A method for detecting the thickness and uniformity of a glue layer in a lens gluing centering process

Abstract

The application discloses a lens gluing centering process glue layer thickness and uniformity detection method, relates to optical element manufacturing detection technical field, and its technical points are: through extracting hidden glue liquid molecule crosslinking chemical bond vibration information in pressure sensor residual error signal, phase coherence analysis is carried out with the micro phase shift component caused by refractive index micro change in laser interference signal, and two-dimensional distribution state is obtained by inversion, and then according to the corresponding relationship between crosslinking density and refractive index, the refractive index of glue liquid of each detection point is corrected point by point, so that the detection result of glue layer thickness can truly reflect the actual state of glue layer; in addition, a two-dimensional joint evaluation system of glue layer real thickness and molecular crosslinking density is constructed; by analyzing the joint distribution characteristics of the two, not only the geometric thickness uniformity of the glue layer can be accurately evaluated, but also the molecular structure uniformity inside the glue layer can be comprehensively reflected.

Description

Technical Field

[0001] This invention relates to the field of optical component manufacturing and testing technology, specifically a method for detecting the thickness and uniformity of the adhesive layer during the lens bonding and centering process. Background Technology

[0002] Lens bonding is a critical process in optical system manufacturing. The quality of the adhesive layer directly determines the imaging accuracy, optical axis consistency, and lifespan of the optical system. Currently, the mainstream method for measuring adhesive layer thickness in the industry is laser interferometry. This method involves emitting a laser of a specific wavelength onto the bonding surface, collecting the interference fringes formed by the reflected light from the upper and lower surfaces of the adhesive layer, and calculating the adhesive layer thickness based on the relationship between the fringe spacing and the laser wavelength. In actual production applications, to simplify the calculation process and improve the detection speed, the industry generally assumes that the refractive index of the adhesive is uniform throughout the bonding surface, and uses only geometric thickness as the sole indicator of adhesive layer quality. However, during the implementation of the above technical solution, at least the following technical problems were discovered: Firstly, during the cross-linking and curing process of the adhesive, the reaction rate in different regions varies significantly due to various factors such as local pressure distribution, temperature field gradient, and adhesive flow state, which will form a radial refractive index gradient inside the adhesive layer. This gradient change will cause the propagation path of the laser in the adhesive layer to deviate nonlinearly, and the resulting system error cannot be eliminated by conventional equipment calibration methods, directly limiting the manufacturing precision of high-precision optical systems. Secondly, it can only identify macroscopic visible defects such as bubbles and excessive thickness, and cannot reflect the molecular structure state inside the adhesive layer; microscopic defects such as insufficient cross-linking and uneven curing gradient will not show obvious abnormalities in geometric thickness, but will generate residual internal stress inside the adhesive layer. With the passage of time and changes in ambient temperature, problems such as delamination and stress cracking are likely to occur, leading to premature failure of the optical system. To this end, we propose a method for detecting the thickness and uniformity of the adhesive layer during the lens bonding centering process. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: A method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering, comprising the following steps: Simultaneously acquire laser interference phase data, machine vision edge data, dual-sampling-rate pressure data, and ambient temperature data during the lens bonding and centering process; A differentiated detection region division scheme is generated based on laser interferometric phase data and historical anomaly distribution data, and multiple batches of spatiotemporally aligned detection sample sets are obtained; The infrasonic vibration time sequence signal corresponding to the chemical bond vibration of the cross-linking reaction of the adhesive molecules was separated from the dual sampling rate pressure data, and the micro-phase drift component caused by the slight change in the refractive index of the adhesive layer was extracted from the laser interference phase data. The sliding window phase coherence eigenvalues ​​of the micro-phase drift component and the infrasonic vibration time series signal are calculated. Then, the two-dimensional distribution matrix of the crosslinking density of the colloidal molecules containing radial distribution information is obtained by inversion through the phase coherence eigenvalues. The two-dimensional distribution matrix of the crosslinking density is temperature corrected by combining the ambient temperature data. The nonlinear correction value of the refractive index gradient at each detection point is calculated using the corrected two-dimensional distribution matrix of crosslinking density. The laser interferometry results are then corrected point by point to obtain the true adhesive layer thickness at each detection point. A two-dimensional joint probability distribution matrix of the true thickness of the adhesive layer and the molecular crosslinking density is constructed. By calculating the two-dimensional information entropy value of this matrix, the comprehensive uniformity evaluation value of the adhesive layer, which simultaneously reflects the geometric thickness uniformity and the molecular structure uniformity, is obtained. The system combines the actual adhesive layer thickness with the comprehensive uniformity assessment value to generate a comprehensive judgment value for bonding quality. This value is then compared with a preset threshold to determine whether the bonding quality is up to standard. For non-compliant products, the system outputs two-dimensional coordinates and type information of the two-dimensional associated defects.

[0004] Furthermore, the multiple sets of detection data collected simultaneously are acquired by laser interferometry, machine vision, pressure and temperature detection devices configured at the gluing centering station, and the four types of detection devices are started to collect data simultaneously through hardware synchronous triggering control, and all output data are timestamped. The pressure detection device operates in a dual sampling rate mode. The first sampling rate is used to collect low-frequency macroscopic pressure changes during the bonding and pressurization process, while the second sampling rate is higher than the first sampling rate and is used to capture high-frequency pressure residual signals containing molecular crosslinking vibration information. Based on the inherent characteristics of each type of data, the raw data are classified and identified to obtain laser interferometric phase data, machine vision edge data, dual sampling rate pressure data, and ambient temperature data.

[0005] Furthermore, the process of generating a differentiated detection region segmentation scheme is as follows: By fitting the optical reference surface of the lens surface with laser interferometry phase data, a three-dimensional Cartesian coordinate reference with the geometric center of the lens as the origin is constructed. The bonded surface is divided radially into three continuous regions: a central circle, a middle ring, and an edge ring. Retrieve the spatial distribution statistics of abnormal crosslinking density and abnormal phase coherence in historical production batches, and calculate the probability of anomalies in each region; By combining the bonding accuracy index and the probability of anomalies in each region, the density of detection points and the detection interval in the corresponding region are determined, and a differentiated detection area division scheme is generated.

[0006] Furthermore, the steps for obtaining multiple batches of spatiotemporal alignment detection sample sets are as follows: Based on the differentiated detection area division scheme, multiple independent detection tasks are generated, and the detection point coordinates, detection time, sampling rate and trigger delay parameters of each task are configured. The original data sources are continuously tested according to the preset detection sequence to obtain the corresponding detection information. Based on the detection time of each detection point, the four types of detection data at the same spatial location are packaged into independent data slices, and all data slices are time stamp verified and spatial coordinate matched to remove invalid slices. The verified data slices are arranged in the detection order to form multiple batches of microsecond-level spatiotemporally aligned detection sample sets.

[0007] Furthermore, the process of separating the infrasound vibration time sequence signal is as follows: The dual sampling rate pressure data corresponding to each detection point is extracted from the detection sample set, and low-frequency macroscopic pressure trend signal is obtained by low-pass filtering. The low-frequency macroscopic pressure trend signal is subtracted from the original dual-sampling-rate pressure data to obtain the high-frequency pressure residual component. Then, the high-frequency residual component is bandpass filtered to retain the signal components within the characteristic frequency range of the adhesive crosslinking reaction. Using the ambient noise collected during the no-load operation of the equipment as a reference signal, adaptive noise cancellation is performed on the characteristic frequency signal to finally obtain the infrasonic vibration time sequence signal corresponding to the chemical bond vibration of the cross-linking reaction of the adhesive molecules.

