A method for detecting OLED screen defects based on optical-magnetic multimodal fusion

Abstract

The present invention relates to the field of defect detection technology, and in particular to an OLED screen defect detection method based on optical-magnetic multimodal fusion, comprising the following steps: S1: synchronous excitation and data acquisition based on pulsed current drive; S2: cross-modal spatial calibration and noise suppression; S3: polarization-magnetic field coupling feature extraction; S4: dynamic correlation decision-making and closed-loop feedback, including realizing defect classification determination through correlation analysis of multimodal data, and verifying the authenticity of defects through current overload testing. This method effectively solves the problems of ambient light interference, missed detection, high cost, low resolution, etc. existing in the prior art, has significant beneficial effects, can improve detection accuracy, efficiency and reliability, meet the needs of large-scale production lines, and can efficiently locate and reduce errors and redundant steps in the detection process.

Description

Technical Field

[0001] The present invention relates to the field of defect detection technology, and in particular to an OLED screen defect detection method based on optical-magnetic multimodal fusion. Background Art

[0002] As a new display technology, OLED screen has the advantages of self-luminescence, wide viewing angle, high contrast, and low power consumption, and has been widely used in smart phones, tablet computers, TVs, etc. However, the manufacturing process of OLED screen is complicated and prone to various defects, such as scratches, dirt, uneven current distribution, micro cracks in the organic layer, etc. These defects will affect the display effect and service life of the screen. Therefore, it is of great significance to accurately detect OLED screen defects.

[0003] Existing OLED defect detection technologies mainly include optical detection technology and magnetic detection technology. Optical detection technologies such as polarized light imaging and multi-spectral analysis can identify surface morphological defects such as scratches and dirt through the difference method of image template matching, but they are less sensitive to internal structural abnormalities and are easily disturbed by ambient light. It is difficult to distinguish between surface dirt and internal material defects, resulting in a high missed detection rate. Magnetic detection technologies such as scanning superconducting quantum interference microscope (SSM) can identify internal current abnormalities by measuring the changes in the magnetic field generated by the OLED current, but there are problems such as limited spatial resolution and high equipment costs, which make it difficult to adapt to the needs of large-scale production lines.

[0004] In addition, existing technologies attempt to combine multiple sensors, such as optics + electricity, but have not effectively integrated the complementary advantages of optics and magnetism. For example, although EDMR (electrically detected magnetic resonance) and ODMR (optically detected magnetic resonance) can simultaneously measure current and luminescence response, they rely on microwave resonators and complex signal processing, making it difficult to achieve fast and high-precision defect location. Traditional multimodal methods are mostly serial detection, resulting in lengthy detection processes and poor data coordination.

[0005] In summary, the existing OLED defect detection technology has the following problems: single optical detection technology has low sensitivity to internal structural abnormalities and is easily disturbed by ambient light; magnetic detection technology has limited spatial resolution and high equipment costs; multimodal fusion technology is not fully developed, the detection process is lengthy and the data coordination is poor. Therefore, there is an urgent need for a multimodal detection method that can effectively integrate the advantages of optics and magnetism to improve the accuracy and efficiency of OLED screen defect detection. Summary of the invention

[0006] Based on the above purpose, the present invention provides an OLED screen defect detection method based on optical-magnetic multimodal fusion, which is characterized by comprising the following steps:

[0007] S1: Synchronous excitation and data acquisition based on pulsed current drive, including transient current excitation of the OLED screen, covering the transient response frequency range of 1-10MHz, and detecting the strain changes on the screen surface and the transmittance changes inside the screen by building dual-channel optical imaging, and detecting the magnetic field gradient changes inside the screen through the magnetic detection system;

[0008] S2: Cross-modal spatial calibration and noise suppression, including extracting high-gradient areas in the magnetic current density map through a dynamic mask generation algorithm and focusing on the defect area; removing noise including ambient light and screen self-luminescence through optical noise filtering, and retaining optical signals related to defects;

[0009] S3: Polarization-magnetic field coupling feature extraction, including wavelet transform of polarization interference fringes to quantify crack depth and distinguish surface scratches from internal cracks; and distinguish surface dirt from internal cracks through time derivative and magnetic field phase lag;

[0010] S4: Dynamic correlation decision-making and closed-loop feedback, including the classification of defects through correlation analysis of multimodal data and the verification of the authenticity of defects through current overload testing.

[0011] Preferably, the OLED screen is subjected to transient current excitation, which specifically includes the following contents:

[0012] Use FPGA to control the current source to generate a square wave pulse sequence with an adjustable amplitude of 0-10mA and a programmable pulse width of 10-100μs. The pulse interval is 200μs.

[0013] Define the pulse rising edge time to be less than or equal to 1μs, and the falling edge time to be less than or equal to 1μs, so that the current transient response frequency covers 1-10MHz;

[0014] The output interface uses a triaxial cable with an inductance of <10nH, and the terminal contact impedance is matched to the equivalent impedance of the OLED screen.

[0015] Preferably, by constructing dual-channel optical imaging, detecting the strain change on the screen surface and the transmittance change inside the screen specifically includes the following contents:

[0016] Channel 1 is polarization interference imaging, and the parameter design includes:

[0017] Laser source: 532nm DPSS laser, output power P=20mW, linear polarization degree>99%;

[0018] Optical path design: incident angle θ = 45°, after being split by Wollaston prism, the optical path difference between the two beams ΔL = λ / 4 = 133nm, and the interference fringe period Λ = λ / 2sinθ = 752nm;

[0019] Image acquisition: A CMOS with a global shutter frame rate of 5000fps, a resolution of 4096×3000, and a pixel size of 2.4μm was used to obtain interference fringes. I ( x );

[0020] Channel 2 is infrared transmission imaging, and the parameter design includes:

[0021] Light source: 850nm LED ring array, the inner diameter of the ring array is 30mm, the outer diameter is 50mm, and the luminous intensity is I 0 = 1500 cd / m 2 , the pulse width is strictly synchronized with the current pulse;

[0022] Transmittance calculation: ;

[0023] For location The transmittance at ;

[0024] For location The detected light intensity at

[0025] is the initial light intensity before incident;

[0026] is the absorption coefficient of the material;

[0027] is the thickness of the material through which the light passes;

[0028] CMOS exposure time is set to t exp =10 μs , start exposure 5μs after the trailing edge of the current pulse.

