LCD (liquid crystal display) defect detection system based on sensor

Abstract

The invention relates to the technical field of LCD (liquid crystal display) detection, and discloses a sensor-based LCD defect detection system, which comprises a frequency sweep synchronous excitation module for generating a frequency sweep excitation signal and an orthogonal reference signal; a pixel addressing driving module; the integrated multi-mode sensing assembly is used for synchronously detecting a pixel transient electric field and acoustic elastic wave response in a non-contact manner; the multi-channel synchronous data acquisition and phase-locked analysis unit is used for performing digital phase-locked demodulation on the response signal; and a master control and data analysis system. And the main control system controls frequency scanning, constructs a multi-physical field dynamic response spectrum for each pixel according to demodulation data, and finally realizes defect detection and classification by comparing and analyzing the response spectrum and a reference spectrum. According to the method, traditional single-point measurement is replaced by dynamic spectral line characterization, and multi-physical field information is fused, so that the detection sensitivity of weak defects is remarkably improved, and automatic and accurate diagnosis of defect types is realized.

Description

Technical Field

[0001] This invention relates to the field of LCD screen inspection technology, specifically to a sensor-based LCD screen defect detection system. Background Technology

[0002] As one of the mainstream technologies in the flat panel display field, the quality and reliability of liquid crystal displays (LCDs) directly depend on the performance consistency of the millions of pixel units that make up the display panel. During the manufacturing process of LCD screens, various micro-defects are inevitably introduced due to factors such as process fluctuations, material defects, or environmental pollution. These defects include threshold voltage drift in thin-film transistors (TFTs), channel leakage, open or short circuits in pixel electrodes, and non-uniformity of the liquid crystal material. These defects are key factors affecting the yield rate, display quality, and long-term operational stability of the display screen.

[0003] Currently, the mainstream method for defect detection in LCD screens is automated optical inspection (AOI). This technology captures and analyzes images of the screen when it is lit, effectively identifying visual defects such as bright spots, dark spots, and murmurs. However, the principle of automated optical inspection dictates that it primarily targets "consequence-type" defects that have already produced optical anomalies. Its detection capability is very limited for subtle degradations in electrical performance or potential physical structural flaws that are in their nascent stages and have not yet caused significant visual changes. These latent defects may not exhibit optical anomalies in the early stages of a product, but they will evolve and deteriorate under the electrical and thermal stresses of long-term operation, eventually leading to pixel or regional display failure and affecting the long-term reliability of the product.

[0004] To compensate for the shortcomings of optical inspection, the industry has explored various electrical testing methods. However, traditional electrical inspection often relies on physical probe contact, which can cause irreversible physical damage to the delicate pixel structure and is inefficient, making it difficult to meet the demands of full-screen, rapid inspection on large-scale production lines. Existing non-contact electrical inspection solutions mostly measure static parameters such as capacitance and resistance of pixels at a single or a few fixed frequency points. This single-point measurement provides very limited information and cannot comprehensively reflect the dynamic response characteristics of pixel units under different operating conditions. The physical health of a pixel is essentially a comprehensive reflection of its complex electrical, mechanical, and thermal characteristics, which are often closely related to the frequency of the excitation signal. Therefore, static or single-frequency measurement methods may miss a large amount of defect feature information hidden within specific frequency bands, leading to misjudgment or missed detection of defect types, making it difficult to meet increasingly stringent quality control requirements. Summary of the Invention

[0005] This invention provides an LCD screen defect detection system and method based on frequency sweep excitation and multi-physics dynamic response spectrum analysis, aiming to solve the problems of insufficient sensitivity to weak defects, single physical feature characterization of defects, and difficulty in accurate and automated classification of defect types in existing LCD screen detection technologies.

[0006] To achieve the above objectives, the present invention provides a sensor-based LCD screen defect detection system, comprising:

[0007] The system includes a frequency sweep synchronization excitation module, a pixel addressing drive module, an integrated multimodal sensing assembly, a multi-channel synchronous data acquisition and phase-locked analysis unit, and a main control and data analysis system.

[0008] The frequency sweep synchronization excitation module is used to generate a periodic excitation signal whose frequency sweeps within a preset frequency set, and two orthogonal digital reference signals that are strictly in phase and frequency with the excitation signal. At any given excitation frequency... The specific signals output by this module are as follows:

[0009] Instantaneous voltage of the excitation signal applied to the LCD pixel :

[0010] In-phase reference signal used for phase-locked loop analysis and quadrature reference signals :

[0011] ;

[0012] ;

[0013] in, The instantaneous voltage of the excitation signal, The amplitude of the excitation signal, The excitation frequency for the current cycle. For time variables, For in-phase reference signal, This is an orthogonal reference signal.

[0014] The pixel addressing driving module receives the excitation signal and, according to the instructions of the main control and data analysis system, applies the excitation to the target pixel of the LCD screen under test.

[0015] The integrated multimodal sensing assembly is disposed above the working surface of the screen under test in a non-contact manner, and is used to simultaneously detect the transient electric field response signal and the acoustic elastic wave response signal generated by the target pixel under excitation.

[0016] The multi-channel synchronous data acquisition and phase-locked analysis unit has its signal input terminal connected to the sensing assembly and its reference signal input terminal connected to the frequency sweep synchronous excitation module. This unit synchronously acquires the raw response signals of both electric and acoustic modes, and uses the input orthogonal digital reference signal to perform real-time digital phase-locked demodulation on the two raw signals, thereby accurately calculating the in-phase and quadrature components of the response signal of each pixel at the current excitation frequency.

[0017] The main control and data analysis system communicates with the frequency sweep synchronization excitation module and the multi-channel synchronous data acquisition and phase-locked analysis unit for control and data exchange. This system sets and controls the frequency sweep protocol, receives the demodulated data, and constructs a multiphysics dynamic response spectrum for each pixel based on this data. Finally, the system performs defect detection, location, and diagnosis by comparing and analyzing the measured response spectrum with a reference spectrum.

[0018] One of the core technical features of this invention is that, by implementing frequency scanning, pixel detection is upgraded from traditional static single-point measurement to a dynamic spectral characterization. The system controls the frequency of the excitation signal to scan point by point within a preset frequency set, and precisely demodulates the multi-physics response amplitude and phase of the pixel at each frequency point. The resulting series of frequency-sorted data points constitute the unique multi-physics dynamic response spectrum of that pixel. This spectrum fully depicts the dynamic behavior of the pixel's physical response as the excitation frequency changes. Its morphology, such as the position of the resonance peak, peak value, full width at half maximum (FWHM), and phase change rate, together constitute the physical state fingerprint of the pixel, thereby enabling more sensitive detection and more refined classification and diagnosis of defects.

