LED Chip Processing Multi-Source Data Fusion Acquisition and Analysis System

Abstract

This invention relates to the field of data fusion technology, specifically to a multi-source data fusion acquisition and analysis system for LED chip processing. The system includes a heterogeneous signal instantaneous synchronization module, a thermal field optical deviation correction module, a mechanical vibration feature mapping module, a processing morphology feature analysis module, and a quality status fusion analysis module. In this invention, by monitoring the transient voltage feedback from a piezoelectric sensor and comparing it with a threshold to trigger real-time image acquisition, an instantaneous synchronization linkage mechanism for acoustic and optical signals is established. This captures the transient morphology features of processing nodes, extracts the platinum resistance thermometer temperature difference to dynamically correct the image plane geometric transformation matrix, offsets the imaging shift caused by thermal field distribution fluctuations, couples the radial runout amplitude of the main axis with the exposure time to establish a spatial displacement correlation, calculates the blur compensation factor and eliminates visual degradation interference, improves the contrast of morphology features and textures, correlates texture features with spatial point mapping, compares the quantization consistency deviation of the reference interval, and completes the processing quality assessment under multi-source parameter coupling.

Description

Technical Field

[0001] This invention relates to the field of data fusion technology, and in particular to a multi-source data fusion acquisition and analysis system for LED chip processing. Background Technology

[0002] The field of data fusion technology mainly involves a working system that performs time alignment and spatial registration of heterogeneous parameters from different levels of sensing nodes, and performs feature splicing and joint calculation according to preset mapping rules to comprehensively obtain multi-dimensional physical quantities of entity objects. The traditional LED chip processing multi-source data fusion acquisition and analysis system refers to the process of collecting and merging machine tool spindle operation parameters, chamber environment parameters, and wafer surface morphology parameters generated during the photolithography, etching, and dicing processes of light-emitting diode wafers. Typically, a programmable logic controller (PLC) reads the current value of the servo motor of the cutting machine and the pulse count of the spindle encoder at a set cycle. A thermocouple is used to measure the temperature value inside the processing chamber. A charge-coupled device (CCD) industrial camera captures a two-dimensional pixel array on the wafer surface. Then, the current value, pulse count, temperature value, and two-dimensional pixel array are transmitted to the industrial control host via a field communication cable. In the industrial control host, a weighted polynomial is used to perform multiplication and summation calculations on the current value, pulse count, and temperature value. A Sobel operator is used to perform matrix convolution calculations on the gray values ​​of the edge pixels in the two-dimensional pixel array. Finally, the summation value and the gray value obtained from the matrix convolution are entered line by line into a relational database table based on timestamps.

[0003] In existing technologies, programmable logic controllers (PLCs) read data based on preset cycles, ignoring high-frequency transient physical pulses during the processing. When faced with sudden processing anomalies, the acquisition response is delayed. Single thermocouple temperature measurement combined with weighted calculation ignores the interference of thermal field refraction on the optical path, resulting in irreversible geometric deformation of the wafer surface imaging. Since the impact of high-speed radial runout of the spindle on pixel sampling is not considered, the original image is accompanied by severe motion blur and feature loss. Heterogeneous data is only entered into the table through time stamps, and the lack of spatial coordinate mapping leads to a break in the physical connection between parameters and physical points. It cannot reflect the consistency difference between micro-processing texture and quality benchmark, causing the processing quality assessment results to deviate from the true state due to the lack of spatial constraints. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a multi-source data fusion acquisition and analysis system for LED chip processing.