[0029] Preferably, detecting the magnetic field gradient change inside the screen by the magnetic detection system specifically includes the following contents:

[0030] Sensor array: The sensor array consists of 64×64 aluminum nitride piezoelectric resonators, with a total of 4096 resonators. The spacing between two adjacent piezoelectric resonators in the x and y directions is 5 microns. The resonant frequency of each piezoelectric resonator is 15 MHz. The quality factor of the piezoelectric resonator is greater than 2000. The change in the resonant frequency is proportional to the magnetic field strength. The distribution of the magnetic field B is determined by measuring the change in the resonant frequency.

[0031] Magnetic field gradient measurement: The magnetic field gradient is obtained by the following formula Distribution map:

[0032] ;

[0033] in,

[0034] : The offset of the resonant frequency;

[0035] : Magnetic field gradient, which indicates the rate of change of magnetic field intensity in space;

[0036] k : sensitivity coefficient, which indicates the resonant frequency shift caused by each unit magnetic field gradient;

[0037] Signal processing: A dual lock-in amplifier is used. The minimum phase difference that the lock-in amplifier can distinguish is 0.01°. The output signal of the lock-in amplifier is: ,in is the amplitude of the signal, represents the intensity of the resonator vibration, is the magnetic field phase, Represents the phase difference between the resonator vibration and the reference signal.

[0038] Preferably, cross-modal spatial calibration and noise suppression comprises the following steps:

[0039] S2.1: Cross-modal spatial calibration and noise suppression:

[0040] S2.1.1: Using Poisson's equation, solve the inverse problem to obtain the magnetic current density map J ( x,y ), the Poisson equation is as follows:

[0041] ;

[0042] in, Represents magnetic field B The Laplace operator describes the second-order rate of change of the magnetic field in space; Indicates current density J The curl multiplied by the vacuum permeability m 0, describes the effect of the rotational characteristics of current density in space on the magnetic field, It is a vector differential operator, which means the curl operation on a vector field;

[0043] S2.1.2: Calculate the gradient amplitude using the following formula:

[0044] ;

[0045] Among them, the current density diagram J ( x,y) is a two-dimensional scalar function, which means that at each location ( x,y ) and its gradient is a vector field representing the current density in x and y The rate of change in direction, the gradient amplitude is the gradient Length;

[0046] S2.1.3: Adaptive threshold segmentation of gradient amplitude is performed to extract >15mA / m m 2 area as defect area;

[0047] S2.1.4: Through the affine transformation matrix M affine Convert the defect area coordinates to the CMOS pixel coordinate system and control the optical zoom lens to reduce the field of view by 20 times;

[0048] S2.2: optical noise filtering;

[0049] S2.2.1: Perform pixel-level difference on two consecutive infrared images:

[0050] ;

[0051] in, To express the transmittance The change in time between two frames, and Respectively represent Frame and Frame at position The transmittance at ;

[0052] S2.2.2: Establishing the noise discrimination matrix N ( x , y ), N ( x , y ) is a binary matrix, which is expressed as follows:

[0053] ;

[0054] in, >2% means the pixel value changes by more than 2%;

[0055] S2.2.3: The noise discriminant matrix N ( x , y ) are marked as ambient noise and are removed in subsequent processing.

[0056] Preferably, quantifying the crack depth specifically comprises the following steps:

[0057] S3.1.1: Polarization interference fringes I ( x ) performs 5-layer discrete wavelet transform;

[0058] S3.1.2: Calculate the 4th layer detail coefficients d 4, where in the wavelet decomposition, the 4th layer detail coefficient d 4 represents the detail information at the 4th level of decomposition, which reflects the high-frequency components in the stripe image. The variance is calculated as:

[0059] ;

[0060] N is the 4th layer detail coefficient d The sample size is 4, that is d The number of data points contained in 4, m d yes d The mean of 4;

[0061] S3.1.3: Establish crack depth through calibration experiments h crack The relationship is:

[0062] ;

[0063] in, is the 4th layer detail coefficient d Variance of 4.

[0064] Preferably, distinguishing surface dirt from internal cracks comprises the following steps:

[0065] S3.2.1: Calculate time derivatives, time derivatives It represents the rate of change of transmittance T over time, and the calculation formula is:

[0066] ;

[0067] in, and denote the transmittance of the n+1th frame and the nth frame respectively, is the time interval =200 m s ;

[0068] S3.2.2: Magnetic field phase lag It represents the change of magnetic field phase over time, and the calculation formula is:

[0069] ;

[0070] in and Respectively represent the magnetic field phases of the n+1th frame and the nth frame, f 0 is the reference frequency;

[0071] S3.2.3: Correlation coefficient R Used to measure the time derivative and magnetic field phase lag The correlation between them is calculated as:

[0072] ;

[0073] The judgment rules are as follows:

[0074] like , then it is judged as surface dirty;

[0075] like , it is judged as an internal crack.

[0076] Preferably, the association analysis of the multimodal data includes a two-level decision tree implementation, specifically including the following:

[0077] First-level decision parameter: Surface damage judgment threshold is <5%, that is, when When <5%, it is judged as a scratch;

[0078] Secondary decision parameters: interface delamination determination threshold;

[0079] Condition 1: The local current peak is greater than 1.2 times the average current density of the same type of defect-free OLED screen at the rated driving current;

[0080] Condition 2: Crack Depth h crack <50 nm ;

[0081] When both conditions 1 and 2 are met, delamination is determined.

[0082] Preferably, when the secondary decision parameters output conflicting results, that is, when condition 1 and condition 2 are not satisfied at the same time, a closed-loop feedback verification is performed, including the following steps:

[0083] S4.1: Apply 1.2 to OLED screen I nominal The overload current, I nominal It is the average current density of the same type of OLED screen without defects at the rated driving current; the duration is 3 pulse cycles;

[0084] S4.2: Transmittance slope calculation:

[0085] During the 2nd to 3rd pulse cycle of current overload, t exp = 1 μs Ultra-short exposure time to collect infrared transmission image sequences T 1 ,T 2 ,...,T n , sampling rate f s =1 MHz ;

[0086] The formula is:

[0087] ;

[0088] t 0: Current overload start time, marked by FPGA hardware;

[0089] =10 μs ;

[0090] T ( t ): average transmittance of defect area;

[0091] S4.3: Magnetic field distortion detection:

[0092] During current overload, f s =10 MHz Sampling rate records the magnetic field time domain signal output by the piezoelectric resonator array B ( t );

[0093] right B ( t ) performs windowed FFT, where the windowing is specifically a Blackman-Harris window with a frequency resolution of Δ f =1 kHz ; Extract the fundamental amplitude A1 and the 3rd harmonic amplitude A3, where the fundamental f 0 =1.2 I nominal Frequency, 3rd harmonic is 3 f 0。

[0094] Preferably, the logic judgment rules for closed-loop feedback verification are as follows:

[0095] when kT >10 4 / s and when A3 / A1<0.1, the surface is judged to be dirty;

[0096] when k T <10 4 / s and A3 / A1>0.1, it is judged as internal crack;

[0097] when k T >10 4 / s and A3 / A1>0.1, it is judged that surface dirt and internal cracks coexist;

[0098] Update defect classification results and mark the final determination type of the disputed area;

[0099] If there is still uncertainty, a third-level alarm is triggered and manual re-inspection is performed, while the overload test data is recorded.