[0019] This invention provides a sensor-based LCD screen defect detection system. It offers the following advantages:

[0020] 1. This invention adopts a combination of frequency sweep excitation and digital phase-locked demodulation technology, which can construct the dynamic response spectrum of the target pixel. This method can not only capture weak defect features that only appear at specific resonance frequency points, but also effectively extract nanovolt-level response signals from strong noise backgrounds by using phase-locked analysis, thereby significantly improving the detection capability of early and small defects, which is difficult to achieve by traditional single-frequency and static measurement methods.

[0021] 2. This invention, through an integrated multimodal sensing assembly, simultaneously acquires the transient electric field response and acoustoelastic wave response of the excited pixel. These two physical modes originate from different physical mechanisms and can reflect the pixel's state from both electrical and mechanical structural perspectives. Combining these two modes for analysis allows for a more comprehensive and in-depth physical profile of the defect, effectively avoiding the information limitations inherent in single-modal detection.

[0022] 3. The main control and data analysis system of this invention not only collects data, but also constructs a multi-physics dynamic response spectrum and extracts a high-dimensional feature vector containing cross-modal coupling information. By comparing this feature vector with a pre-constructed defect spectrum library and analyzing it using a trained classifier, the system can automatically output a clear defect classification result. This process transforms complex spectral analysis into a standardized data processing workflow, realizing an intelligent upgrade from "defect detection" to "defect type identification". Attached Figure Description

[0023] Figure 1 This is a system structure block diagram of one embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram illustrating the complete workflow of a system according to an embodiment of the present invention;

[0025] Figure 3 This is a block diagram of the internal structure of a frequency sweep synchronization excitation module according to an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of the hierarchical structure of an integrated multimodal sensing assembly according to an embodiment of the present invention;

[0027] Figure 5 This is a schematic diagram showing the correspondence between a micro-area transient electric field sensing array and an LCD pixel unit according to an embodiment of the present invention;

[0028] Figure 6 This is a schematic diagram illustrating the working principle of a transparent piezoelectric thin film sensing unit according to an embodiment of the present invention;

[0029] Figure 7 This is a hardware structure block diagram of a multi-channel synchronous data acquisition and phase-locked analysis unit according to an embodiment of the present invention;

[0030] Figure 8 This is a schematic diagram of a defect diagnosis process based on a defect spectrum library and a machine learning classifier, according to an embodiment of the present invention.

[0031] The components include: 10. Sweep frequency synchronous excitation module; 11. Direct digital frequency synthesizer; 12. High-stability reference clock source; 13. Microcontroller unit; 14. Low-pass filter; 15. Programmable gain amplifier; 16. Digital logic and level conversion circuit; 20. Pixel addressing drive module; 30. Integrated multimodal sensing assembly; 31. Optical-grade transparent substrate; 32. Transparent piezoelectric thin film array; 321. Piezoelectric sensing unit; 33. Transparent insulating dielectric layer; 34. Micro-area transient electric field sensing array; 341. Sensing electrode; 35. Transparent electromagnetic shielding layer; 40. Multi-channel synchronous data acquisition and phase-locked loop analysis unit; 41. Multi-channel analog front end; 42. High-speed synchronous analog-to-digital converter group; 43. Field-programmable gate array; 44. High-speed data buffer; 45. Data interface; and 50. Main control and data analysis system. Detailed Implementation

[0032] The technical solutions in 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] See attached document Figure 1 This invention provides a sensor-based LCD screen defect detection system, which may include:

[0034] The system includes a frequency sweep synchronization excitation module 10, a pixel addressing drive module 20, an integrated multimodal sensing assembly 30, a multi-channel synchronous data acquisition and phase-locked analysis unit 40, and a main control and data analysis system 50.

[0035] The frequency sweep synchronization excitation module 10 is used to generate a periodic excitation signal whose frequency sweeps within a preset frequency set, and two orthogonal digital reference signals that are strictly in phase and frequency with the excitation signal. At any given excitation frequency of the current period... The specific signals output by this module are as follows:

[0036] Instantaneous voltage of the excitation signal applied to the LCD pixel :

[0037] In-phase reference signal used for phase-locked loop analysis and quadrature reference signals :

[0038] ;

[0039] ;

[0040] in, The instantaneous voltage of the excitation signal, The amplitude of the excitation signal, The excitation frequency for the current cycle. For time variables, For in-phase reference signal, This is an orthogonal reference signal. The excitation signal output of this module is connected to the pixel addressing drive module 20, and the reference signal output is connected to the multi-channel synchronous data acquisition and phase-locked analysis unit 40.

[0041] The pixel addressing driver module 20 has its input connected to the sweep frequency synchronization excitation module 10 and communicates with the main control and data analysis system 50 via commands. This module can be physically integrated into or externally connected to the driver circuit board of the LCD screen under test. Its core function is to precisely control the gate driver and source driver chips of the screen under test according to the target pixel coordinate commands (e.g., row address and column address) issued by the main control and data analysis system 50. It controls the gate driver to turn on the thin-film transistors (TFTs) of a specific row, and simultaneously the source driver to transmit the instantaneous voltage received from the excitation signal of the sweep frequency synchronization excitation module 10. Loaded into a specific column, this module enables precise gating and excitation application to any single or multiple target pixels.

[0042] An integrated multimodal sensing assembly 30 is non-contactly positioned above the working surface of the LCD screen under test. This assembly integrates a micro-region transient electric field sensing array and a transparent piezoelectric thin film array. The micro-region transient electric field sensing array is used to detect the transient electric field response signal generated by the pixel thin-film transistor (TFT) operating under high-frequency excitation. The transparent piezoelectric thin film array is used to detect the acoustoelastic wave response signal induced by electrostriction and thermoelastic effects. This assembly also works in conjunction with an externally mounted high-speed infrared thermal imager and high-resolution optical camera to simultaneously acquire thermal radiation signals and optical images.

[0043] The multi-channel synchronous data acquisition and phase-locked analysis unit 40 has its signal input terminal connected to the integrated multimodal sensing assembly 30 and external sensing modules, and its reference signal input terminal connected to the sweep frequency synchronous excitation module 10. This unit is used to synchronously acquire the raw response signals of three modes: electric field, sound field, and thermal field, and to perform real-time digital phase-locked demodulation of the above three raw signals using the input reference signal, thereby calculating the in-phase component and quadrature component of the response signal of each pixel at the current excitation frequency.