[0008] Furthermore, the specific steps for calculating the phase coherence eigenvalues ​​of the sliding window are as follows: The laser interference phase data corresponding to each detection point is extracted from the detection sample set. After polynomial fitting detrending processing, the macroscopic phase change caused by the overall lens displacement and ambient temperature drift is removed, and the micro-phase drift component caused by the slight change in the refractive index of the adhesive layer is obtained. The micro-phase drift component and the infrasonic vibration time sequence signal are divided into multiple sliding time windows of equal length, with the sliding step size not exceeding one-fifth of the window length; Perform Hilbert transform on the two signals in each window to obtain analytic signals, calculate the cross-correlation function of the two analytic signals, and determine the phase difference based on the peak position of the cross-correlation function; The phase coherence coefficient of each window is calculated based on the phase difference. Then, a weighting coefficient is set according to the time distance between the window and the detection time. The coherence coefficients of all windows are weighted and averaged to obtain the sliding window phase coherence characteristic value of the detection point.

[0009] Furthermore, the process of inverting the two-dimensional distribution matrix of the crosslinking density of the adhesive layer molecules is as follows: Beforehand, the phase coherence characteristic value and the crosslinking density of the adhesive layer molecules were established by calibrating the standard adhesive sample through phase coherence test and thermal analysis, and the initial crosslinking density value of each detection point was obtained according to the mapping relationship. An adaptive noise template is constructed based on historical qualified batch data to perform noise correction on the initial crosslinking density value, thereby eliminating the influence of system noise and random noise. Spatial interpolation is performed on the corrected crosslinking density values ​​to generate a two-dimensional distribution matrix of crosslinking density of the adhesive layer molecules containing radial distribution information.

[0010] Furthermore, the process for obtaining the true adhesive layer thickness is as follows: Based on the corrected two-dimensional crosslinking density distribution matrix, the crosslinking density gradient value between each detection point and its neighboring detection points is calculated; The actual refractive index of the adhesive at the laser interferometric detection wavelength was determined in advance under different crosslinking density gradients and different ambient temperatures, and a nonlinear correspondence between crosslinking density gradient, ambient temperature and refractive index change was established. Based on the corresponding relationship, the nonlinear correction value of the refractive index gradient at each detection point is obtained, and the standard refractive index of the adhesive is corrected point by point to obtain the actual refractive index at each detection point. The initial true adhesive layer thickness is calculated by combining the optical path difference obtained from the laser interference phase data with the actual refractive index of each detection point. Then, outliers in the initial thickness are removed by iterative weighted least squares method to obtain the final true adhesive layer thickness at each detection point.

[0011] Furthermore, the steps to obtain the overall uniformity evaluation value of the adhesive layer are as follows: The actual adhesive layer thickness values ​​of all test points are divided into multiple continuous equal-width intervals, and the crosslinking density values ​​of all test points are divided into the same number of continuous equal-width intervals; The number of detection points in each thickness interval and cross-linking density interval is counted, and the combined probability value of each combination is obtained by dividing the counted number by the total number of detection points. Arrange all joint probability values ​​in interval order to construct a two-dimensional joint probability distribution matrix of the true thickness of the adhesive layer and the molecular crosslinking density; The two-dimensional information entropy value of the matrix is ​​calculated and normalized to obtain the overall uniformity evaluation value of the adhesive layer.

[0012] Furthermore, the process of generating a comprehensive judgment value for adhesive quality is as follows: The average adhesive layer thickness and average crosslinking density of the three regions—the central circle, the middle ring, and the edge ring—were calculated respectively, and the thickness deviation and crosslinking density deviation of each region were obtained by comparing them with the preset standard values. Multiply the thickness deviation value of each region by the crosslinking density deviation value to obtain the comprehensive deviation value of each region; The weight coefficients of each region are determined by the analytic hierarchy process, and the weighted summation of the comprehensive deviation values ​​of each region is used to obtain the overall adhesive layer thickness assessment value. Weighting coefficients are assigned to the thickness assessment value and the overall uniformity assessment value respectively, and then the overall judgment value of the bonding quality is obtained by weighted summation.

[0013] This invention provides a method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering, which has the following advantages: By extracting the vibrational information of cross-linked chemical bonds in the adhesive molecules hidden in the residual signal of the pressure sensor, and performing phase coherence analysis with the micro-phase drift component caused by the slight change in refractive index in the laser interference signal, the two-dimensional distribution of the cross-linking density of the adhesive layer is obtained. Then, based on the correspondence between cross-linking density and refractive index, the refractive index of the adhesive at each detection point is corrected point by point, thereby eliminating the systematic error caused by the refractive index gradient, so that the detection result of the adhesive layer thickness can truly reflect the actual state of the adhesive layer. In addition, geometric thickness is no longer used as the sole criterion for judging the quality of the adhesive layer. Instead, a two-dimensional joint evaluation system of the true thickness of the adhesive layer and the molecular cross-linking density is constructed. By analyzing the joint distribution characteristics of the two, not only can the geometric thickness uniformity of the adhesive layer be accurately evaluated, but the uniformity of the molecular structure inside the adhesive layer can also be comprehensively reflected, and microscopic hidden defects such as insufficient local cross-linking and uneven curing gradient that cannot be detected by traditional methods can be identified in advance.

[0014] Secondly, based on the degree of influence of different areas of the lens on optical imaging quality and combined with the spatial distribution pattern of defects in historical production data, a differentiated detection point density setting is adopted. The detection point density is increased in the central area with the greatest impact on imaging quality to ensure the detection accuracy of key areas. The detection point density is appropriately reduced in the edge areas with less impact on imaging quality to reduce unnecessary detection workload. While ensuring overall detection quality, the detection efficiency is improved, which can better meet the needs of mass industrial production.

[0015] Since the overall solution focuses on data processing and analysis methods, there is no need to modify the existing hardware equipment such as laser interferometers and pressure sensors on the production line. It can be achieved simply by upgrading the software algorithm. This allows for the rapid upgrading of the existing detection system without a large increase in equipment investment. The upgrade cost is low, the technology is highly compatible, and it is easy to promote and apply it widely in the industry. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of lens adhesive layer detection in this invention. Figure 2 This is a flowchart of the dual-sampling-rate pressure data processing and infrasound signal extraction in this invention; Figure 3 This is a schematic diagram illustrating the principle of sliding window phase coherence calculation in this invention. Figure 4 This is a graph of pressure data with dual sampling rates in this invention. Detailed Implementation