[0100] Beneficial effects of the present invention: 1. The present invention significantly improves the accuracy of detection by integrating optical and magnetic technologies and combining their complementary advantages. Optical detection can effectively identify surface defects (such as scratches, dirt, etc.), while magnetic detection can identify internal structural abnormalities (such as uneven current distribution). The complementarity of the two technologies enables more comprehensive and accurate detection of overall defects of OLED screens.

[0101] 2. Optical detection is easily affected by ambient light in traditional methods, resulting in unstable defect detection results. By combining with magnetic technology, magnetic field detection is not affected by ambient light, so it can effectively compensate for the shortcomings of optical detection, solve the problem of ambient light interference, and thus improve detection stability.

[0102] 3. Traditional optical detection methods are prone to missed detection due to their low sensitivity to internal defects. Magnetic detection technology can detect abnormalities in internal current distribution, ensuring that the risk of missed detection is reduced. In addition, the combination of the two technologies can provide more accurate detection results, effectively avoid the occurrence of false detection, and further improve the reliability of detection.

[0103] 4. Magnetic detection technology can identify tiny internal defects by improving spatial resolution, while avoiding the high equipment cost and limited resolution problems of traditional technology. Through the fusion of optics and magnetism, high-precision defect detection can be performed at a lower cost to meet the needs of large-scale production lines.

[0104] 5. The present invention uses multimodal fusion technology to simultaneously obtain optical images and magnetic field data in the same detection process, avoiding the lengthy serial processing process in traditional multimodal detection. In this way, the detection process is greatly simplified, defects can be located quickly and accurately, and the overall detection efficiency is improved.

[0105] 6. Traditional multimodal technology usually faces the problem of poor data coordination, that is, it is difficult to effectively combine the detection data of different modes, resulting in insufficient and inaccurate information. The present invention ensures the efficient coordination of optical and magnetic data by designing an effective optical-magnetic data fusion algorithm, making the detection results more comprehensive and consistent.

[0106] 7. Due to the integration of the advantages of optics and magnetism, the detection method of the present invention can be flexibly adjusted according to the specific needs of the OLED screen, and is suitable for the detection of screens of different types and specifications, and has strong adaptability. This flexibility enables the technology to operate efficiently in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0108] Figure 1 is a flow chart of the steps of the method of the present invention;

[0109] Figure 2 is a flow chart of the steps of method S3.1 of the present invention;

[0110] Figure 3 This is a flow chart of the steps of method S3.2 of the present invention. DETAILED DESCRIPTION

[0111] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0112] See also Figure 1-Figure 3The embodiment of the present invention provides an OLED screen defect detection method based on optical-magnetic multimodal fusion, which provides an efficient and accurate defect detection solution through fine step arrangement and technical integration. Each step is closely linked and complements each other, from signal acquisition to defect extraction, and then to the final decision feedback, which not only improves the accuracy and efficiency of OLED screen defect detection, but also provides strong support for subsequent quality control and product optimization.

[0113] Specifically, in step S1, the OLED screen is transiently excited by pulsed current driving, covering a frequency range of 1-10MHz. The function of this stimulation signal is to stimulate strain and magnetic field gradient changes inside and on the surface of the screen. Through a dual-channel optical imaging system, the changes in screen surface strain and internal transmittance are detected respectively to collect optical signals. At the same time, the magnetic detection system provides magnetic signals for subsequent analysis by detecting changes in magnetic field gradients inside the OLED screen. This step lays the foundation for subsequent multimodal data fusion.

[0114] Furthermore, step S2 processes the magnetic current density map through a dynamic mask generation algorithm to extract high-gradient areas, which are usually related to the defect locations of the screen. This processing will focus on possible defect areas and improve the accuracy of defect identification. At the same time, through the noise filtering technology in the optical signal, the interference caused by ambient light and OLED self-luminescence is eliminated, and the optical signal related to the defect is retained. This step effectively improves the signal-to-noise ratio of subsequent analysis and provides clear data for accurate defect extraction and judgment.

[0115] Furthermore, in step S3, the polarization interference fringes are analyzed by wavelet transform technology to quantify the depth of the screen cracks and accurately distinguish between surface scratches and internal cracks. The different characteristics of surface dirt and internal cracks can be further distinguished by analyzing the phase lag of the magnetic field. This process can accurately extract the characteristic information related to the defect, providing a key basis for the analysis of the type and degree of the defect.

[0116] Finally, multimodal data is used for correlation analysis to achieve defect classification. The authenticity of the defect can be further confirmed by verifying the current overload test. Based on this data, the system can dynamically adjust the detection strategy and perform closed-loop feedback to optimize the accuracy and efficiency of defect identification.

[0117] In one possible implementation, an FPGA (field programmable gate array) is used as the control core to generate a square wave pulse sequence with an adjustable amplitude of 0-10mA and a programmable pulse width of 10-100μs through programming. The pulse interval is 200μs, and the rising and falling edge times of the pulse are less than or equal to 1μs. The high-precision control of these pulse signals ensures that the current transient response frequency can cover the range of 1-10MHz. This precise pulse signal generates transient excitation for the current response of the OLED screen, driving the rapid change of the current density in the screen, causing it to generate magnetic field and strain signals related to the defect, thereby providing diagnostic information of the defect.