[0044] The main control and data analysis system 50 communicates with the frequency sweep synchronization excitation module 10 and the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40 for control and data exchange. This system sets and controls the frequency sweep protocol of the frequency sweep synchronization excitation module 10, receives demodulated data from the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40, and constructs a multi-physics dynamic response spectrum for each pixel based on this data. Finally, the system performs defect detection, location, and diagnosis by comparing and analyzing the measured response spectrum with a reference spectrum.

[0045] See attached document Figure 1 The overall workflow of the system is uniformly scheduled and closed-loop controlled by the main control and data analysis system 50. In a complete detection process, the main control and data analysis system 50 first sends an instruction to the frequency scanning synchronization excitation module 10 according to the preset frequency scanning protocol, specifying the excitation frequency of the current cycle. Simultaneously, the main control and data analysis system 50 sends instructions to the pixel addressing driver module 20, specifying the coordinates of the target pixel to be excited.

[0046] After receiving the frequency command, the frequency sweep synchronization excitation module 10 accurately generates and outputs two types of signals: the first type is an analog excitation signal, whose instantaneous voltage varies with time. The signal is sent to the pixel addressing driver module 20 and applied to a specified pixel on the screen under test; the second type is a digital in-phase reference signal. Orthogonal reference signal These two signals, serving as pure time and phase references, are directly sent to the multi-channel synchronous data acquisition and phase-locked analysis unit 40.

[0047] When a specified pixel is driven by a high-frequency excitation signal, it will produce a variety of weak physical responses. The integrated multimodal sensing assembly 30 synchronously acquires these responses and transmits these multi-channel raw analog signals, which contain response components at the same frequency as the excitation signal and background noise, to the multi-channel synchronous data acquisition and phase-locked analysis unit 40.

[0048] The core function of the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40 is to perform phase-sensitive detection. It uses the digital reference signal received from the frequency sweep synchronous excitation module 10 to perform mixing and low-pass filtering on the multiple noisy raw response signals received from the integrated multimodal sensing assembly 30. This process effectively suppresses noise unrelated to the excitation frequency and accurately extracts the components in the response signal that are in phase with and quadrature with the excitation. The output of this unit is no longer a time-domain waveform, but rather numerical data representing amplitude and phase.

[0049] The quantified data is transmitted back to the main control and data analysis system 50. The main control and data analysis system 50 uses this data point as the excitation frequency of the current pixel in the current cycle. The response characteristics are stored. Subsequently, the system will iteratively execute this process, whereby the main control and data analysis system 50 issues the excitation frequency for the next cycle. The process follows the instructions until all frequency points within the preset frequency set are scanned. The entire process forms a precise closed loop of instruction issuance, excitation application, response acquisition, phase-locked demodulation, and data feedback, and is repeatedly executed on each pixel in a frequency scanning manner.

[0050] To further clarify the present invention, the specific structure and implementation of each functional module in the system will be described in detail below.

[0051] See attached document Figure 2 Before the test begins, the LCD screen to be tested is pre-processed, including placing it securely on the test platform, completing the electrical connection with the pixel addressing drive module 20, and adjusting the position of the integrated multimodal sensing assembly 30 to ensure a preset, precise, parallel detection distance between it and the surface of the screen under test. Simultaneously, the operator sets various parameters for the test, including the starting frequency of the frequency scan, through the user interface of the main control and data analysis system 50. Termination frequency Frequency step size And the screen area to be detected.

[0052] After the detection process is initiated, the system strictly follows the technical framework of excitation, sensing, demodulation, and analysis. In the excitation phase, the main control and data analysis system 50 controls the frequency sweep synchronization excitation module 10 and the pixel addressing drive module 20 at the preset excitation frequency for the first cycle. Next, apply a high-frequency excitation to the specified first pixel.

[0053] In the sensing stage, when an excitation signal is applied to a pixel, the integrated multimodal sensing assembly 30 and the external sensing module immediately and synchronously acquire multiple raw response signals of the pixel in the electric field, sound field, and thermal field. The intensity of these signals is usually extremely weak and is drowned out by strong background noise.

[0054] In the demodulation stage, the acquired multiple raw response signals and the clean reference signal provided by the frequency sweep synchronization excitation module 10 are sent together to the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40. This unit executes the core phase-sensitive detection algorithm, matching the excitation frequency of the first cycle in the raw signal. The strictly synchronized response components are precisely demodulated, and their amplitude and phase information are calculated. This process greatly improves the signal-to-noise ratio.

[0055] In the analysis phase, the demodulated excitation frequency of the pixel in the first cycle is obtained. The amplitude and phase data are transmitted to the main control and data analysis system 50 for storage. The system then iteratively executes the above-mentioned excitation, sensing, and demodulation process until all preset excitation frequency points have been traversed. At this point, the complete multiphysics dynamic response spectrum of that pixel is constructed. Afterward, the system moves to the next pixel and repeats the entire spectrum construction process until the entire test area is covered.

[0056] Once the dynamic response spectrum data of all target pixels has been collected and constructed, the main control and data analysis system 50 enters the final diagnostic stage. The system compares the measured response spectrum of each pixel with the standard defect-free reference spectrum pre-stored in the database, calculates the deviation, and identifies abnormal pixels. For abnormal pixels, the system further extracts the key features of their spectral lines and matches them with the defect spectrum library, thereby outputting a diagnostic report on the location, type, and related physical parameters of the defects.

[0057] To further clarify the present invention, the specific structure and implementation of each functional module in the system will be described in detail below.

[0058] A core technical feature of this invention is that by implementing a frequency scanning operating mode, pixel detection is upgraded from traditional static single-point measurement to a dynamic spectral characterization. Traditional detection methods are typically performed at a single fixed excitation frequency, and the measurement result is only a response amplitude and phase at that specific frequency point. This single-point data has limitations in characterizing the complete electrical and physical properties of a pixel, and may miss defective features that only appear at other frequencies.

[0059] This invention controls the frequency sweep synchronization excitation module 10 to set the excitation frequency of its output excitation signal for the current period. In a preset frequency set Intra-point scanning. For each pixel within the test area, the system performs a complete frequency scan. During this process, the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40 will perform a frequency scan at the excitation frequency of each current cycle. The amplitude and phase of the multiphysics response of the pixel are accurately demodulated.

[0060] After a complete frequency scan, the system obtains a series of response data points ordered by frequency for that pixel. Connecting these data points forms the unique multiphysics dynamic response spectrum of that pixel, which fully depicts the dynamic behavior of the pixel's physical response as the excitation frequency changes.