[0017] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: Please see Figure 1 This embodiment provides a method for detecting the thickness and uniformity of the adhesive layer during the centering process of lens bonding. Taking the bonding process of a K9 glass lens with a diameter of 50mm and a designed adhesive layer thickness of 10μm as an example, the method involves preheating the equipment for a period of time until the temperature stabilizes before starting the detection to avoid system errors introduced by environmental fluctuations. The specific implementation process is as follows: A testing system was built at the gluing centering station. This system included a laser interferometer, an industrial camera, and a high-precision piezoelectric pressure sensor. The laser interferometer, a helium-neon laser interferometer, was vertically mounted 120mm above the gluing lens. The coaxiality error between the beam center and the lens's geometric center did not exceed 0.02mm, the beam divergence angle was less than 0.5mrad, and the phase measurement resolution reached 0.01rad. The industrial camera, a global shutter CMOS camera, was mounted beside the laser interferometer, with its lens at a 45° angle to the lens surface. It was used to acquire interference fringe images and assist in... Positioning the lens center; a high-precision piezoelectric pressure sensor using a quartz crystal piezoelectric sensor is used. Three sensors are evenly distributed in an equilateral triangle at the contact interface between the annular pressure head and the lens of the centering fixture to form a pressure monitoring array, which can simultaneously monitor pressure changes in three directions and capture pressure fluctuations. The sensor output is connected to a charge amplifier to effectively amplify weak infrasound signals. In addition, in order to understand the temperature environment, a PT1000 platinum resistance temperature sensor is also used, which is installed close to the fixture base, no more than 5mm away from the bonding surface, for real-time monitoring of the ambient temperature and fixture temperature during the bonding process.

[0019] A FPGA hardware synchronization trigger card is used for nanosecond-level synchronous control of four types of devices. The four output channels of the trigger card are connected to the external trigger input of the laser interferometer, the trigger input of the industrial camera, the sampling trigger of the pressure sensor, and the sampling trigger of the temperature sensor, respectively. The trigger card outputs four synchronous TTL level signals, and the rising edge triggers all devices to start acquisition simultaneously. All output data are directly stamped by the hardware clock of the trigger card during data acquisition, ensuring that the timing synchronization error of different types of data is small. In contrast, traditional detection often uses software synchronization, which has a large synchronization error and cannot meet the synchronous analysis requirements of infrasound vibration and micro-phase drift. This method of directly marking the original data with hardware timestamps avoids the synchronization error caused by software transmission delay and is a necessary prerequisite for realizing subsequent phase coherence calculation.

[0020] The pressure sensor employs a dual-sampling-rate mode for parallel operation, primarily implemented through a frequency divider circuit within the FPGA. Both sampling data streams share the same analog front-end, ensuring data consistency and avoiding sensitivity matching errors caused by multiple sensors. Specifically, the first sampling rate is set to 100Hz to acquire low-frequency macroscopic pressure changes from 0.1Hz to 10Hz during the adhesive bonding process, monitoring the stability of the clamp pressurization. An alarm is automatically triggered when pressure fluctuations exceed ±0.1N. The second sampling rate is set to 5000Hz, 50 times the first rate. This rate satisfies the Nyquist sampling theorem, enabling the complete capture of 5-20Hz infrasonic signals generated by the cross-linking reaction of adhesive molecules. While conventional pressure sensors are only used to monitor macroscopic pressure and their high-frequency response characteristics are never utilized to capture molecular-level reaction signals, this dual-sampling-rate mode allows for the simultaneous acquisition of macroscopic process parameters and microscopic reaction information without increasing hardware costs, achieving a revolutionary expansion of the sensor's applications.

[0021] like Figure 2 As shown, the collected raw data is classified according to its inherent characteristics. For example, 16-bit binary data with continuous phase change characteristics transmitted by the phase output interface of the laser interferometer is identified as laser interferometric phase data; 8-bit grayscale image data containing edge grayscale abrupt change information and output by the industrial camera is identified as machine vision edge data; floating-point pressure signals containing both 100Hz and 5000Hz sampling rates are identified as dual sampling rate pressure data; and 16-bit temperature digital signals that change slowly over time are identified as ambient temperature data.

[0022] Furthermore, zero-point calibration and wavelength calibration are required before each test to improve accuracy. Zero-point calibration involves placing a standard 10μm thick quartz plate on the fixture, collecting 10 seconds of laser interference phase data and pressure data, calculating and storing the system's zero-point deviation, which is automatically deducted from subsequent test data. Wavelength calibration is performed by calibrating the interferometer's wavelength using a standard helium-neon laser source. The calibration cycle is every 8 hours, and recalibration is required if the adhesive batch is changed, the fixture is adjusted, or the equipment is shut down for more than 2 hours. The lens surface roughness should be less than 0.5nm; otherwise, scattering errors will be introduced, so surface cleaning and roughness checks are necessary before testing. In addition, before each batch of production, all sensors need to be calibrated for sensitivity using standard pressure and temperature sources to ensure that the calibration error is controlled within ±0.1%. The sensitivity differences of pressure sensors are compensated for by calibration coefficients, and the nonlinearity error of temperature sensors is corrected by third-order polynomial fitting.

[0023] Using the initial 100 frames of phase data acquired by a laser interferometer, the optical reference plane of the upper surface of the lens is fitted using the least squares method. The root mean square residual of the fit is less than 0.01 μm, ensuring the accuracy of the coordinate reference. Simultaneously, the outer contour ellipse of the lens is extracted using machine vision edge data. The tilt angle of the lens is calculated by fitting the ellipse. The formula for calculating the tilt angle is θ = arctan(b / a), where θ is the tilt angle of the lens in degrees, and a and b are the major and minor semi-axes of the outer contour ellipse of the lens extracted by machine vision, respectively. Based on the calculated tilt angle, the incident angle of the laser interferometer is adjusted to ensure that the beam is perpendicularly incident on the cemented surface. When the deviation between the optical center and the geometric center exceeds 0.02 mm, the fixture position is automatically adjusted to reposition the lens. A three-dimensional Cartesian coordinate reference is established with the geometric center of the lens as the origin, where the XY plane is parallel to the lens surface, and the Z-axis is perpendicular to the lens surface and points to the laser interferometer.

[0024] The bonded surface is radially divided into three continuous regions: a central circular region, a middle annular region, and an edge annular region. The central circular region, with a radial range of 0-16.7 mm (corresponding to 0-1 / 3 of the lens diameter), is located at the center of the optical system's field of view and contributes over 70% of wavefront aberration, requiring the highest bonding precision with an allowable thickness deviation of ±0.5 μm. The middle annular region, with a radial range of 16.7-33.3 mm (corresponding to 1 / 3-2 / 3 of the lens diameter), contributes approximately 25% of wavefront aberration, and requires the highest bonding precision, with an allowable thickness deviation of ±0.5 μm. The next highest precision requirement is a thickness deviation of ±1μm; the radial range of the edge annular region is 33.3-50mm, corresponding to 2 / 3-1 times the lens diameter, with wavelet aberration contribution of less than 5%, and the bonding precision requirement is the lowest, with a thickness deviation of ±2μm; by retrieving the defect database of the past 1200 production batches, the number of occurrences of abnormal crosslinking density and phase coherence of the adhesive layer in each region was counted, and the probability of anomalies in the central region was calculated to be 0.082, in the middle region 0.037, and in the edge region 0.011.