[0118] Through the precise pulse sequence controlled by FPGA, the amplitude and pulse width of the current excitation are adjustable, and the rising and falling edge times are extremely short, ensuring that the high-frequency current excitation signal can cover the frequency range of 1-10MHz, which is crucial for the transient response of OLED screen materials. High-frequency pulse signals can cause tiny physical changes inside and on the surface of the material, enhance the response characteristics of the defective area, and make defect detection more sensitive and accurate.

[0119] Furthermore, the output interface uses a triaxial cable with an inductance of less than 10nH to ensure high-quality signal transmission and avoid signal attenuation and distortion during transmission. The triaxial cable design effectively suppresses electromagnetic interference and ensures that the current excitation signal is stably transmitted to the OLED screen. The end of the cable is impedance matched at the connection with the OLED screen to ensure that the signal matches the equivalent impedance of the OLED screen to avoid signal reflection or loss caused by impedance mismatch. This impedance matching design optimizes the transmission efficiency of the current excitation signal and maximizes the accurate response of the OLED screen to the pulse signal.

[0120] The triaxial cable adopts a low inductance design to reduce signal loss and interference during transmission, ensuring high-quality transmission of the excitation signal. This design not only enhances the integrity of the signal, but also avoids misjudgment or defect identification errors caused by signal distortion. The impedance matching design of the signal interface further optimizes the transmission efficiency of the current signal and improves the overall performance of the system.

[0121] The transient current excitation of the pulse signal can stimulate tiny strain and magnetic field changes in the OLED screen. These changes will be captured and analyzed in subsequent optical and magnetic inspections, becoming an important basis for defect location and identification. The precise control of the current excitation signal and the high-quality signal transmission path lay the foundation for subsequent defect detection.

[0122] By applying transient current to the OLED screen, its current response, strain response and magnetic field changes can be triggered. These physical signals can be efficiently captured by subsequent optical and magnetic detection equipment for accurate analysis. This excitation method can significantly improve the accuracy of defect recognition, especially for the detection of tiny defects or deep defects, and improve the sensitivity of the detection system.

[0123] In a possible implementation, channel 1 is polarization interference imaging, wherein the laser source is selected as follows: a DPSS laser with a wavelength of 532nm, an output power of 20mW, and a linear polarization degree of more than 99%. This ensures the high beam quality of the laser light source, which can accurately interact with the surface of the OLED screen to produce interference fringes that can be used for defect detection.

[0124] Light path design: The incident angle is set to 45°. After being split by the Wollaston prism, the two beams are separated by an optical path difference of ΔL=λ / 4=133nm, making the period of the interference fringes Λ=λ / 2sinθ=752nm. This design can produce high-resolution interference fringes and accurately reflect the slight deformation or strain changes on the surface of the OLED screen.

[0125] Image acquisition: A global shutter CMOS camera was used with a frame rate of 5000fps, 4096×3000, and a pixel size of 2.4μm to capture interference fringes with high precision. I ( x ) can display the tiny strain or deformation caused by defects on the screen surface, reflecting the unevenness of the screen surface structure.

[0126] Furthermore, channel 2 is infrared transmission imaging, in which the light source design is: select an LED ring array light source with a wavelength of 850nm, the inner diameter of the ring array is 30mm, the outer diameter is 50mm, and the luminous intensity is I 0 is 1500cd / m², and the current pulse of the ring array is strictly synchronized with the pulse width of the light source. This design ensures uniform illumination of the light source, can effectively penetrate the OLED screen material, and complements the optical interference imaging channel.

[0127] Transmittance calculation: According to the absorption characteristics of the screen surface and interior, the transmittance By formula Calculate; where For location The actual detected light intensity at is the initial light intensity before incident; is the absorption coefficient of the material; The thickness of the material through which light passes; this transmittance calculation can reveal defects or uneven areas within the material, especially for subtle damage or invisible defects inside the screen.

[0128] The exposure time is set to t exp =10 μs , start exposure 5μs after the trailing edge of the current pulse.

[0129] Exposure time setting: The exposure time of the CMOS camera is set to 10μs, and the exposure is started at 5μs after the trailing edge of the current pulse to ensure that the transmitted light image synchronized with the excitation signal is captured. This high time precision control can synchronously track the transient response of the material and improve the detection sensitivity.

[0130] Polarization interference imaging and infrared transmission imaging are used together to comprehensively detect defects in OLED screens from two aspects: surface strain and internal transmittance. Polarization interference imaging can capture surface deformation or strain caused by defects, while infrared transmission imaging can provide further information about internal defects, especially tiny cracks, bubbles or poor interlayer bonding in the screen material.

[0131] The two channels can complement each other. Through the comprehensive analysis of interference fringe images and transmittance images, the accuracy and reliability of defect detection can be greatly improved. For example, surface interference fringes may reveal areas of stress concentration, while infrared transmission images may reveal hidden defects inside the screen. The combination of the two can more comprehensively evaluate the overall quality of the screen.

[0132] By using a 5000fps high-speed CMOS camera, combined with 10 μs With precise exposure time and synchronous pulse signals, the system can capture and process the dynamic changes of the OLED screen in real time, especially the transient response of the screen under the action of the excitation signal. This high timeliness design makes the detection system more sensitive to tiny defects and rapid changes, significantly improving the sensitivity and accuracy of detection.

[0133] In one possible implementation, by combining an aluminum nitride piezoelectric resonator array, magnetic field gradient calculation, and precise signal processing, it is possible to effectively detect tiny defects inside the OLED screen and abnormal magnetic field distribution, thereby improving the accuracy and reliability of screen defect detection, and providing a powerful tool for quality control of the OLED screen. Specifically, the sensor array consists of 64×64 aluminum nitride (AlN) piezoelectric resonators, a total of 4096 piezoelectric resonators. Each piezoelectric resonator has a resonant frequency of 15MHz and a high quality factor (greater than 2000), which makes the resonator extremely sensitive to tiny frequency changes. The spacing between adjacent piezoelectric resonators in the x-direction and y-direction is 5 microns, which provides precise spatial resolution for detailed detection of the screen. The resonant frequency change of each resonator is proportional to the magnetic field strength at the location. By measuring the frequency shift of these resonators, the magnetic field distribution inside the screen can be inferred, and the magnetic field gradient therein can be further analyzed. Because the resonator is very sensitive to changes in the magnetic field, the system can accurately capture subtle changes in the magnetic field gradient.