[0061] The morphology of spectral lines, such as the position of resonance peaks, peak height, full width at half maximum (FWHM), and the rate of phase change with frequency, collectively constitutes the unique physical fingerprint of a pixel. Different types of minute defects within a pixel alter the reactance and susceptance of its equivalent circuit in different ways, resulting in morphologically distinctive changes in its dynamic response spectrum that can be precisely identified. Therefore, by analyzing the complete response spectrum rather than a single data point, this system can achieve more sensitive defect detection and more refined classification and diagnosis.

[0062] To further clarify the present invention, the specific structure and implementation of each functional module in the system will be described in detail below.

[0063] See attached document Figure 3 The frequency sweep synchronization excitation module 10 is based on Direct Digital Synthesis (DDS) technology. Physically, this module may include a direct digital frequency synthesizer 11, a high-stability reference clock source 12, a microcontroller unit 13, a low-pass filter 14, a programmable gain amplifier 15, and a digital logic and level conversion circuit 16.

[0064] A high-stability reference clock source 12, such as a temperature-compensated crystal oscillator (TCXO), provides a highly stable main system clock with extremely low phase noise for the direct digital frequency synthesizer 11, which is the basis for ensuring the frequency accuracy and phase purity of the generated signal.

[0065] The microcontroller unit 13, serving as the local control core of this module, is connected to the main control and data analysis system 50 via a communication interface. It is responsible for receiving the excitation frequency for the current cycle from the main control system. and the amplitude of the excitation signal Upon receiving the instruction, the microcontroller unit 13 converts the frequency instruction into a frequency control word in a specific format required by the direct digital frequency synthesizer 11, and converts the amplitude instruction into a gain control signal for the programmable gain amplifier 15.

[0066] The direct digital frequency synthesizer 11 is the core of signal generation. Based on the frequency control word provided by the microcontroller unit 13 and the reference clock from the high-stability reference clock source 12, it generates a digitized sine wave data stream and two digital quadrature square wave signals that are strictly synchronized with the waveform through an internal phase accumulator and sine lookup table.

[0067] The digitized sinusoidal waveform data stream is fed into the digital-to-analog converter (DAC) inside the direct digital frequency synthesizer 11, converting it into a stepped analog voltage signal. This signal first passes through a low-pass filter 14 to filter out the Nyquist image frequency and harmonic components generated during the digital-to-analog conversion, thus forming a smooth sine wave. Subsequently, the smoothed sine wave signal enters the programmable gain amplifier 15, which precisely adjusts its output amplitude according to the instructions of the microcontroller unit 13, ultimately generating an excitation signal whose instantaneous voltage is... And the amplitude of this excitation signal satisfies Requirements.

[0068] Simultaneously, the two digital quadrature square wave signals generated by the direct digital frequency synthesizer 11 are sent to the digital logic and level conversion circuit 16. This circuit shapes and levels the signals to ensure that the output in-phase reference signal is in line with the circuit's parameters. and quadrature reference signals It is compatible with the digital input interface standard of the multi-channel synchronous data acquisition and phase-locked analysis unit 40.

[0069] To further clarify the present invention, the working principle and signal generation process of the frequency sweep synchronization excitation module 10 will be described in detail below.

[0070] The synchronization and accuracy of the output signal of the frequency sweep synchronization excitation module 10 are rooted in the direct digital frequency synthesis technology adopted by the module.

[0071] The synchronicity of signal output is ensured through a mechanism of co-source generation. This includes the excitation signal and the in-phase reference signal. Orthogonal reference signal These three outputs all originate from the same digital processing core within the direct digital frequency synthesizer 11. Inside the direct digital frequency synthesizer 11, a phase accumulator driven by a high-stability reference clock source 12 provides a unified, discrete phase angle reference for the generation of all signals. The digital value of the sine wave used to generate the excitation signal, along with the in-phase reference signal... and quadrature reference signals The logic levels of all three output signals are derived from this unified phase angle reference. This common-source generation mechanism fundamentally ensures that there is no relative phase drift between the three output signals.

[0072] The accuracy of signal output is reflected in two dimensions: frequency and amplitude. First, the frequency stability of the high-stability reference clock source 12 directly determines the excitation frequency of the current cycle of all output signals. The accuracy of the excitation signal. Secondly, the amplitude of the excitation signal. The accuracy is ensured by the bit depth of the digital-to-analog converter inside the direct digital frequency synthesizer 11 and the gain resolution of the programmable gain amplifier 15. The microcontroller unit 13 precisely sets the gain value of the programmable gain amplifier 15 according to the instructions of the main control and data analysis system 50, thereby achieving precise control over the amplitude of the excitation signal. To perform precise control via software.

[0073] After explaining the frequency sweep synchronization excitation module 10 responsible for generating the excitation signal, the integrated multimodal sensing assembly 30 responsible for sensing the multi-physics response signal will be described in detail below.

[0074] See attached document Figure 4 The integrated multimodal sensing assembly 30 is designed as a compact, multi-layered stacked composite structure that enables the synchronous detection of the response signals of two physical fields, electric field and acoustoelastic wave, in space.

[0075] The bottom layer of the integrated multimodal sensing assembly 30 is an optically grade transparent substrate 31, which can be made of high-transmittance quartz glass or polymer material. The optically grade transparent substrate 31 provides stable mechanical support for the entire sensing assembly and allows an external high-resolution optical camera to penetrate the sensing assembly and acquire optical images of the pixels of the LCD screen under test without obstruction.

[0076] A transparent piezoelectric thin film array 32 is deposited and patterned on the upper surface of an optical-grade transparent substrate 31. The material of this array can be a polyvinylidene fluoride-trifluoroethylene (PVDF-TrFE) copolymer, which possesses excellent piezoelectric and acoustic properties. Using standard photolithography or laser etching processes, the continuous piezoelectric thin film is divided into multiple independent sensing units. The physical dimensions of each sensing unit match the dimensions of one or more pixels of the LCD screen under test, thereby achieving high spatial resolution acoustoelastic wave detection capability.

[0077] A transparent insulating dielectric layer 33 is disposed above the transparent piezoelectric thin film array 32, which serves to achieve electrical isolation between the upper and lower sensing arrays.

[0078] A micro-region transient electric field sensing array 34 is fabricated on the transparent insulating dielectric layer 33. This array consists of a series of precisely patterned microelectrodes, used to non-contactly sense the local electric field changes formed between the gate and source of the pixel thin-film transistor (TFT) during operation.

[0079] To shield against external electromagnetic interference and reduce interlayer crosstalk, the top layer of the entire sensing array is covered with a grounded transparent electromagnetic shielding layer 35, which can be made of transparent conductive materials such as indium tin oxide (ITO).