[0005] On the one hand, a multi-source data fusion acquisition and analysis system for LED chip processing is provided, which includes: The heterogeneous signal instantaneous synchronization module detects the transient voltage signal fed back by the piezoelectric sensor, compares the transient voltage signal with the noise trigger voltage threshold, and triggers the image sensor to acquire data when the threshold is exceeded, thus obtaining an audio-visual synchronized pixel matrix. The thermal field optical deviation correction module reads the temperature value of the platinum resistance thermometer and extracts the temperature difference change. Combined with the acousto-optic synchronized pixel matrix, it corrects the image plane geometric transformation matrix according to the temperature difference mapping to obtain the thermally induced geometric correction coordinates. The mechanical vibration feature mapping module monitors the radial runout amplitude of the spindle, combines the thermally induced geometric correction coordinates, matches the exposure time and rotational frequency, extracts the imaging displacement deviation and associates it with the spatial offset orientation, and obtains the vibration motion blur compensation factor. The processing morphology feature analysis module acquires image pixel features and extracts grayscale distribution extreme value features. Based on the vibration motion blur compensation factor, it corrects visual degradation components and obtains static feature texture contrast. The quality status fusion analysis module, based on the static feature texture contrast, retrieves the spatial point information corresponding to the thermally induced geometric correction coordinates, associates the texture features with the spatial point mapping, compares the quality benchmark interval and quantifies the consistency deviation, and obtains the multi-source fusion processing quality evaluation result.

[0006] As a further aspect of the present invention, the acoustic-optical synchronization pixel matrix includes a synchronization frame number, pixel grayscale, trigger time stamp, and array index; the thermally induced geometric correction coordinates include lateral correction coordinates, longitudinal correction coordinates, thermal drift calibration terms, and image plane geometric residuals; the vibration motion blur compensation factor includes radial blur weight, tangential trailing weight, displacement sensitivity coefficient, and phase correction coefficient; the static feature texture contrast includes texture peak and valley amplitude, edge sharpness, grayscale level, and local contrast coefficient; and the multi-source fusion processing quality evaluation result includes consistency deviation, processing offset, defect risk category, and judgment identifier.

[0007] As a further aspect of the present invention, the image plane geometric transformation matrix refers to the coordinate transformation matrix that performs geometric correction on the pixel coordinates of the original image based on thermal drift and refraction offset. The imaging displacement deviation refers to the offset of the image target position caused by the combined effects of spindle radial runout, rotational frequency, and exposure time at the moment of exposure acquisition.

[0008] As a further aspect of the present invention, the visual degradation component refers to the image contrast reduction, ghosting, and feature blur caused by non-processing factors such as vibration blur, thermal drift, and imaging offset. The quantification consistency deviation refers to the degree of numerical deviation obtained by mapping texture features to spatial points and comparing them with a preset processing quality benchmark range.

[0009] As a further aspect of the present invention, the heterogeneous signal instantaneous synchronization module includes: The voltage acquisition submodule detects the transient voltage signal fed back by the piezoelectric sensor at the LED epitaxial wafer dicing station, reads the potential of continuous sampling points according to the sensing sampling clock, and performs normalization processing based on the potential change amplitude of adjacent sampling points to obtain the transient voltage amplitude. The noise discrimination submodule extracts the noise trigger voltage reference based on the transient voltage amplitude, converts the transient voltage amplitude into a normalized amplitude, and compares the normalized amplitude with a preset threshold reference to obtain the noise exceeding the limit criterion. The pixel acquisition submodule calls the noise limit criterion, maps the positive criterion interval to charge transfer gating timing, calibrates the timing start point according to the sensor sampling clock, drives the image sensor to perform charge transfer acquisition, receives the pixel charge in the exposure window and maps it to grayscale data according to row and column addresses, and obtains the audio-visual synchronized pixel matrix.

[0010] As a further aspect of the present invention, the thermal field optical deviation correction module includes: The temperature acquisition submodule reads the real-time temperature reading from the platinum resistance sensor on the inner wall of the objective lens barrel in the LED scribing machine. Combined with the aforementioned acoustic-optical synchronized pixel matrix, the real-time temperature reading is compared with the ambient reference temperature reading point by point according to the sampling clock to obtain the temperature deviation range. The refractive mapping submodule calls the temperature deviation range, associates it with the thermal coefficient of the refractive index of the optical glass, and derives the thermal drift state of the lens barrel according to the positive correspondence between the temperature deviation range and the thermal coefficient of the refractive index to obtain the refractive offset coefficient. The coordinate correction submodule maps the image plane geometric transformation matrix according to the refraction offset coefficient, reverses the acousto-optic synchronized pixel matrix coordinates, and synchronously writes the horizontal and vertical coordinates into the matrix row and column addresses to obtain the thermally induced geometric correction coordinates.