[0134] Magnetic field gradient By shifting the resonant frequency To calculate the frequency offset The relationship with the magnetic field gradient is given by the formula (k=3.2×10^3 Hz / (T / m)), where k is a sensitivity coefficient. The sensitivity coefficient k is selected so that each unit change in magnetic field gradient can cause an accurate frequency offset. Through this formula, the system can convert the frequency offset into a magnetic field gradient, thereby obtaining a distribution map of the magnetic field changes inside the screen.

[0135] By detecting the frequency shift of each sensor array, the system can draw a distribution map of the magnetic field gradient. This distribution map provides valuable information for detecting potential defects inside the OLED screen, especially in the inner layer of the screen or hidden tiny cracks and defect areas, which can reflect the abnormal magnetic field distribution.

[0136] Furthermore, the signal processing part uses a dual phase-locked amplifier. The phase-locked amplifier can distinguish extremely small phase differences (the minimum is 0.01°), which greatly improves the sensitivity of the signal. The output signal of the phase-locked amplifier is ,in is the amplitude of the signal, represents the intensity of the resonator vibration, is the magnetic field phase, Represents the phase difference between the resonator vibration and the reference signal. By analyzing the change in phase difference, the change in magnetic field can be further accurately calculated, thereby obtaining the change in magnetic field gradient inside the screen. The signal output by the phase-locked amplifier can accurately interpret the frequency change of each resonator and further obtain detailed information on the magnetic field distribution. This step not only improves the reliability and accuracy of the signal, but also can accurately reveal the magnetic field changes caused by defects inside the screen in complex OLED structures.

[0137] In one possible implementation, the steps of cross-modal spatial calibration and noise suppression are to accurately identify and remove noise by combining magnetic and optical information to improve detection accuracy. The relationship between magnetic field and current density is calculated using the Poisson equation, and then the gradient amplitude is used to accurately identify the defect area, and finally the defect area is further confirmed and processed in combination with the optical image.

[0138] Specifically, in step S2.1.1, the inverse problem is solved using the Poisson equation to obtain the current density map J(x, y). By solving the Poisson equation, the relationship between the magnetic field and the current density is accurately reflected. Each point in the current density map corresponds to the current intensity distribution at a location. This information is crucial for defect detection of OLED screens.

[0139] In step S2.1.2, by calculating the gradient amplitude of the current density map J(x, y), the rate of change at each position is further analyzed to identify areas with large current density changes, which are usually associated with physical defects of the OLED screen (such as breakage, short circuit, etc.). Areas with gradient amplitudes greater than a certain threshold (here 15mA / μm²) are extracted as potential defect areas.

[0140] Step S2.1.4 converts the coordinates of the defective area of ​​the magnetic image into the coordinate system of the optical image through an affine transformation matrix, so that the optical imaging can accurately correspond to the magnetic information. Reducing the field of view helps to improve the image resolution, thereby making the defective area clearer.

[0141] Step S2.2 mainly processes the noise in the optical image. First, the transmittance change of two consecutive infrared images is calculated by the pixel-level difference method to ensure that the change in the image is due to defects rather than environmental noise. Then, the noise discrimination matrix N(x, y) distinguishes noise from actual defect signals by setting a change threshold (such as 2%). By marking and removing the noise area, misjudgment can be avoided.

[0142] The noise area is eliminated through step S2.2.3, and the noise discrimination matrix marks the pixels that do not meet the defect conditions as noise. This can effectively reduce the interference of environmental factors (such as lighting changes, noise of the equipment itself, etc.) on the detection results, thereby ensuring the accuracy of the final detection results.

[0143] Through the cross-modal fusion of optics and magnetism, the complementarity of the two signals can be fully utilized to further improve the sensitivity and accuracy of defect detection. The magnetic method can screen out possible defect areas in advance, while the supplement of optical images provides higher spatial resolution. Noise suppression technology can effectively eliminate errors caused by environmental or equipment noise, thereby ensuring the reliability of detection results. This multimodal fusion strategy not only improves the accuracy of detection, but also greatly improves the processing speed, providing strong support for large-scale defect detection in practical applications.

[0144] In one possible implementation, in step S3.1.1, by performing polarization interference fringe image I ( x ) performs 5 layers of discrete wavelet transform, which can extract detailed information of the image from different scales. Discrete wavelet transform (DWT) is an effective signal analysis method that can decompose image signals into multiple frequency bands. Each layer of wavelet transform provides details of different frequency components, and high-frequency components reflect the faster changing parts of the image, which is especially important for the detection of small defects such as cracks. By decomposing the stripe image into 5 layers of wavelet signals, the details of the cracks, especially tiny cracks, can be captured at multiple scales, and more accurate analysis can be performed at different scales.

[0145] Furthermore, in step S3.1.2, calculating the variance of the detail coefficient of the 4th layer is a key step. The detail coefficient of the 4th layer mainly represents the image features at a higher resolution, which can capture more subtle crack information. By calculating the variance of this coefficient, the degree of change of the crack in the image can be quantified. The variance reflects the discreteness of the data, that is, the volatility of the crack area in the image. A high variance means that the crack features in the image are more obvious, and a low variance means that the defects are relatively gentle or crack-free. Calculating the variance helps to evaluate the intensity and performance of the cracks, and provides a quantitative basis for subsequent crack depth analysis. The formula for calculating the variance is:

[0146] ; N is the 4th layer detail coefficient d The sample size is 4, that is d The number of data points contained in 4, m d yes d The mean of 4;

[0147] In step S3.1.3, the relationship between the crack depth and the variance of the fourth layer detail coefficient was established through a calibration experiment. Through this relationship, the depth of the crack can be inferred using the calculated variance value. In the experiment, cracks of different depths were set and their corresponding variance values ​​were measured to obtain a fitting relationship between the variance and the crack depth. Through this relationship, combined with the variance value, accurate quantification of the crack depth can be achieved. The establishment of the formula ensures that the crack information obtained from the image is not limited to the recognition of surface morphology, but can also go deep into the actual depth of the crack. The calculation formula is:

[0148] ;in, is the 4th layer detail coefficient d Variance of 4.

[0149] An efficient crack quantification process was established through wavelet transform, variance calculation and crack depth calibration. Wavelet transform provides detailed information at different scales and can capture crack characteristics more accurately, especially for the detection of small cracks. Variance calculation further quantifies the characteristic intensity of the crack and provides an effective indicator for depth quantification. The depth-variance relationship established through the calibration experiment enables the crack depth to be linked to the volatility of the crack in the image, thereby achieving high-precision depth estimation. This method can improve the accuracy of OLED screen defect detection, especially in the detection of small cracks and depth quantification, and has important application value.