[0080] The output electrodes of all sensing units are connected to a flexible printed circuit board (FPC) interface via high-density micro-leads, ultimately exporting the raw response signals of the multi-channel to the multi-channel synchronous data acquisition and phase-locked analysis unit 40.

[0081] To further clarify the present invention, the working principle and signal acquisition process of each sensor array in the integrated multimodal sensing assembly 30 will be described in detail below.

[0082] See attached document Figure 5 The working principle of the micro-region transient electric field sensing array 34 is to capture, non-contactly, the transient edge electric field leaked into the surrounding space by a single pixel thin-film transistor (TFT) during operation through capacitive coupling effect. When an excitation signal is applied to the pixel, the voltage between the gate and source of the TFT fluctuates at a high frequency, and this fluctuation generates a local electric field that changes synchronously with time around the TFT device.

[0083] This sensing array consists of a series of physically independent sensing electrodes 341. These sensing electrodes 341 are fabricated by depositing and patterning a transparent conductive material (such as indium tin oxide) on a transparent insulating dielectric layer 33. In terms of design, the size, shape, and spatial position of each sensing electrode 341 are precisely calculated to accurately correspond to one or a small group of pixel units of the LCD screen under test. This one-to-one or one-to-many precise spatial correspondence is the basis for achieving high-resolution defect localization.

[0084] Because the edge electric field strength generated by the TFT is extremely weak, the charge or voltage signal directly coupled from the sensing electrode 341 is also extremely weak. To effectively extract this signal, the output of each sensing electrode 341 is connected to a dedicated preamplifier circuit with high input impedance and low noise characteristics. The preamplifier circuit can be a charge amplifier or a voltage follower, its function being to convert the weak sensed electrical signal into a voltage signal with higher amplitude and stronger driving capability, while simultaneously performing impedance matching to facilitate subsequent signal transmission and processing. These preamplifier circuits can be integrated onto the flexible printed circuit board interface of the sensing assembly to shorten the signal transmission path and minimize noise coupling.

[0085] To further clarify the present invention, the working principle of the transparent piezoelectric thin film array will be described in detail below.

[0086] See attached document Figure 6The working mechanism of the transparent piezoelectric thin film array 32 lies in converting the acoustoelastic wave response signal generated by the excited pixel into an electrical signal. When the excitation signal is applied to the pixel under test, the material inside the pixel will generate weak mechanical vibrations synchronized with the frequency of the excitation signal due to electrostriction and thermoelasticity. This mechanical vibration propagates outward in the form of acoustoelastic waves, passes through the detection gap, and finally reaches the surface of the transparent piezoelectric thin film array 32.

[0087] This array consists of multiple independent piezoelectric sensing units 321. Physically, a continuous transparent conductive film can be deposited on an optically grade transparent substrate 31 as a common electrode. Then, a polyvinylidene fluoride-trifluoroethylene (PVDF-TrFE) copolymer film is coated on top of this film. Finally, a series of independent, patterned transparent conductive films are fabricated on the top surface of the copolymer film using photolithography to serve as sensing electrodes. The area defined by each sensing electrode and the common electrode below constitutes an independent piezoelectric sensing unit 321.

[0088] When the pressure wavefront of an acoustoelastic wave impacts the piezoelectric sensing unit 321, it causes deformation of the polyvinylidene fluoride-trifluoroethylene copolymer film. According to the piezoelectric effect of the material, this mechanical deformation generates a charge proportional to the stress between the sensing electrode and the common electrode. Because the intensity of the acoustoelastic wave is extremely weak, the amount of induced charge generated is also very small. To effectively extract this signal, the sensing electrode of each piezoelectric sensing unit 321 is connected to a dedicated high-sensitivity charge amplifier. The charge amplifier converts the weak induced charge signal into a voltage signal with significantly increased amplitude, while minimizing noise interference, and then outputs the amplified signal to the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40.

[0089] To achieve simultaneous acquisition of multiple modal signals such as electric field, acoustic elastic wave, thermal radiation and optical image, this invention has specially designed the spatial layout of the integrated multimodal sensing assembly and the external sensing module. The following will describe this collaborative working scheme in detail.

[0090] The core objective of spatial layout and optical path design is to achieve non-interference in physical space detection and signal acquisition among the integrated multimodal sensing assembly 30, high-speed infrared thermal imager and high-resolution optical camera in the same target area of ​​the LCD screen under test.

[0091] To achieve common-path detection, the integrated multimodal sensor assembly 30, the high-speed infrared thermal imager, and the high-resolution optical camera are arranged on a common main optical axis perpendicular to the surface of the LCD screen under test. Thanks to its overall transparent design, the integrated multimodal sensor assembly 30 allows optical and thermal radiation signals from the pixels under test to pass through without attenuation.

[0092] A dichroic beamsplitter is positioned above the integrated multimodal sensing assembly 30 on the main optical axis. A dichroic beamsplitter is a special optical element that separates light paths based on wavelength. In this configuration, the dichroic beamsplitter is designed to efficiently reflect infrared radiation of a specific wavelength band while efficiently transmitting visible light. The thermal radiation signal emitted by the pixel under test, after passing through the integrated multimodal sensing assembly 30, is reflected by the dichroic beamsplitter to a high-speed infrared thermal imager on the measuring optical axis. Simultaneously, the visible light image signal emitted by the pixel under test, after passing through the integrated multimodal sensing assembly 30, is transmitted through the dichroic beamsplitter and received by a high-resolution optical camera on the main optical axis. This optical path design allows two external cameras to observe the same target area without obstructing each other.

[0093] To ensure long-term stability of spatial alignment, the integrated multimodal sensor assembly 30, dichroic beam splitter, high-speed infrared thermal imager, and high-resolution optical camera are all fixed on a high-rigidity optical platform. Before detection begins, the system establishes a pixel-level precise mapping relationship between the infrared image, optical image, and sensor assembly array unit through a standard calibration procedure, thereby ensuring the accuracy of subsequent data analysis.

[0094] To further clarify the present invention, the structure and working principle of the multi-channel synchronous data acquisition and phase-locked analysis unit 40 will be described in detail below.

[0095] See attached document Figure 7 The hardware core of the multi-channel synchronous data acquisition and phase-locked analysis unit 40 is a high-speed parallel signal processing system built around a field-programmable gate array 43 (FPGA). This architecture provides independent, real-time digital phase-locked amplification processing capabilities for each sensing channel from the integrated multimodal sensing assembly 30.

[0096] The hardware configuration of the unit may physically include a multi-channel analog front end 41, a high-speed synchronous analog-to-digital converter group 42, a field-programmable gate array 43 (FPGA), a high-speed data buffer 44, and a data interface 45 for communicating with the main control and data analysis system 50.