[0011] As a further aspect of the present invention, the mechanical vibration feature mapping module includes: The amplitude monitoring submodule monitors the radial runout amplitude fed back by the diamond dicing cutter spindle accelerometer. Combined with the thermally induced geometric correction coordinates, it reads the radial acceleration waveform according to the spindle sampling clock and performs dimensional normalization processing on the amplitude of the peak and trough intervals to obtain the radial runout amplitude spectrum. The displacement definition submodule calls the radial runout amplitude spectrum, matches the exposure duration parameter and the rotation speed frequency parameter, and derives the instantaneous displacement deviation of imaging according to the correspondence between the exposure window span and the rotation speed period, thus obtaining the imaging displacement deviation trajectory. The blur compensation submodule, based on the imaging displacement deviation trajectory, associates the spatial orientation offset information under the thermally induced geometric correction coordinates, maps the horizontal and vertical components of the coordinates to the exposure direction respectively, and obtains the vibration motion blur compensation factor.

[0012] As a further aspect of the present invention, the processing morphology feature analysis module includes: The extreme value extraction submodule extracts the extreme value features of the gray value distribution of the pixel set based on the vibration motion fuzzy compensation factor, reads the gray value peak reading and gray value valley reading according to the pixel row and column address, compares the gray value span with the gray value discrete sequence of the pixel neighborhood, and obtains the gray value extreme value spectrum. The benchmark matching submodule calls the gray-level extreme value spectrum, matches the contrast benchmark value, and converts the gray-level extreme value spectrum and the contrast benchmark value according to the proportional relationship to obtain the contrast offset domain. The degradation correction submodule, based on the contrast offset domain, applies a vibration motion blur compensation factor to correct the visual degradation component in the contrast reference value, and subtracts the degradation ratio according to the compensation factor and the contrast offset domain to obtain the static feature texture contrast.

[0013] As a further aspect of the present invention, the quality status fusion analysis module includes: The point retrieval submodule retrieves spatial point information corresponding to the thermally induced geometric correction coordinates based on the static feature texture contrast, processes the trajectory sampling position according to the coordinate row and column address index, compares the texture contrast readings of adjacent sampling positions, and obtains a spatial point index set. The trajectory mapping submodule calls the spatial point index set, associates texture features with the processing trajectory dimension and maps spatial points, and matches the texture comparison readings with the point coordinate sequence according to the trajectory sampling order to obtain the texture point mapping matrix; The quality judgment submodule compares the texture point mapping matrix with the preset processing quality benchmark range, makes a quantitative judgment on the processing consistency deviation, assigns a quality status code according to the position of the deviation falling into the benchmark range, and obtains the multi-source fusion processing quality evaluation result.