[0025] The density of each region is determined according to the principle that the density of detection points equals the accuracy requirement coefficient multiplied by the anomaly probability coefficient. The accuracy requirement coefficient is inversely proportional to the accuracy requirement, with a value of 5 for the central region, 3 for the middle region, and 1 for the edge region. The anomaly probability coefficient is directly proportional to the anomaly probability, with a value of 1.64 for the central region, 0.74 for the middle region, and 0.22 for the edge region. The final result is 5 points / mm in the central region. 2 3 points / mm in the middle area 2 1 point / mm in the edge region 2 The total number of detection points is approximately 1200, which reduces the detection workload by 40% compared to the 2000 points used for uniform detection across the entire area, while ensuring the detection accuracy in key areas.

[0026] Traditional inspection methods employ uniform density full-area scanning, resulting in low efficiency and unsuitability for current needs. Therefore, we optimize resource allocation by combining optical contribution and historical defect distribution to improve efficiency while maintaining accuracy. For different types of optical systems, the area division and inspection point density can be flexibly adjusted. A large field-of-view optical system refers to one that covers the entire effective area of ​​the measured object. For example, in the inspection of large-diameter optical elements, the field of view must be ≥ the diameter of the measured object. Increasing the optical contribution of the edge regions allows for a higher inspection point density of up to 2 points / mm in these areas. 2 High-precision laser focusing systems require higher accuracy in the central area, and can increase the detection point density in the central area to 8 points / mm. 2 For small lenses with a diameter of less than 10mm, they can be combined into one area for detection; for large lenses with a diameter of more than 200mm, a transition ring area can be added to further refine the density of detection points.

[0027] For different glass materials (such as ZF7 and BK7), the optical path parameters of the laser interferometer need to be adjusted according to their refractive index; for different types of adhesives (such as epoxy and silicone), the mapping relationship table between phase coherence and crosslinking density, as well as the correspondence table between refractive index gradient and temperature, need to be re-established according to the standardized process; for spherical lenses, the focusing position of the laser interferometer is adjusted according to the radius of curvature to ensure that the beam is perpendicularly incident on the cemented surface; for aspherical lenses, the density of detection points is appropriately increased in areas with large curvature changes.

[0028] Based on the differentiated detection area division scheme, the entire detection process is decomposed into 1200 independent detection tasks, each corresponding to a detection point. Parameters such as the three-dimensional coordinates of the detection point, detection time, laser interferometer sampling rate, pressure sensor dual sampling rate, and temperature sensor sampling rate are configured for each task. The accuracy of the three-dimensional coordinates of the detection point is ±0.001mm, and the delay time accuracy between the detection time and the trigger time is ±1μs. Continuous detection is performed sequentially from the center area to the edge area according to a preset detection sequence. The detection order from center to edge is chosen because the adhesive spreads from the center to the edge, and the cross-linking reaction in the center area reaches a stable state first. Detecting the center area first reduces the overall detection time and improves production efficiency. Furthermore, the detection sequence can be dynamically adjusted based on historical defect data. Areas with a high historical defect incidence rate are prioritized for detection. When 100 consecutive products in a certain area are defect-free, the detection point density in that area can be appropriately reduced to further improve detection efficiency. At the detection time of each detection point, four types of data are simultaneously collected. Based on the detection time, the four types of data at the same spatial location and the same timestamp are packaged into an independent data slice.

[0029] The data slices are stored in little-endian mode. The first 8 bytes are timestamps, the next 12 bytes are spatial coordinates (4 bytes each for x, y, and z, single-precision floating-point numbers), followed by 200 bytes of laser interferometric phase data (100 double-precision floating-point numbers), 4000 bytes of high-frequency pressure data (500 double-precision floating-point numbers), 80 bytes of low-frequency pressure data (10 double-precision floating-point numbers), and 8 bytes of temperature data (1 double-precision floating-point number). The slice size is approximately 2.2 MB.

[0030] After slicing is completed, all data slices are double-verified. First, the timestamp is verified by calculating the difference between the slice timestamp and the preset detection time. Slices with a difference exceeding 1 μs are considered invalid. Then, the spatial coordinates are verified by calculating the deviation between the actual coordinates of the slice and the preset coordinates. Slices with a deviation exceeding 1 μm are considered invalid. The verified data slices are arranged in the detection order to form multiple batches of microsecond-level spatiotemporally aligned detection sample sets. The sample sets are stored in binary format, compared to text format.

[0031] Lossless compression algorithms are used to compress the test data, achieving a compression ratio of approximately 3:1, reducing data storage while preserving all original information. Multiple batches of test sample sets can be used to construct product quality trend models. By analyzing test data from consecutive batches, process drift can be detected in advance, production parameters can be adjusted in a timely manner, and batch defects can be prevented. Traditional test data is stored separately by type, which cannot guarantee spatiotemporal consistency. However, by using data slicing and packaging and double verification, strict alignment of multi-source data at the same spatiotemporal location is ensured, laying the foundation for subsequent cross-physical quantity coupling analysis.

[0032] Example 2: Based on Example 1, this embodiment also provides a detection system for the thickness and uniformity of the adhesive layer during lens bonding and centering. The system is divided into data acquisition, signal processing, quality judgment, data storage, and human-machine interaction. Data acquisition has the highest priority to ensure real-time performance. GPU parallel computing technology is used to accelerate the signal processing module to meet the needs of high-speed production lines. The system has an anomaly handling mechanism: when a sensor disconnection is detected, detection automatically stops and an audible and visual alarm is issued; when data transmission is interrupted for more than 1 second, data at that detection point is automatically reacquired; when the lens center deviation exceeds 0.02mm, the fixture position is automatically adjusted and detection is re-inspected; when the ambient temperature exceeds the range of 15℃-35℃, the system automatically issues an alarm and stops detection. All detection data is bound to a unique product identifier and stored in a local database, while also being uploaded to the production management system. The data is retained for at least 3 years, and all detection history can be queried through the product identifier.

[0033] Extract the dual sampling rate stress data corresponding to each detection point from the sample set, such as Figure 4 As shown, the average pressure data is first subtracted to eliminate sensor zero-point drift. Then, an 8th-order Butterworth low-pass filter is applied with a cutoff frequency of 10Hz to obtain a low-frequency macroscopic pressure trend signal, reflecting the overall change in clamp pressure. The original 5000Hz pressure data is subtracted from the low-frequency trend signal to obtain a high-frequency pressure residual component. The high-frequency residual component is then subjected to an 8th-order Butterworth bandpass filter with a passband of 5Hz to 20Hz to retain signal components within the characteristic frequency range of the adhesive crosslinking reaction. The 8th-order filter is chosen because the 8th-order Butterworth filter has the flattest amplitude response in the passband, and the transition band is narrow enough to effectively separate the target signal and interference signal. If the order is too low, the transition band will be too wide, introducing interference. If the order is too high, it will increase the computational load and produce phase distortion.