[0097] The multi-channel analog front-end 41 is the first stage through which signals enter this unit. This front-end includes a programmable gain amplifier (PGA) and an anti-aliasing low-pass filter, each configured independently for each input channel. The PGA adaptively adjusts the signal gain based on the signal strength, ensuring that the signal amplitude optimally matches the dynamic range of the subsequent analog-to-digital converter. The anti-aliasing low-pass filter filters out noise components above the Nyquist frequency before signal digitization, preventing aliasing distortion in the signal spectrum.

[0098] The high-speed synchronous analog-to-digital converter group 42 is responsible for converting the conditioned multiple analog signals into digital data streams. All analog-to-digital conversion channels within this converter group share the same sampling clock. This sampling clock is strictly synchronized with the system master clock of the sweep frequency synchronous excitation module 10, thereby ensuring that the signal sampling of all sensing channels is perfectly aligned in time.

[0099] The Field Programmable Gate Array (FPGA) 43 is the core of the entire unit's computation. This FPGA receives parallel digital data streams from the high-speed synchronous analog-to-digital converter group 42, and simultaneously receives in-phase reference signals from the swept-frequency synchronous excitation module 10. and quadrature reference signals The parallel processing architecture of FPGAs allows for the instantiation of a completely independent digital lock-in amplifier logic for each sensing channel within the chip. This design ensures that data processing for all channels is performed in parallel and in real time, without the time delay caused by channel switching or multiplexing.

[0100] A high-speed data buffer 44, such as double data rate synchronous dynamic random access memory (DDR SDRAM), is used to temporarily store a large amount of amplitude and phase data obtained after processing by the field-programmable gate array 43. A data interface 45, such as a PCI Express (PCIe) or 10 Gigabit Ethernet interface, is responsible for transmitting the data in the buffer to the main control and data analysis system 50 at high speed for subsequent data fusion and defect diagnosis.

[0101] To further clarify the present invention, the core algorithm of the digital phase-locked amplification of this unit will be described in detail below.

[0102] Inside the Field Programmable Gate Array 43 (FPGA), leveraging its inherent parallel processing capabilities, a completely independent digital lock-in amplifier logic is instantiated for each sensing channel. The signal processing flow of all channels is performed synchronously, thereby realizing true full-channel parallel real-time analysis.

[0103] For any sensing channel, the digital signal received from the high-speed synchronous analog-to-digital converter group 42 is first sent to a dual-channel processing path. In the first path, the digital signal is compared with the in-phase reference signal from the frequency sweep synchronous excitation module 10. Perform point-by-point digital multiplication. In the second path, the digital signal and the orthogonal reference signal... The same digital multiplication operation is performed. Both multiplication operations are implemented by the hardware multiplier logic resources inside the FPGA, ensuring extremely high operation speed.

[0104] The output of both multiplication operations includes a DC component proportional to the amplitude and phase of the signal under test, and an excitation frequency equal to the current cycle. Twice the high-frequency components. To extract the required DC component, both signals are passed through a digital low-pass filter. This filter preferably employs a finite impulse response (FIR) filter structure because FIR filters have stable characteristics and linear phase response. The filter order and cutoff frequency, among other parameters, can be dynamically configured by the main control and data analysis system 50 to adapt to different sweep speeds and signal-to-noise ratio requirements.

[0105] After filtering, two stable DC values ​​are finally obtained: the in-phase component. Orthogonal components This pair ( The numerical value fully characterizes the excitation frequency in the current cycle. Below, the amplitude and phase information of the response signal received by the sensing channel relative to the excitation signal. All channels ( The data pairs are packaged together and transmitted to the high-speed data buffer 44.

[0106] To further clarify the present invention, the data processing flow and defect diagnosis algorithm of the main control and data analysis system will be described in detail below.

[0107] One of the core tasks of the main control and data analysis system 50 is to build and maintain a dedicated multi-physics dynamic response spectrum data set for each pixel of the LCD screen under test in real time throughout the entire frequency sweep detection process.

[0108] The main control and data analysis system 50 continuously receives data streams from the multi-channel synchronous data acquisition and phase-locked loop analysis unit 40 via a high-speed data interface 45. This data stream contains the excitation frequency for each discrete current cycle. Below, the in-phase components of all micro-region transient electric field sensing channels and all transparent piezoelectric thin film sensing channels. Orthogonal components At the same time, the main control and data analysis system 50 also synchronously triggers the high-speed infrared thermal imager to acquire the excitation frequency in the same current cycle. Below is the thermal response image of the corresponding pixel area.

[0109] Based on a pre-defined spatial mapping relationship, the system precisely correlates the data from each sensing channel to a specific pixel coordinate on the LCD screen under test. For each pixel, the main control and data analysis system 50 first processes the in-phase components of the received electric field and acoustic response. Orthogonal components Perform calculations to determine the amplitude of the electric field response. With phase and the amplitude of the acoustic elastic wave response. With phase For the thermal image sequences acquired by the infrared thermal imager, the system also extracts the amplitude of the thermal response signal using a phase-locked loop analysis algorithm. With phase .

[0110] Subsequently, the system maintains a dedicated data structure for each pixel in memory. This data structure contains six independent, dynamically growing numerical arrays. These six arrays are: electric field response amplitude spectrum vector, electric field response phase spectrum vector, acoustoelastic wave response amplitude spectrum vector, acoustoelastic wave response phase spectrum vector, thermal response amplitude spectrum vector, and thermal response phase spectrum vector.

[0111] As the frequency sweep process progresses from the starting frequency to the ending frequency, the system at each frequency point... After completing the above calculations, the six values ​​obtained will be... Each of the six arrays in the corresponding pixel data structure is appended to the end of the array. When the entire frequency sweep process is completed, each pixel's data structure will have stored six dynamic response spectra covering the entire frequency sweep range.

[0112] To further clarify the present invention, the following will provide a detailed description of the feature extraction and fusion algorithm based on these multiphysics dynamic response spectra.

[0113] Within the main control and data analysis system 50, there is a dedicated software algorithm module. The core function of this module is to transform the multiphysics dynamic response spectrum of each pixel, which consists of six original spectral vectors, into a numerical feature vector with lower dimension but higher information density. This transformation is a prerequisite for achieving efficient and accurate defect diagnosis.