[0150] In one possible implementation, by combining multimodal information of optics and magnetism, the dynamic characteristics of defects are quantified using time derivatives and magnetic field phase lags, greatly improving the accuracy of OLED screen defect detection. First, the time derivative provides a preliminary basis for distinguishing cracks and dirt by capturing changes in transmittance. Then, the introduction of magnetic field phase lag further enhances the ability to distinguish cracks and dirt, because dirt generally affects the transmittance of light but has little effect on the magnetic field, while cracks not only change the transmittance of light, but also change the magnetic field characteristics inside the material. Finally, by calculating the correlation coefficient R, defects can be more accurately classified in a quantitative manner.

[0151] Specifically, in step S3.2.1, the temporal changes of the screen surface are captured by calculating the derivative of the transmittance over time. The transmittance T reflects the degree of light passing through the OLED screen and is usually related to defects on the screen surface (such as dirt and cracks). By calculating the time derivative, the rate of change of the transmittance can be understood, thereby effectively distinguishing different types of defects. The calculation formula of the time derivative is:

[0152] ;in, and denote the transmittance of the n+1th frame and the nth frame respectively, is the time interval =200 μs ; The purpose of this step is to determine the rate of change of transmittance in a short period of time in order to further analyze the type of defects.

[0153] In step S3.2.2, the dynamic changes of the magnetic field are reflected by calculating the change in the magnetic field phase. Magnetic field phase lag refers to the time lag effect of the magnetic field phase, which is closely related to the material properties of the surface and interior of the object. By calculating the change in magnetic field phase lag, the magnetic field characteristics related to defects (such as cracks and dirt) can be identified. The calculation formula for magnetic field phase lag is:

[0154] ;in and Respectively represent the magnetic field phases of the n+1th frame and the nth frame, f 0 is the reference frequency; this step helps analyze the change of the magnetic field over time and further provides information about the nature of the defect (such as surface or internal defects).

[0155] In step S3.2.3, the nature of the defect is determined by calculating the correlation coefficient between the time derivative and the magnetic field phase lag. The correlation coefficient R is used to quantify the correlation between the transmittance change and the magnetic field phase lag, and the formula is as follows:

[0156] ;

[0157] This formula calculates the linear relationship between the time derivative and the magnetic field phase lag. When the R value is low (R < -0.7), it indicates a strong inverse correlation between the two, which is usually a characteristic of surface contamination, because contamination usually causes a strong inverse relationship between the change in transmittance and the change in magnetic field. When the R value is high (R > 0.5), it indicates a strong positive correlation between the two, which is usually a characteristic of internal cracks, because cracks cause changes in the internal structure, resulting in more consistent changes in transmittance and magnetic field phase.

[0158] In one possible implementation, the implementation of the secondary decision tree achieves accurate classification of defect types through two main decision thresholds (primary decision parameters and secondary decision parameters). In the primary decision stage, it is first determined whether it is surface damage (such as scratches), and the change in transmittance is determined by ΔJ. If the scratch condition is met, there is no need to enter the secondary decision; if not, enter the secondary decision step to further determine whether it is an interface delamination defect. This two-level determination method effectively reduces the computational complexity while ensuring high accuracy of the defect type. Through the judgment of current density and crack depth, the secondary decision effectively distinguishes between minor cracks and interface delamination. Interface delamination involves not only surface defects, but also the internal structure and electrical properties of the screen. This determination provides more comprehensive information than optical or magnetic analysis alone through the joint analysis of current changes and crack depth, thereby improving the reliability of detection.

[0159] First-level decision: the surface damage judgment threshold is Δ J <5%. Specifically, firstly, the judgment parameter Δ of the OLED screen surface damage is obtained by calculating the change of the optical signal. J . Δ J It indicates the ratio of the amplitude of transmittance change to the reference value during the detection process. J When it is less than 5%, it means that there are relatively minor scratches or surface damage on the surface of the OLED screen. At this time, the damage type can be preliminarily judged as "scratches" based on the first-level decision parameters.

[0160] After the first-level decision is determined to be non-scratch, the second-level decision step is entered. The determination of interface delamination needs to be judged by two conditions:

[0161] Condition 1: The local current peak is greater than 1.2 times the average current density of a defect-free OLED screen of the same model at the rated driving current.

[0162] This condition is used to determine whether an abnormal increase in local current has occurred. The increase in current peak is usually caused by delamination or separation of the material interface. By comparing with the reference current density of a defect-free OLED screen, if the local current peak exceeds 1.2 times the reference value, it indicates that interface delamination may exist.

[0163] Condition 2: Crack Depth h crack <50 nm.

[0164] Crack depth h crack Used to judge the severity of cracks. When the crack depth is less than 50 nm When the crack is 0.04mm, it indicates that the crack has not caused obvious structural damage and is usually accompanied by interface delamination rather than a simple surface crack.

[0165] When these two conditions are met at the same time (i.e., the local current peak is large and the crack depth is shallow), it can be determined as an interface delamination defect.

[0166] In a possible implementation, when the secondary decision parameters output conflicting results, closed-loop feedback verification is used to further improve the accuracy of defect identification. The specific closed-loop feedback verification steps include applying an overload current to the OLED screen, calculating the slope of the sudden change in transmittance, and detecting magnetic field distortion.

[0167] Specifically, first, an overload current is applied to the OLED screen, which is the rated current. I nominal 1.2 times (i.e. 1.2 I nominal ). I nominal This is the driving current of a defect-free OLED screen of the same model under normal working conditions. The purpose of this step is to simulate extreme working conditions through the action of external overload current to detect whether there are potential defects, especially those that are difficult to show under normal working conditions. The overload current lasts for 3 pulse cycles. The overload current can stimulate defects caused by excessive current, such as delamination of material layers or surface cracks.

[0168] During the second and third pulse cycles of the current overload, an ultra-short exposure time ( t exp = 1 μs ) Collect infrared transmission image sequences of OLED screens ( T 1 ,T 2 ,...,T n ), the sampling rate is 1 MHz This process monitors the rapid change of transmittance and calculates the slope of the sudden change of transmittance. The specific formula is:

[0169] ;

[0170] in, t 0: Current overload start time, marked by FPGA hardware;

[0171] Δ t =10 μs ; is the time difference (10μs), T(t): the average transmittance of the defective area. Under current overload conditions, a sharp change in transmittance usually indicates that there are obvious structural problems in the OLED screen material (such as thermal stress damage and cracks caused by current overload). Through this ultra-short time scale transmittance change, the rapid response under current impact can be detected, providing a more accurate optical signal for subsequent defect determination.