[0034] Using the ambient noise collected during 30 minutes of no-load operation of the equipment as the reference signal, an 8th-order FIR least mean square adaptive noise cancellation algorithm is employed to remove interference signals generated by environmental vibration and equipment operation. The algorithm step size is set to 0.01, and the weight update formula is w(n+1) = w(n) + 2μe(n)x(n), where w(n) is the filter weight vector of the nth iteration, w(n+1) is the filter weight vector of the (n+1)th iteration, μ is the algorithm step size (set to 0.01), e(n) is the error signal of the nth sampling point, and x(n) is the reference input signal of the nth sampling point, with units of Newtons. The algorithm is considered to have converged when the root mean square error signal of 100 consecutive sampling points is less than 0.0001N, achieving an optimal balance between convergence speed and steady-state error. After processing, a pure infrasonic vibration time-series signal corresponding to the cross-linking reaction of the adhesive molecules is obtained. The cross-linking reaction of the adhesive is an exothermic process, and changes in the local reaction rate will cause small temperature fluctuations, which in turn cause thermal expansion and contraction of the adhesive, generating infrasonic vibrations with a frequency of 5-20Hz. This vibration can be transmitted to the pressure sensor through a lens. Through a high-precision pressure sensor and signal processing, non-invasive detection of this micro-vibration signal can be achieved.

[0035] Simultaneously, the laser interference phase data corresponding to each detection point is extracted, and a third-order polynomial fitting detrending process is used to remove the macroscopic phase change caused by the overall lens displacement and ambient temperature drift. The fitting residual is the micro-phase drift component caused by the slight change in the refractive index of the adhesive layer, and its amplitude is usually between 0.01 rad and 0.1 rad.

[0036] The micro-phase drift component and the infrasonic vibration timing signal were divided into multiple sliding time windows of equal length. The window length was set to 100 ms, and the sliding step size was set to 10 ms, not exceeding one-fifth of the window length. Each window contained 100 laser phase points and 500 pressure points. The window length of 100 ms was chosen because the time constant of the adhesive crosslinking reaction is about 50 ms, and a 100 ms window can completely cover one reaction cycle. The sliding step size of 10 ms was chosen not only to balance time resolution and computational load, but more importantly, to avoid spectral leakage in phase coherence calculation. When the step size is too large, the overlap rate of adjacent windows is insufficient, which will lead to increased fluctuations in the coherence coefficient and increased measurement error.

[0037] Perform Hilbert transforms on the two signals within each window to convert the real signals into analytic signals, thereby obtaining the instantaneous phase and instantaneous amplitude of the signals. The Hilbert transform can set the negative frequency components of the real signal to zero while retaining the positive frequency components, thus obtaining a unique instantaneous phase. This is a key step in calculating phase coherence. Calculate the cross-correlation function of the two analytic signals: , where R xy(τ) represents the cross-correlation function value of signals x and y under a delay of τ, where τ is the time delay in milliseconds, N is the number of sampling points within the window, x(t) is the analytic signal of the micro-phase drift component in radians (rad), y(t) is the analytic signal of the infrasonic vibration time sequence signal in Newtons, and y∗(t) is the conjugate complex number of y(t). The peak position of the cross-correlation function corresponds to the optimal matching delay of the two signals, and the phase difference ϕ between the two signals is determined by the peak position. The phase coherence coefficient of each window is calculated based on the phase difference. Where γ(τ) is the phase coherence coefficient at a delay of τ, and R xx (0) and R yy (0) represents the values ​​of the autocorrelation functions of signals x and y at zero points, respectively. This formula obtains the coherence coefficient with a value range of 0 to 1 by normalizing the amplitude of the cross-correlation function. The larger the value, the higher the phase coupling degree of the two signals. The sliding window cross-correlation phase coherence is chosen because it has a small computational load, good real-time performance, and stronger adaptability to non-stationary signals, making it suitable for online detection in industrial production.

[0038] like Figure 3 As shown, the decreasing weighting coefficient is set according to the time distance between the window and the detection time: , where w k t represents the weighting coefficient for the k-th window. k Let t0 be the detection time, τ0 be the time constant (set to 50ms), and e be the natural constant. Windows closer to the detection time have larger weighting coefficients because crosslinking is an irreversible exothermic process, and signals closer to the detection time better reflect the current crosslinking state. A weighted average of the coherence coefficients of all windows is used to obtain the sliding window phase coherence characteristic value for that detection point. Changes in the crosslinking reaction rate are the common cause of phase drift and pressure vibration. A faster reaction rate leads to more drastic changes in refractive index and stronger vibrations from thermal expansion. Therefore, higher phase coherence indicates a more vigorous crosslinking reaction and higher crosslinking density. When the phase coherence coefficient of a detection point is below 0.3 and the coefficients of adjacent detection points are all above 0.6, it is determined to be interference from bubbles or impurities. The data at this point is replaced by interpolation using the average of eight adjacent points. The molecular crosslinking density is retrieved through the phase coherence inversion of two classical signals, achieving online non-destructive testing without any chemical analysis. Crosslinking density, also known as curing degree, refers to the proportion of crosslinked monomer units to the total monomer units.

[0039] For new adhesive types, a mapping relationship is established according to a standardized process, as follows: First, 50 standard samples with different curing times were prepared, and three parallel differential scanning calorimetry (DSC) tests were performed on each sample. The crosslinking density of the adhesive was calculated by the area of ​​the curing exothermic peak. At the same time, phase coherence tests were performed on the same samples, and the corresponding phase coherence characteristic values ​​were recorded. Finally, a mapping relationship table between the phase coherence characteristic values ​​and the crosslinking density of the adhesive layer molecules was established by linear regression analysis. The calibration process should be carried out at the same ambient temperature. For different batches of the same type of adhesive, a rapid calibration method can be used, which only requires the preparation of 10 standard samples. The original mapping relationship table can be fine-tuned by using a transfer learning algorithm, which can shorten the calibration time.

[0040] An adaptive noise template was constructed based on data from 300 historical qualified batches. The mean μ and standard deviation σ of the phase coherence characteristics of the detection points at the same location in the historical qualified batches were statistically analyzed. Values ​​exceeding the range of μ±3σ were identified as noise. This template was used to correct the initial crosslinking density value for noise, eliminating the influence of system noise and random noise. The corrected crosslinking density value was spatially interpolated using the natural boundary cubic spline interpolation method, i.e., the second derivative at the boundary is 0. This boundary condition is suitable for the smooth distribution characteristics of the crosslinking density of the adhesive layer. The interpolation error is less than 0.005μm, generating a two-dimensional distribution matrix of the crosslinking density of the adhesive layer molecules containing radial distribution information. The interpolation resolution is 0.1mm.

[0041] Temperature correction was performed on the two-dimensional distribution matrix of crosslinking density using ambient temperature data. A quadratic polynomial relationship between the temperature and crosslinking density correction coefficients was established experimentally beforehand: k(T) = aT 2 +bT+c, where k(T) is the crosslinking density correction coefficient at temperature T, T is the real-time ambient temperature in degrees Celsius, and a is the quadratic coefficient, taken as 0.00001 / ℃. 2 b is the coefficient of the first term, taken as 0.0018 / ℃, c is the constant term, taken as -0.016, applicable temperature range 15℃-35℃, the crosslinking density value of each detection point is corrected according to the real-time collected ambient temperature to eliminate the influence of ambient temperature on the crosslinking reaction rate.

