coefficient F h It is calculated from the pre-processed coating thickness time series data h(t), and the calculation process is as follows: Where Δt is the sampling time interval, h i is a single thickness sampling value, is the mean thickness of the current batch, and N is the total number of samples.

4. The method for detecting the coating quality of a coating machine according to claim 3, wherein: In S3, the process of constructing a quality classification model and outputting the coating quality classification result and quality confidence level includes: Taking the two-dimensional vector composed of the dominant wavelength feature and chromatic uniformity as input, a quality classification model framework is constructed. The input feature space is divided into qualified and unqualified areas through nonlinear mapping. At the same time, a confidence quantification mechanism is established. The particle swarm algorithm is used to iteratively optimize the kernel function parameters and penalty coefficients of the support vector machine. Each particle position represents a set of candidate parameters. The classification accuracy of the validation set is used as the fitness evaluation indicator. After position updates and global optimal tracking, the optimal parameter configuration is obtained, and the nonlinear relationship between input features and quality categories is modeled. The signed distance between the sample output by the optimized support vector machine decision function and the classification hyperplane is input into the Sigmoid probability function and converted into a quality confidence in the interval [0,1]. Positive distances correspond to qualified classes and the quality confidence is greater than 0.

5. The closer to 1, the higher the confidence. Negative distances correspond to unqualified classes and the confidence is less than 0.

5. The closer to 0, the lower the confidence.

5. The method for detecting the coating quality of a coating machine according to claim 4, characterized in that: In S3, the process of constructing a thickness prediction model and outputting a thickness prediction value and a prediction error range includes: Input a two-dimensional matrix consisting of deposition rate and thickness fluctuation coefficient, extract orthogonal principal components through covariance matrix decomposition, and output the eigenvalues ​​and eigenvectors after dimensionality reduction; The eigenvalues ​​and eigenvectors are input into the input layer of the back-propagation neural network architecture. The hidden layer nodes capture the complex relationship between the input features through nonlinear activation functions. The output layer generates thickness prediction values ​​through linear combination and uses a hybrid loss function to balance the absolute error and relative error. This establishes a thickness prediction model that nonlinearly maps the input deposition rate, thickness fluctuation coefficient, and coating layer thickness. Based on the residual distribution of historical thickness prediction values, the standard deviation σ of the statistical thickness prediction value h , combined with the confidence coefficient to determine the symmetric prediction error interval, and output the thickness prediction value and the prediction error interval 80% of the data input to the thickness prediction model is divided into a training set and 20% of the data is divided into a validation set. The training set is used to iteratively update the back-propagation neural network weights, and the validation set is used to monitor the generalization performance of the thickness prediction model. When the validation error does not improve for five consecutive iterations, the early stopping mechanism is triggered.

6. The method for detecting the coating quality of a coating machine according to claim 5, characterized in that: In S3, the process of performing classification and thickness determination and formulating a corresponding compensation mechanism includes: The classification judgment includes: when the quality classification result is qualified and the confidence level C q ≥0.9, passed the optical performance test, if there is quality classification as unqualified and confidence level C q <0.9, it is judged as an optical defect; The thickness determination includes that the thickness prediction value must meet Among them, μ h is the average value of the thickness prediction value of the historical batch, and the prediction error interval must be included in the process standard range [h min ,[h max ], the thickness test is passed; if the thickness prediction value and the prediction error interval do not meet the conditions, the thickness test is failed; When the thickness prediction value deviates from the target value h target When , calculate the evaporation source power adjustment ΔP, the calculation process is as follows: Among them, K p With K i is the compensation coefficient, t is the length of the integration time window; If the optical judgment is qualified but the thickness judgment fails, the evaporation source temperature is adjusted Prioritize adjusting the deposition rate. If the thickness passes but the quality confidence level is in the range [0.7, 0.9), trigger the plasma excitation frequency compensation.