[0114] This feature extraction process is performed independently on the six spectral vectors of each pixel, and mainly includes the following levels of calculation:

[0115] First, the algorithm analyzes each individual amplitude spectrum vector. It automatically searches for one or more formants in the spectrum. For each identified formant, the algorithm extracts three key parameters: the center frequency of the formant, the peak amplitude of the formant, and the full width at half maximum (FWHM) of the formant. The center frequency reflects the inherent response frequency of the pixel under a specific excitation, the peak amplitude characterizes the intensity of the response, and the FWHM quantifies the sharpness of the resonance, directly related to the system's energy dissipation or damping characteristics. Furthermore, the algorithm calculates the integral of the entire amplitude spectrum over the sweeping frequency range; this integral value represents the total response energy of the pixel throughout the entire excitation process.

[0116] Secondly, the algorithm analyzes each individual phase spectrum vector. It focuses on regions near the resonance peak frequency of the amplitude spectrum where the phase changes drastically. The algorithm extracts the frequency of the point of fastest phase change and the phase value at that frequency. This phase value characterizes the time delay of the response signal relative to the excitation signal.

[0117] Finally, cross-modal features are extracted. The algorithm calculates the correlation characteristics between responses from different physical fields. For example, it calculates the difference between the resonant peak frequencies of the acoustoelastic wave response amplitude spectrum and the electric field response amplitude spectrum. It also calculates the difference between the phase of the thermal response signal and the phase of the acoustoelastic wave response signal at the same key frequency point (e.g., the principal resonant frequency). These cross-modal features reveal the coupling strength and temporal relationships between different physical effects, which is crucial for identifying certain types of deep-seated defects.

[0118] All the values ​​obtained through the above calculations (e.g., resonance frequency, peak value, full width at half maximum, total energy, phase delay, cross-modal frequency difference and phase difference, etc.) are finally combined into a high-dimensional numerical feature vector. Each vector uniquely corresponds to a pixel and serves as the input for the subsequent defect diagnosis model.

[0119] The preceding text described the algorithm for converting dynamic response spectra into numerical feature vectors. To further clarify this invention, the following text will list and explain in detail the specific numerical features that can be extracted in this process.

[0120] To transform the raw spectral data into input that can be effectively utilized by machine learning models, the feature extraction module calculates a series of numerical parameters with clear physical meaning for each pixel's six spectral vectors. These parameters together constitute a high-dimensional feature vector. The extractable features are mainly divided into the following three categories:

[0121] I. Amplitude Spectrum Characteristics: These characteristics mainly describe the intensity, energy, and frequency selectivity of the response signal.

[0122] center frequency of resonance peak ( ): The frequency point where the maximum value appears in the amplitude spectrum. This frequency is directly related to the local physical properties of the pixel unit (such as mass, elastic modulus).

[0123] Resonance peak value ( ): The amplitude at the center frequency of the resonance peak. This value reflects the intensity of the pixel's response in the resonance state.

[0124] Full width at half maximum (FWHM): The frequency range where the amplitude drops to half the peak value (or -3 dB) on both sides of the resonance peak in the amplitude spectrum. This parameter characterizes the sharpness of the resonance peak.

[0125] Quality factor ( Value): via formula Calculated by FWHM. The resonance value is a dimensionless key indicator for measuring the energy dissipation or damping of a system; a sharp resonance peak corresponds to a high [value]. The value is extremely sensitive to areas with energy dissipation defects (such as microcracks).

[0126] Spectral integral value: The integral of the amplitude spectrum over the entire swept frequency range. This value represents the total energy absorbed or responded to by the pixel during the entire excitation process.

[0127] II. Phase Spectrum Characteristics: These characteristics mainly describe the time delay relationship between the response signal and the excitation signal.

[0128] Resonance point phase value ( ): The phase value of the phase spectrum at the center frequency of the resonance peak in the amplitude spectrum. This value reflects the inherent delay of the system at resonance.

[0129] Maximum phase change rate ( ): The maximum value of the phase spectrum slope, which usually occurs near the resonance frequency. This parameter is also related to the sharpness of the resonance.

[0130] Phase delay at a specific frequency point: The phase value read at one or more fixed frequency points (e.g., the start or end frequency) within the sweep frequency range. This feature can be used to evaluate the system response delay in non-resonant states.

[0131] III. Cross-modal coupling characteristics: These characteristics are used to quantify the interrelationships between responses of different physical fields and are crucial for identifying complex defects.

[0132] Resonance frequency difference: For example, the difference between the resonant frequency of the acoustoelastic wave response and the resonant frequency of the electric field response ( ),in, The resonant frequency representing the acoustic elastic wave response spectrum; The resonant frequency represents the electric field response spectrum.

[0133] Peak amplitude ratio: For example, the ratio between the peak amplitude of the thermal response and the peak amplitude of the acoustoelastic wave response at their respective resonance points ( ),in, The peak amplitude represents the thermal response spectrum; This represents the peak amplitude of the acoustic elastic wave response spectrum.

[0134] Phase difference at a specific frequency: For example, at the principal resonant frequency, the difference between the phase of the electric field response and the phase of the acoustoelastic wave response ( ),in, Represents the frequency at the center of the resonance peak. The phase of the electric field response; Represents the frequency at the center of the resonance peak. The phase of the acoustic elastic wave response.

[0135] To further illustrate the present invention, the following will provide a detailed description of the machine learning model for automatic defect diagnosis and classification using these high-dimensional feature vectors.

[0136] See attached document Figure 8 Within the main control and data analysis system 50, a "defect spectrum library" is pre-built and fixed. This library is constructed during an offline calibration phase. The calibration phase uses a series of LCD screen samples with known and verified typical defects using other detection methods (such as scanning electron microscopy), as well as several "gold samples" (i.e., defect-free, good products). The system performs a complete frequency sweep detection on these samples and extracts complete high-dimensional feature vectors for each pixel at a known defect location and for normal pixels. Finally, these high-dimensional feature vectors with clear category labels (e.g., "normal," "Mura defect," "pixel open circuit," "line short circuit," etc.) are stored, thus forming the defect spectrum library.

[0137] In the actual online detection process, after the system extracts the unique high-dimensional feature vector of a pixel to be tested, it calls a pre-trained multi-class classifier to analyze the feature vector.

[0138] One feasible implementation is to use a Support Vector Machine (SVM) classifier. During the offline calibration phase, the SVM model is trained using all labeled feature vectors from the defect spectrum library. The goal is to find the optimal hyperplane in the high-dimensional feature space that can separate data points of different defect categories with the maximum margin. After training, this classifier model is embedded in the main control and data analysis system 50. During online detection, the feature vector of the pixel to be detected is input into the trained SVM model. The model determines which region the feature vector falls into, defined by the hyperplane, and directly outputs a clear defect classification result.