[0042] Based on the corrected two-dimensional distribution matrix of crosslinking density, the central difference method was used to calculate the crosslinking density gradient between each detection point and its eight adjacent detection points. The actual refractive index of the adhesive at a wavelength of 632.8 nm was pre-determined under different crosslinking density gradients and different ambient temperatures. Using an Abbe refractometer, the refractive index of the adhesive under different crosslinking density gradients was measured every 1 °C within the range of 15 °C to 35 °C, and a nonlinear correspondence table of crosslinking density gradient, ambient temperature and refractive index change was established. The traditional laser interferometry method uses a single standard refractive index and ignores the refractive index change caused by the crosslinking density gradient. In fact, for every 0.1 / mm change in the crosslinking density gradient, the refractive index will change by about 0.001. For a 10 μm thick adhesive layer, this will introduce a systematic error of about 0.0067 μm. Although the error at a single point is small, the cumulative error of the entire adhesive surface will cause the wavefront aberration to exceed the standard.

[0043] The nonlinear correction value Δn(x,y) of the refractive index gradient at each detection point is obtained by querying the corresponding relationship table. The standard refractive index of the adhesive, n0=1.5, is corrected point by point to obtain the actual refractive index at each detection point: n(x,y) = n0 + Δn(x,y), where n(x,y) is the actual refractive index at detection point (x,y), and n0 is the standard refractive index of the adhesive. The optical path difference is calculated using laser interference phase data: ΔL(x,y) = λΔϕ / 2π, where ΔL(x,y) is the optical path difference of the adhesive layer at detection point (x,y), in micrometers; λ is the laser wavelength, in nanometers; and Δϕ is the phase difference of the reflected light from the upper and lower surfaces of the adhesive layer, in radians. To be applicable to adhesive layer thicknesses of 5μm-20μm, when the thickness exceeds this range, an interface reflection correction term ΔLr needs to be introduced into the optical path difference formula. The corrected optical path difference formula is ΔL(x,y) = λΔϕ / 2π + ΔLr. r ΔL r The interface reflection correction term is in micrometers to ensure measurement accuracy. The initial true adhesive layer thickness is calculated as: d0(x,y) = ΔL(x,y) / 2n(x,y). This formula takes into account the round-trip propagation of the laser in the adhesive layer and the actual refractive index of each detection point, thus eliminating the systematic error caused by the refractive index gradient compared to the traditional formula.

[0044] Iterative weighted least squares method is used to remove outliers from the initial thickness. Tukey's double-weighting function is employed; this robust statistical function effectively suppresses the impact of outliers on statistical results and is widely used in data cleaning and outlier removal. The weighting formula is as follows: Where w(r) is the weight when the residual is r, r is the ratio of the residual to the absolute deviation of the median, and 4.685 is a robust statistical constant; the initial value of the iteration is the median of all data, and the iteration process is repeated until the change of the median between two iterations is less than 0.001 μm, so as to obtain the final true adhesive layer thickness at each detection point.

[0045] To address the total measurement error, the root-squaring method was used for synthesis. A laser wavelength drift of 0.001 nm resulted in a thickness measurement error of 0.0016 μm, a lens surface roughness of 0.5 nm introduced a scattering error of 0.002 μm, a sensor sensitivity deviation of 0.1% resulted in a crosslinking density measurement error of 0.2%, a time synchronization error of 1 μs introduced a phase error of 0.0005 μm, and the interface reflection error was less than 0.001 μm (applicable to adhesive layer thicknesses of 5 μm-20 μm). Ultimately, the total thickness measurement error was ≤ ±0.8%, and the total crosslinking density measurement error was ≤ ±1%, meeting the manufacturing requirements of high-precision optical systems.

[0046] Traditional uniformity assessment methods can only measure the distribution of thickness and crosslinking density separately, failing to reflect the correlation between the two. In reality, the distributions of thickness and crosslinking density are interdependent; localized thicker thickness is often accompanied by lower crosslinking density. Two-dimensional information entropy can simultaneously consider the joint distribution characteristics of both variables, thus providing a more comprehensive assessment of the overall uniformity of the adhesive layer. A two-dimensional joint probability distribution matrix of the true thickness and molecular crosslinking density of the adhesive layer is constructed. The true thickness values ​​of all detection points are divided into 10 consecutive equal-width intervals, each interval being 1% of the standard adhesive layer thickness (0.1 μm). Similarly, the crosslinking density values ​​of all detection points are divided into 10 consecutive equal-width intervals, each interval being 1% of the standard crosslinking density. The number of detection points Nij in each combination of thickness and crosslinking density intervals is counted. Dividing this count by the total number of detection points N yields the joint probability value for each combination: P. ij =N ij / N, where P ij N represents the joint probability that a detection point falls within the i-th thickness interval and the j-th crosslinking density interval. ij Let N be the number of detection points falling within the i-th thickness interval and the j-th crosslinking density interval, and let N be the total number of detection points. Arrange all joint probability values ​​in interval order to construct a 10×10 two-dimensional joint probability distribution matrix.

[0047] Calculate the two-dimensional information entropy value of this matrix: Where H is the two-dimensional joint distribution information entropy in bits, and 10 represents the number of intervals. Information entropy is a fundamental indicator in information theory for measuring the degree of disorder in a system. When both thickness and crosslinking density are uniformly distributed, the joint probability distribution is close to uniform, and the information entropy value is minimum. When local defects exist, the joint probability distribution will show a peak, and the information entropy value will increase significantly. The calculated two-dimensional information entropy value is normalized to obtain the overall uniformity evaluation value of the adhesive layer: U = 1 - H / H max Where U is the overall uniformity evaluation value of the adhesive layer, H is the two-dimensional joint distribution information entropy, and H max H represents the maximum value of the two-dimensional information entropy when all joint probability values ​​are equal. max =log2(100)≈6.64; The smaller the information entropy value, the larger the normalized comprehensive uniformity evaluation value, indicating that the geometric thickness and molecular structure uniformity of the adhesive layer are better; Traditional uniformity evaluation only calculates the standard deviation of thickness, which cannot reflect the uniformity of molecular structure. For example, some areas have uniform thickness but uneven crosslinking density. The standard deviation may show normal, but the information entropy will increase significantly. However, by using the information entropy of dual-dimensional joint distribution, the uniformity of geometric thickness and molecular structure are evaluated at the same time, which can detect microscopic defects that cannot be detected by traditional methods in advance.

[0048] The average adhesive layer thickness and average crosslinking density of the three regions—the central circular region, the middle annular region, and the edge annular region—were calculated separately. These values ​​were compared with preset standard values ​​to obtain the thickness deviation and crosslinking density deviation for each region. The absolute value of the thickness deviation was multiplied by the absolute value of the crosslinking density deviation to obtain the comprehensive deviation value for each region. The weighting coefficient for insufficient crosslinking was 1.2, and the weighting coefficient for excessive crosslinking was 0.8. Analysis of 200 failed samples revealed that insufficient crosslinking accounted for 65% of failures, while excessive crosslinking accounted for 35%. Therefore, the weighting coefficient for insufficient crosslinking was set to a higher value to highlight its impact on the long-term reliability of the product.