[0172] Recording the magnetic field time domain signal of an OLED screen using a piezoelectric resonator array during current overload B ( t ), and collect data at a high sampling rate of 10MHz. B ( t ) signal is subjected to windowed FFT (Fast Fourier Transform), using Blackman-Harris window with a frequency resolution of Δf of 1kHz. FFT analysis can extract the frequency domain characteristics of the signal, especially the fundamental amplitude A1 and the third harmonic amplitude A3, where the fundamental frequency f 0 =1.2 I nominal , the third harmonic is 3 f 0 The distortion of magnetic field signals is caused by changes in the electromagnetic properties of the screen material when the current inside the OLED screen is overloaded. In the defective area of ​​the OLED screen, the current overload will cause local changes in the magnetic field, resulting in frequency shifts and amplitude changes, especially changes in harmonic signals. By detecting these frequency domain signals, it is possible to further determine whether there are physical defects caused by current overload, such as material delamination or structural cracks.

[0173] In a possible implementation, by analyzing the transmittance mutation slope (kT) and the third harmonic amplitude ratio (A3 / A1) of the magnetic field signal and combining the data features of the two, more accurate defect determination can be achieved.

[0174] Specifically, when the transmittance slope suddenly changes k T >10 4 / s and the third harmonic amplitude ratio A3 / A1<0.1, the system will judge that the surface is dirty. At this time, the sharp change in transmittance reflects the sudden change in the optical properties of the screen surface caused by external pollutants (such as dust or oil stains), while the third harmonic amplitude ratio of the magnetic field signal is small, which means that the internal electromagnetic properties have not changed significantly, indicating that the defect mainly comes from external pollutants.

[0175] When the slope of the transmittance suddenly changes k T <10 4 / s and A3 / A1>0.1, the system will judge it as an internal crack. In this case, the transmittance changes relatively slowly, indicating that the defect is located in the internal layer of the OLED screen, and the third harmonic amplitude ratio of the magnetic field signal increases, indicating that the internal electromagnetic characteristics change significantly, indicating that there may be cracks, material layer peeling or other structural defects.

[0176] When the slope of the transmittance suddenly changes k T >10 4 / s and A3 / A1>0.1, it is judged that surface dirt and internal cracks coexist. In this case, the sudden change in transmittance reflects the surface contamination, while the increase in the third harmonic amplitude ratio indicates the existence of internal defects. Therefore, it is judged that the two coexist and the defects are more complicated.

[0177] According to the above judgment rules, the system will update the defect classification results in a timely manner and make a final judgment on the difficult areas. For areas where there is uncertainty in multiple judgment rules (for example, some areas may meet multiple judgment criteria at the same time), the system will mark them as disputed areas and provide reference for manual re-inspection.

[0178] If the system still cannot determine the defect type in certain areas or there is great uncertainty in the judgment, the system will trigger a level 3 alarm. This alarm mechanism can remind humans to conduct further re-inspections and help humans make accurate judgments when making decisions by recording the data of overload tests.

[0179] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0180] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting OLED screen defects based on optical-magnetic multimodal fusion, characterized in that: The steps include: S1: Synchronous excitation and data acquisition based on pulsed current drive, including transient current excitation of the OLED screen, covering the transient response frequency range of 1-10MHz, and detecting the strain changes on the screen surface and the transmittance changes inside the screen by building dual-channel optical imaging, and detecting the magnetic field gradient changes inside the screen through the magnetic detection system; S2: Cross-modal spatial calibration and noise suppression, including extracting high-gradient areas in the magnetic current density map through a dynamic mask generation algorithm and focusing on the defect area; removing noise including ambient light and screen self-luminescence through optical noise filtering, and retaining optical signals related to defects; S3: Polarization-magnetic field coupling feature extraction, including wavelet transform of polarization interference fringes to quantify crack depth and distinguish surface scratches from internal cracks; and distinguish surface dirt from internal cracks through time derivative and magnetic field phase lag; S4: Dynamic correlation decision-making and closed-loop feedback, including the classification of defects through correlation analysis of multimodal data and the verification of the authenticity of defects through current overload testing.

2. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 1 is characterized in that: The OLED screen is excited by transient current, including the following: Use FPGA to control the current source to generate a square wave pulse sequence with an adjustable amplitude of 0-10mA and a programmable pulse width of 10-100μs. The pulse interval is 200μs. Define the pulse rising edge time to be less than or equal to 1μs, and the falling edge time to be less than or equal to 1μs, so that the current transient response frequency covers 1-10MHz; The output interface uses a triaxial cable with an inductance of <10nH, and the terminal contact impedance is matched to the equivalent impedance of the OLED screen.

3. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 2 is characterized in that: By constructing dual-channel optical imaging, the strain changes on the screen surface and the transmittance changes inside the screen are detected, including the following: Channel 1 is polarization interference imaging, and the parameter design includes: Laser source: 532nm DPSS laser, output power P=20mW, linear polarization degree>99%; Optical path design: incident angle θ = 45°, after being split by Wollaston prism, the optical path difference between the two beams ΔL = λ / 4 = 133nm, and the interference fringe period Λ = λ / 2sinθ = 752nm; Image acquisition: A CMOS with a global shutter frame rate of 5000fps, a resolution of 4096×3000, and a pixel size of 2.4μm was used to obtain interference fringes. I ( x ); Channel 2 is infrared transmission imaging, and the parameter design includes: Light source: 850nm LED ring array, the inner diameter of the ring array is 30mm, the outer diameter is 50mm, and the luminous intensity is I 0 = 1500 cd / m 2 , the pulse width is strictly synchronized with the current pulse; Transmittance calculation: ; For location The transmittance at ; For location The detected light intensity at is the initial light intensity before incident; is the absorption coefficient of the material; is the thickness of the material through which the light passes; CMOS exposure time is set to t exp =10 μs , start exposure 5μs after the trailing edge of the current pulse.

4. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 3 is characterized in that: The magnetic field gradient changes inside the screen are detected by the magnetic detection system, including the following: Sensor array: The sensor array consists of 64×64 aluminum nitride piezoelectric resonators, with a total of 4096 resonators. The spacing between two adjacent piezoelectric resonators in the x and y directions is 5 microns. The resonant frequency of each piezoelectric resonator is 15 MHz. The quality factor of the piezoelectric resonator is greater than 2000. The change in the resonant frequency is proportional to the magnetic field strength. The distribution of the magnetic field B is determined by measuring the change in the resonant frequency. Magnetic field gradient measurement: The magnetic field gradient is obtained by the following formula Distribution map: ； in, : The offset of the resonant frequency; : Magnetic field gradient, which indicates the rate of change of magnetic field intensity in space; k : sensitivity coefficient, which indicates the resonant frequency shift caused by each unit magnetic field gradient; Signal processing: A dual lock-in amplifier is used. The minimum phase difference that the lock-in amplifier can distinguish is 0.01°. The output signal of the lock-in amplifier is: ,in is the amplitude of the signal, is the magnetic field phase.

5. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 4 is characterized in that: Cross-modal spatial calibration and noise suppression includes the following steps: S2.1: Cross-modal spatial calibration and noise suppression: S2.1.1: Using Poisson's equation, solve the inverse problem to obtain the magnetic current density map J ( x,y ), the Poisson equation is as follows: ； in, Represents magnetic field B The Laplace operator describes the second-order rate of change of the magnetic field in space; Indicates current density J The curl multiplied by the vacuum permeability μ 0, describes the effect of the rotational characteristics of current density in space on the magnetic field, It is a vector differential operator, which means the curl operation on a vector field; S2.1.2: Calculate the gradient amplitude using the following formula: ； Among them, the current density diagram J ( x,y ) is a two-dimensional scalar function, which means that at each location ( x,y ) and its gradient is a vector field representing the current density in x and y The rate of change in direction, the gradient amplitude is the gradient Length; S2.1.3: Adaptive threshold segmentation of gradient amplitude is performed to extract >15mA / μ m 2 area as defect area; S2.1.4: Through the affine transformation matrix M affine Convert the defect area coordinates to the CMOS pixel coordinate system and control the optical zoom lens to reduce the field of view by 20 times; S2.2: optical noise filtering; S2.2.1: Perform pixel-level difference on two consecutive infrared images: ； in, To express the transmittance The change in time between two frames, and Respectively represent Frame and Frame at position The transmittance at ; S2.2.2: Establishing the noise discrimination matrix N ( x , y ), N ( x , y ) is a binary matrix, which is expressed as follows: ； in, >2% means the pixel value changes by more than 2%; S2.2.3: The noise discriminant matrix N ( x , y ) are marked as ambient noise and are removed in subsequent processing.

6. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 5, characterized in that: Quantifying crack depth specifically includes the following steps: S3.1.1: Polarization interference fringes I ( x ) performs 5-layer discrete wavelet transform; S3.1.2: Calculate the 4th layer detail coefficients d 4, where in the wavelet decomposition, the 4th layer detail coefficient d 4 represents the detail information at the 4th level of decomposition, which reflects the high-frequency components in the stripe image. The variance is calculated as: ； N is the 4th layer detail coefficient d The sample size is 4, that is d The number of data points contained in 4, μ d yes d The mean of 4; S3.1.3: Establish crack depth through calibration experiments h crack The relationship is: ； in, is the 4th layer detail coefficient d Variance of 4.

7. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 6, characterized in that: Differentiating between surface dirt and internal cracks involves the following steps: S3.2.1: Calculate time derivatives, time derivatives It represents the rate of change of transmittance T over time, and the calculation formula is: ； in, and denote the transmittance of the n+1th frame and the nth frame respectively, is the time interval =200 μs ; S3.2.2: Magnetic field phase lag It represents the change of magnetic field phase over time, and the calculation formula is: ； in and Respectively represent the magnetic field phases of the n+1th frame and the nth frame, f 0 is the reference frequency; S3.2.3: Correlation coefficient R Used to measure the time derivative and magnetic field phase lag The correlation between them is calculated as: ； The judgment rules are as follows: like , then it is judged as surface dirty; like , it is judged as an internal crack.

8. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 7, characterized in that: The association analysis of the multimodal data includes a secondary decision tree implementation, which specifically includes the following: First-level decision parameter: Surface damage judgment threshold is <5%, that is, when When <5%, it is judged as a scratch; Secondary decision parameters: interface delamination determination threshold; Condition 1: The local current peak is greater than 1.2 times the average current density of the same type of defect-free OLED screen at the rated driving current; Condition 2: Crack Depth h crack <50 nm ; When both conditions 1 and 2 are met, delamination is determined.

9. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 8, characterized in that: When the secondary decision parameters output conflicting results, that is, when condition 1 and condition 2 are not met at the same time, a closed-loop feedback verification is performed, including the following steps: S4.1: Apply 1.2 to OLED screen I nominal The overload current, I nominal It is the average current density of the same type of OLED screen without defects at the rated driving current; the duration is 3 pulse cycles; S4.2: Transmittance slope calculation: During the 2nd to 3rd pulse cycle of current overload, t exp = 1 μs Ultra-short exposure time to collect infrared transmission image sequences T 1 ,T 2 ,...,T n , sampling rate f s =1 MHz ; The formula is: ； t 0: Current overload start time, marked by FPGA hardware; D t =10 μs ; T ( t ): average transmittance of defect area; S4.3: Magnetic field distortion detection: During current overload, f s =10 MHz Sampling rate records the magnetic field time domain signal output by the piezoelectric resonator array B ( t ); right B ( t ) performs windowed FFT, where the windowing is specifically a Blackman-Harris window with a frequency resolution of Δ f =1 kHz ; Extract the fundamental amplitude A1 and the 3rd harmonic amplitude A3, where the fundamental f 0 =1.2 I nominal Frequency, 3rd harmonic is 3 f 0。 10. The OLED screen defect detection method based on optical-magnetic multimodal fusion according to claim 9, characterized in that: The logical judgment rules for closed-loop feedback verification are as follows: when k T >10 4 / s and when A3 / A1<0.1, the surface is judged to be dirty; when k T <10 4 / s and A3 / A1>0.1, it is judged as internal crack; when k T >10 4 / s and A3 / A1>0.1, it is judged that surface dirt and internal cracks coexist; Update defect classification results and mark the final determination type of the disputed area; If there is still uncertainty, a third-level alarm is triggered and manual re-inspection is performed, while the overload test data is recorded.

Images