[0139] Another approach is to use a deep neural network (DNN) classifier. A multilayer perceptron network is constructed, where the number of neurons in the input layer is the same as the dimension of the feature vector, and the number of neurons in the output layer corresponds to the total number of all known defect categories (including the "normal" category). During the calibration phase, the neural network is trained using data from a defect spectral library, and the network weights are adjusted through backpropagation, enabling the network to learn the complex nonlinear mapping relationship between the feature vector and the defect category. During online detection, the feature vector of the pixel to be detected is propagated forward through this trained network, and the category corresponding to the neuron with the highest activation value in the output layer is the final diagnostic result.

[0140] In addition to the classifier-based discrimination method mentioned above, the system can also employ a distance-based method. For example, it can calculate the Euclidean or Mahalanobis distance between the feature vector of the pixel to be tested and all feature vectors stored in the defect spectrum library, and then classify the pixel to be tested into the category of one or more nearest samples according to the nearest neighbor or K-Nearest Neighbor (k-NN) algorithm.

[0141] Through any of the above methods, the system can ultimately output a high-confidence classification label for each pixel, thereby achieving automated and high-precision root cause diagnosis of defects.

Claims

1. A sensor-based LCD liquid crystal panel defect detection system, characterized by, The application relates to a multi-physical field dynamic response spectrum testing system for LCD liquid crystal screens. The application comprises: a sweep frequency synchronous excitation module for generating a periodic excitation signal with a frequency scanning in a preset frequency set and two strictly co-frequency and co-phase digital reference signals of the excitation signal; a pixel addressing and driving module for receiving the periodic excitation signal and applying the excitation signal to target pixels of a to-be-tested LCD liquid crystal screen; an integrated multi-modal sensing assembly arranged in a non-contact mode above a working surface of the to-be-tested LCD liquid crystal screen for synchronously detecting a transient electric field response signal and an acoustic elastic wave response signal generated by the target pixels under the action of the excitation signal; a multi-channel synchronous data acquisition and phase-locked analysis unit for synchronously acquiring the transient electric field response signal and the acoustic elastic wave response signal, and performing digital phase-locked demodulation on the acquired signals by using the two co-frequency and co-phase digital reference signals to calculate in-phase components and quadrature components of the response signals; 2. The sensor-based LCD panel defect detection system of claim 1, wherein, a main control and data analysis system for controlling the sweep frequency synchronous excitation module to perform frequency scanning, receiving the in-phase components and quadrature components, constructing a multi-physical field dynamic response spectrum for the target pixels, and finally realizing defect detection by comparing and analyzing the multi-physical field dynamic response spectrum and a reference spectrum. The sweep frequency synchronous excitation module physically comprises: a direct digital frequency synthesizer, a high-stability reference clock source, a microcontroller unit, a low-pass filter, a programmable gain amplifier and a digital logic and level conversion circuit; 3. The sensor-based LCD panel defect detection system of claim 1, wherein, the direct digital frequency synthesizer generates a digital sine waveform data stream and two digital quadrature square wave signals, the data stream is processed by the low-pass filter and the programmable gain amplifier to form the periodic excitation signal, and the two digital quadrature square wave signals are processed by the digital logic and level conversion circuit to form the two co-frequency and co-phase digital reference signals. The integrated multi-modal sensing assembly is a multi-layer stacked composite structure, and the composite structure comprises, from bottom to top: an optical-grade transparent substrate; a transparent piezoelectric film array deposited on the upper surface of the optical-grade transparent substrate; a transparent insulating medium layer arranged above the transparent piezoelectric film array; a micro-transient electric field sensing array made on the transparent insulating medium layer; and 4. The sensor-based LCD panel defect detection system of claim 3, wherein, a grounded transparent electromagnetic shielding layer covering the top layer.

5. The sensor-based LCD panel defect detection system of claim 3, wherein, The micro-transient electric field sensing array is composed of a series of physically independent sensing electrodes, the size, shape and spatial position of each sensing electrode accurately correspond to one or a group of pixel units of the to-be-tested LCD liquid crystal screen, and the transient edge electric field leaked by the thin film transistor of the pixel unit in the working state is captured through the capacitive coupling effect.

6. The sensor-based LCD panel defect detection system of claim 1, wherein, The transparent piezoelectric film array is composed of a plurality of independent piezoelectric sensing units, and each piezoelectric sensing unit is used for converting the acoustic elastic wave response signal generated by the excited pixel due to the electrostrictive effect and the thermoelastic effect into an electric signal through the piezoelectric effect. The hardware structure of the multi-channel synchronous data acquisition and phase-locked analysis unit comprises: A multi-channel analog front-end, a high-speed synchronous analog-to-digital converter group, a field programmable gate array as an operation core, a high-speed data buffer, and a data interface for communicating with a host and data analysis system.

7. A sensor-based LCD liquid crystal panel defect detection system according to claim 6, wherein, The multi-channel analog front-end comprises programmable gain amplifiers and anti-aliasing low-pass filters configured independently for each input channel; the high-speed synchronous analog-to-digital converter group is responsible for converting the multi-channel analog signals processed by the multi-channel analog front-end into digital data streams and transmitting them to the field programmable gate array.

8. The sensor-based LCD panel defect detection system of claim 1, wherein, The host and data analysis system maintains a dedicated data structure for each pixel in memory, which includes six independent numerical arrays: an electric field response amplitude spectrum vector, an electric field response phase spectrum vector, an acoustic elastodynamic wave response amplitude spectrum vector, an acoustic elastodynamic wave response phase spectrum vector, a thermal response amplitude spectrum vector, and a thermal response phase spectrum vector; The six numerical arrays completely store the six dynamic response spectra covering the entire sweep range.

9. A sensor-based LCD liquid crystal panel defect detection system according to claim 8, wherein, The host and data analysis system is also used to execute a feature extraction algorithm to extract and combine a high-dimensional numerical feature vector from the six dynamic response spectra, the numerical feature vector including at least one cross-modal coupling feature selected from: the difference between the resonance frequency of the acoustic elastodynamic wave response and the resonance frequency of the electric field response, the ratio between the peak amplitude of the thermal response and the peak amplitude of the acoustic elastodynamic wave response, or the difference between the phase of the electric field response and the phase of the acoustic elastodynamic wave response at the main resonance frequency point.

10. A sensor-based LCD liquid crystal panel defect detection system according to claim 9, wherein, The host and data analysis system has a pre-constructed and solidified defect spectrum library inside, which stores a set of high-dimensional feature vectors with clear category labels; the host and data analysis system is also used to call a multi-class classifier trained using the defect spectrum library to analyze the high-dimensional numerical feature vector and output a clear defect classification result.

Images