[0049] The weight coefficients of each region were determined using the analytic hierarchy process (AHP) to construct a judgment matrix. The importance ratio of the central region to the middle region was 2:1, the central region to the edge region was 5:1, and the middle region to the edge region was 3:1. The maximum eigenvalue and eigenvector of the judgment matrix were calculated. After a consistency test with a consistency ratio of 0.05 (less than 0.1), the weight coefficients for each region were determined as follows: 0.5 for the central circular region, 0.3 for the middle annular region, and 0.2 for the edge annular region. The weighted sum of the comprehensive deviation values ​​of each region yielded the overall adhesive layer thickness assessment value D. A weight of 0.6 was assigned to the thickness assessment value, and 0.4 was assigned to the comprehensive uniformity assessment value. The two values ​​are weighted and summed to obtain the comprehensive judgment value of the bonding quality: Q = 0.6D + 0.4U. The calculated comprehensive judgment value of bonding quality Q is compared with the preset qualified threshold. The qualified threshold of 0.85 is determined by receiver operating characteristic (ROC) analysis. ROC analysis is a commonly used method for evaluating the performance of classification models. By selecting different thresholds, a balance can be achieved between precision and recall. For high-precision optical systems, the threshold can be increased to 0.9; for ordinary optical systems, it can be reduced to 0.8. If Q > 0.85, the lens bonding quality is judged to be qualified; if Q ≤ 0.85, it is judged to be unqualified.

[0050] When a defect is deemed unqualified, the areas with excessive adhesive layer thickness, abnormal crosslinking density, and overlapping defects are located separately. Excessive thickness is defined as a thickness deviation exceeding ±1 μm; insufficient crosslinking is defined as a crosslinking density below 95% of the standard value; excessive crosslinking is defined as a crosslinking density above 105% of the standard value; and poor uniformity is defined as a comprehensive uniformity assessment value below 0.8. A contour extraction algorithm is used to extract the contours of the areas with excessive thickness and abnormal crosslinking density, and the intersection area of ​​the two contours is calculated. The ratio of the intersection area to the minimum area of ​​the two contours is the overlap ratio. A two-dimensionally related defect area is defined as an area with an overlap ratio exceeding 30%. When the overlap ratio exceeds 50%, it is determined to be an abnormal crosslinking caused by excessive thickness, and rework should be carried out by adjusting the pressure distribution. When the overlap ratio is between 30% and 50%, it is determined to be mutually influential, and both pressure and curing time should be adjusted simultaneously. When the overlap ratio is below 30%, it is determined to be an independent defect, and each should be handled separately.

[0051] Output the two-dimensional coordinate information, defect type information and defect severity level information of all defect areas. The defect information is output in JSON format, including fields such as defect ID, defect type, defect center coordinate, defect area, and defect severity, which can be directly imported into the production management system for data analysis and traceability. The defect severity is divided into three levels: slight, normal and severe, corresponding to deviations less than 1.5 times the tolerance, 1.5 to 3 times the tolerance, and greater than 3 times the tolerance respectively. The processing procedures for defects of different levels are as follows: slight defects do not require rework and directly flow into the next process, but need to be marked in the production management system to track their subsequent usage. Normal defects are reworked by local heating at a heating temperature of 60°C for a heating time of 30 seconds, and re-inspected after rework. Severe defects are directly scrapped, and the location and type of the defects are recorded at the same time for process improvement.

[0052] Assuming that the comprehensive judgment value of the lens in the inspection is 0.92, it is judged as qualified, the average thickness of the central area is 9.92μm, the thickness deviation is 0.08μm, the average value of the crosslinking density is 0.86, the deviation is 0.02, and the comprehensive uniformity evaluation value is 0.94. This method can be directly integrated into the full-automatic gluing and centering production line, realizing the full-process automation from loading, gluing, centering, inspection to unloading. The manipulator places the lens to be glued on the fixture to automatically position the center, starts the gluing and pressurization program while starting the inspection system. After the inspection is completed, the system automatically judges the quality, qualified products flow into the next process, unqualified products are automatically sorted to the rework area, and all inspection data are automatically uploaded to the production management system to generate a quality report.

[0053] Based on the aforementioned real-time detection method, we can feed the detection results back to the bonding and centering control system, realizing a shift from passive detection to active control. By adjusting the pressure distribution and curing time in real time, we can actively correct the thickness deviation and crosslinking density deviation of the adhesive layer, further improving the consistency of product quality. The detection results are transmitted in real time to the bonding and centering PLC control system via the Modbus TCP protocol. The control system generates corresponding pressure and curing time control signals according to the preset control algorithm, ensuring real-time control. During the bonding and centering process, real-time detection is performed every 100ms to obtain the crosslinking density distribution and thickness distribution of the adhesive layer. When a local crosslinking density is detected to be too low, the local pressure of the clamping head is adjusted. An additional 0.5N pressure is applied for 2 seconds in areas with low crosslinking density to promote adhesive flow and crosslinking reaction. When a local adhesive layer thickness exceeding 1μm is detected, the clamping force of the centering fixture is adjusted to apply greater clamping force to expel excess adhesive from the thicker areas. When the overall crosslinking density of the adhesive layer reaches a preset threshold of 0.85, the pressurization automatically stops and the UV curing stage begins. During UV curing, a detection is performed every second to monitor changes in crosslinking density in real time. When the crosslinking density reaches a preset threshold, the UV light source is automatically turned off to prevent over-curing. Through this real-time feedback control, the thickness and crosslinking density deviations of the adhesive layer can be actively corrected, significantly improving the product qualification rate.

[0054] Common defects in the production process can also be accurately located and addressed using this method: Insufficient cross-linking in the central area is usually caused by poor flow of adhesive in the central area or insufficient curing time, which can be improved by appropriately extending the pressurization time or adding local heating in the central area; Excessive thickness in the edge area is usually caused by insufficient pressure in the edge area and insufficient extrusion of adhesive, which can be improved by adjusting the pressure distribution of the clamp head and increasing the pressure in the edge area; Poor overall uniformity is usually caused by uneven mixing of adhesive or excessive pressurization speed, which can be improved by extending the mixing time of adhesive and slowing down the pressurization speed.

[0055] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples 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 implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0056] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering, characterized in that, The steps of this method are as follows: Simultaneously acquire laser interference phase data, machine vision edge data, dual-sampling-rate pressure data, and ambient temperature data during the lens bonding and centering process; A differentiated detection region division scheme is generated based on laser interferometric phase data and historical anomaly distribution data, and multiple batches of spatiotemporally aligned detection sample sets are obtained; The infrasonic vibration time sequence signal corresponding to the chemical bond vibration of the cross-linking reaction of the adhesive molecules was separated from the dual sampling rate pressure data, and the micro-phase drift component caused by the slight change in the refractive index of the adhesive layer was extracted from the laser interference phase data. The sliding window phase coherence eigenvalues ​​of the micro-phase drift component and the infrasonic vibration time series signal are calculated. Then, the two-dimensional distribution matrix of the crosslinking density of the colloidal molecules containing radial distribution information is obtained by inversion through the phase coherence eigenvalues. The two-dimensional distribution matrix of the crosslinking density is temperature corrected by combining the ambient temperature data. The nonlinear correction value of the refractive index gradient at each detection point is calculated using the corrected two-dimensional distribution matrix of crosslinking density. The laser interferometry results are then corrected point by point to obtain the true adhesive layer thickness at each detection point. A two-dimensional joint probability distribution matrix of the true thickness of the adhesive layer and the molecular crosslinking density is constructed. By calculating the two-dimensional information entropy value of this matrix, the comprehensive uniformity evaluation value of the adhesive layer, which simultaneously reflects the geometric thickness uniformity and the molecular structure uniformity, is obtained. The system combines the actual adhesive layer thickness with the comprehensive uniformity assessment value to generate a comprehensive judgment value for bonding quality. This value is then compared with a preset threshold to determine whether the bonding quality is up to standard. For non-compliant products, the system outputs two-dimensional coordinates and type information of the two-dimensional associated defects.

2. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The multiple sets of detection data collected simultaneously are acquired by the laser interferometry, machine vision, pressure and temperature detection devices configured at the gluing centering station. The four types of detection devices are started and collected synchronously through hardware synchronous triggering control, and all output data are timestamped. The pressure detection device operates in a dual sampling rate mode. The first sampling rate is used to collect low-frequency macroscopic pressure changes during the bonding and pressurization process, while the second sampling rate is higher than the first sampling rate and is used to capture high-frequency pressure residual signals containing molecular crosslinking vibration information. Based on the inherent characteristics of each type of data, the raw data are classified and identified to obtain laser interferometric phase data, machine vision edge data, dual sampling rate pressure data, and ambient temperature data.

3. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The process of generating a differential detection region segmentation scheme is as follows: By fitting the optical reference surface of the lens surface with laser interferometry phase data, a three-dimensional Cartesian coordinate reference with the geometric center of the lens as the origin is constructed. The bonded surface is divided radially into three continuous regions: a central circle, a middle ring, and an edge ring. Retrieve the spatial distribution statistics of abnormal crosslinking density and abnormal phase coherence in historical production batches, and calculate the probability of anomalies in each region; By combining the bonding accuracy index and the probability of anomalies in each region, the density of detection points and the detection interval in the corresponding region are determined, and a differentiated detection area division scheme is generated.

4. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The steps to obtain multiple batches of spatiotemporal alignment detection sample sets are as follows: Based on the differentiated detection area division scheme, multiple independent detection tasks are generated, and the detection point coordinates, detection time, sampling rate and trigger delay parameters of each task are configured. The original data sources are continuously tested according to the preset detection sequence to obtain the corresponding detection information. Based on the detection time of each detection point, the four types of detection data at the same spatial location are packaged into independent data slices, and all data slices are time stamp verified and spatial coordinate matched to remove invalid slices. The verified data slices are arranged in the detection order to form multiple batches of microsecond-level spatiotemporally aligned detection sample sets.

5. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 4, characterized in that, The process of separating the infrasound vibration time sequence signal is as follows: The dual sampling rate pressure data corresponding to each detection point is extracted from the detection sample set, and low-frequency macroscopic pressure trend signal is obtained by low-pass filtering. The low-frequency macroscopic pressure trend signal is subtracted from the original dual-sampling-rate pressure data to obtain the high-frequency pressure residual component. Then, the high-frequency residual component is bandpass filtered to retain the signal components within the characteristic frequency range of the adhesive crosslinking reaction. Using the ambient noise collected during the no-load operation of the equipment as a reference signal, adaptive noise cancellation is performed on the characteristic frequency signal to finally obtain the infrasonic vibration time sequence signal corresponding to the chemical bond vibration of the cross-linking reaction of the adhesive molecules.

6. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 4, characterized in that, The specific steps for calculating the phase coherence eigenvalues ​​of the sliding window are as follows: The laser interference phase data corresponding to each detection point is extracted from the detection sample set. After polynomial fitting detrending processing, the macroscopic phase change caused by the overall lens displacement and ambient temperature drift is removed, and the micro-phase drift component caused by the slight change in the refractive index of the adhesive layer is obtained. The micro-phase drift component and the infrasonic vibration time sequence signal are divided into multiple sliding time windows of equal length, with the sliding step size not exceeding one-fifth of the window length; Perform Hilbert transform on the two signals in each window to obtain analytic signals, calculate the cross-correlation function of the two analytic signals, and determine the phase difference based on the peak position of the cross-correlation function; The phase coherence coefficient of each window is calculated based on the phase difference. Then, a weighting coefficient is set according to the time distance between the window and the detection time. The coherence coefficients of all windows are weighted and averaged to obtain the sliding window phase coherence characteristic value of the detection point.

7. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The process of inverting the two-dimensional distribution matrix of the crosslinking density of the colloidal layer molecules is as follows: Beforehand, the phase coherence characteristic value and the crosslinking density of the adhesive layer molecules were established by calibrating the standard adhesive sample through phase coherence test and thermal analysis, and the initial crosslinking density value of each detection point was obtained according to the mapping relationship. An adaptive noise template is constructed based on historical qualified batch data to correct the noise in the initial crosslinking density value; Spatial interpolation is performed on the corrected crosslinking density values ​​to generate a two-dimensional distribution matrix of crosslinking density of the adhesive layer molecules containing radial distribution information.

8. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 7, characterized in that, The process of obtaining the true adhesive layer thickness is as follows: Based on the corrected two-dimensional crosslinking density distribution matrix, the crosslinking density gradient value between each detection point and its neighboring detection points is calculated; The actual refractive index of the adhesive at the laser interferometric detection wavelength was determined in advance under different crosslinking density gradients and different ambient temperatures, and a nonlinear correspondence between crosslinking density gradient, ambient temperature and refractive index change was established. Based on the corresponding relationship, the nonlinear correction value of the refractive index gradient at each detection point is obtained, and the standard refractive index of the adhesive is corrected point by point to obtain the actual refractive index at each detection point. The initial true adhesive layer thickness is calculated by combining the optical path difference obtained from the laser interference phase data with the actual refractive index of each detection point. Then, outliers in the initial thickness are removed by iterative weighted least squares method to obtain the final true adhesive layer thickness at each detection point.

9. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The steps to obtain the overall uniformity evaluation value of the adhesive layer are as follows: The actual adhesive layer thickness values ​​of all test points are divided into multiple continuous equal-width intervals, and the crosslinking density values ​​of all test points are divided into the same number of continuous equal-width intervals; The number of detection points in each thickness interval and cross-linking density interval is counted, and the combined probability value of each combination is obtained by dividing the counted number by the total number of detection points. Arrange all joint probability values ​​in interval order to construct a two-dimensional joint probability distribution matrix of the true thickness of the adhesive layer and the molecular crosslinking density; The two-dimensional information entropy value of the matrix is ​​calculated and normalized to obtain the overall uniformity evaluation value of the adhesive layer.

10. The method for detecting the thickness and uniformity of the adhesive layer during lens bonding and centering according to claim 1, characterized in that, The process for generating a comprehensive judgment value for adhesive quality is as follows: The average adhesive layer thickness and average crosslinking density of the three regions—the central circle, the middle ring, and the edge ring—were calculated respectively, and the thickness deviation and crosslinking density deviation of each region were obtained by comparing them with the preset standard values. Multiply the thickness deviation value of each region by the crosslinking density deviation value to obtain the comprehensive deviation value of each region; The weight coefficients of each region are determined by the analytic hierarchy process, and the weighted summation of the comprehensive deviation values ​​of each region is used to obtain the overall adhesive layer thickness assessment value. Weighting coefficients are assigned to the thickness assessment value and the overall uniformity assessment value respectively, and then the overall judgment value of the bonding quality is obtained by weighted summation.

Images