7. The method for detecting the coating quality of a coating machine according to claim 6, characterized in that: In S4, the process of obtaining the real-time risk index and performing graded warning and spatiotemporal correlation warning includes: Define the real-time risk index RI as the quality confidence attenuation term (1-C q ) and thickness offset The weighted sum of and the extreme value normalization of the thickness offset term; When RI ≥ 0.8 and When RI<0.8, it is judged as a first-level red warning, the coating machine will be shut down immediately and the vacuum chamber pressure relief protection will be activated. When it is judged as a Level 2 orange warning, the coating machine will slow down and activate the auxiliary evaporation source. When 0.4≤RI<0.6 and C q When <0.7, it is judged as a level 3 yellow warning, the coating machine keeps running but records the parameter deviation trend; Detect the moving average slope of the thickness prediction value. When the absolute value of the moving average slope is greater than 0.5σ h / min for 3 minutes, the deposition rate instability warning is triggered; If there are C q If the standard deviation exceeds 0.15, an early warning of film uniformity degradation is triggered; Obtain the correlation coefficient between the quality confidence and the thickness prediction value. When the correlation coefficient is lower than -0.5 and persists for 5 sampling cycles, it is determined that the decoupling of optical and mechanical properties is abnormal. After each batch of production, the real-time risk index RI is updated based on historical data thres =μ RI +2σ RI , where μ RI and σ RI The mean and standard deviation of the real-time risk index of this batch; A delayed confirmation mechanism is introduced. The response action will be executed only when the warning condition is triggered three times in a row with an interval of ≤10 seconds.

8. The method for detecting the coating quality of a coating machine according to claim 7, wherein: In S5, the process of designing a 3D visualization dashboard includes: The standardized thickness deviation is defined as the standard deviation of the thickness prediction value relative to the historical mean. The X-axis is defined as the standardized thickness deviation, the Y-axis as the quality confidence, and the Z-axis as the normalized real-time risk index. The parameter dimensions are unified through the 4σ principle and linear mapping to construct a three-dimensional data space base. Real-time data points are colored according to the warning level, and the quality judgment status is distinguished by spheres, cubes and cones. The historical data track is displayed in a semi-transparent band. The bandwidth is dynamically adjusted according to the thickness fluctuation coefficient to reflect the trend of process stability changes. The green qualified zone is delineated by the ellipsoid equation. The semi-axis length is set based on the film thickness tolerance, quality confidence threshold and real-time risk index upper limit. The spatial boundary represents the safe range of process parameters, and a visual alarm is triggered when the boundary is penetrated. The data point coordinates are refreshed every 100 milliseconds, and a 5-minute sliding window history is retained. When a real-time data point breaks through the qualified area and associated warning rules, boundary flashing and spatial annotation are activated; It supports free rotation in three-dimensional space and 0.5-5x zoom, as well as filtering and displaying data by time and parameter intervals, providing a multi-dimensional process analysis perspective.

9. A coating quality detection system for a coating machine, used to implement the coating quality detection method for a coating machine according to any one of claims 1 to 8, characterized in that: It includes a coating data acquisition module, a coating feature extraction module, a coating quality detection module and a dual-parameter coupling early warning module, wherein the modules are connected with electrical signals.

10. The coating quality detection system of a coating machine according to claim 9, characterized in that: The coating data acquisition module collects and pre-processes the spectral reflectance data and film thickness oscillation frequency data of the coating surface; The coating feature extraction module extracts coating quality features from the pre-processed spectral reflectance data and film thickness oscillation frequency data; The coating quality detection module, in combination with the coating quality characteristics, constructs a quality classification model and a thickness prediction model, and outputs the coating quality classification result, quality confidence, thickness prediction value and prediction error range respectively, and then performs classification judgment and thickness judgment, and formulates a corresponding compensation mechanism; The dual-parameter coupling early warning module obtains a real-time risk index based on the coating quality classification results and thickness prediction values, and simultaneously performs graded early warning and spatiotemporal correlation early warning. The quality inspection visualization module combines quality confidence, thickness prediction value and real-time risk index to design a three-dimensional visualization dashboard.

Images