[0014] As a further aspect of the present invention, during the process of comparing the texture contrast readings of adjacent sampling positions, the static feature texture contrast carried by two adjacent sampling points in the processing trajectory sampling position is obtained, the static feature texture contrast is subjected to differential subtraction calculation and the absolute value deviation is extracted, and when the absolute value deviation reaches the preset texture mutation limit value, the two adjacent sampling points are stored in the spatial point index set. In the process of quantifying the processing consistency deviation, the spatial geometric straight-line distance between each coordinate point in the texture point mapping matrix and the center position of the preset processing quality benchmark interval is measured. The spatial geometric straight-line distance is defined as the processing consistency deviation. The processing consistency deviation is compared with the coverage area of ​​the preset processing quality benchmark interval. The target position of the processing consistency deviation falling into the preset processing quality benchmark interval is mapped to the corresponding quality status code to obtain the multi-source fusion processing quality evaluation result.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by monitoring the transient voltage feedback from the piezoelectric sensor and comparing it with the threshold to trigger real-time image acquisition, a synchronous linkage mechanism for acoustic and optical signals is established to capture the transient morphological features of the processing nodes. The dynamic correction image plane geometric transformation matrix of the platinum resistance temperature difference is extracted to offset the imaging shift caused by thermal field distribution fluctuations. The spatial displacement correlation is established by coupling the radial runout amplitude of the main axis with the exposure time. The blur compensation factor is calculated and visual degradation interference is eliminated to improve the contrast of morphological features and texture. The texture features are correlated with the spatial point mapping. The quantization consistency deviation of the reference interval is compared to complete the processing quality assessment under multi-source parameter coupling. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a flowchart of the heterogeneous signal instantaneous synchronization module in this invention; Figure 4 This is a flowchart of the thermal field optical deviation correction module in this invention; Figure 5 This is a flowchart of the mechanical vibration feature mapping module in this invention; Figure 6 This is a flowchart of the processing morphology feature analysis module in this invention; Figure 7 This is a flowchart of the quality status fusion analysis module in this invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] This invention provides a multi-source data fusion acquisition and analysis system for LED chip processing, such as... Figure 1-2 The diagram shown illustrates a multi-source data fusion acquisition and analysis system for LED chip processing. This system includes: The heterogeneous signal instantaneous synchronization module detects the transient voltage signal fed back by the piezoelectric sensor at the LED epitaxial wafer dicing station, compares the transient voltage signal with the noise trigger voltage threshold, and triggers the image sensor to perform charge transfer acquisition when the transient voltage signal exceeds the noise trigger voltage threshold, thereby obtaining an acoustic-optical synchronized pixel matrix. The thermal field optical deviation correction module, based on the acousto-optic synchronized pixel matrix, reads the real-time temperature value fed back by the platinum resistance sensor on the inner wall of the objective lens barrel in the LED scribing machine, extracts the deviation range between the real-time temperature value and the ambient reference temperature value, correlates the thermal coefficient of the refractive index of the optical glass, maps the refractive index offset to the image plane geometric transformation matrix, and reverses the coordinates of the acousto-optic synchronized pixel matrix to obtain the thermally induced geometric correction coordinates. The mechanical vibration feature mapping module monitors the radial runout amplitude fed back by the accelerometer of the diamond dicing cutter spindle. Combined with the thermally induced geometric correction coordinates, it matches the exposure time parameter and the rotation speed frequency parameter, defines the physical displacement deviation at the moment of imaging, and associates the spatial orientation offset information under the thermally induced geometric correction coordinates to obtain the vibration motion fuzziness compensation factor. The processing morphology feature analysis module extracts the extreme value features of the gray-level distribution of the pixel set based on the vibration motion blur compensation factor. It matches the gray-level distribution features with the contrast benchmark value and applies the vibration motion blur compensation factor to correct the visual degradation components in the contrast benchmark value to obtain the static feature texture contrast. The quality status fusion analysis module, based on static feature texture contrast, retrieves the spatial point information corresponding to the thermally induced geometric correction coordinates, associates texture features with spatial point mapping in the processing trajectory dimension, compares with the preset processing quality benchmark range, makes a quantitative judgment on the processing consistency deviation, and obtains the multi-source fusion processing quality evaluation result.

[0021] The audio-visual synchronization pixel matrix includes synchronization frame number, pixel grayscale, trigger time stamp, and array index; the thermally induced geometric correction coordinates include lateral correction coordinates, longitudinal correction coordinates, thermal drift calibration terms, and image plane geometric residuals; the vibration motion blur compensation factor includes radial blur weight, tangential trailing weight, displacement sensitivity coefficient, and phase correction coefficient; the static feature texture contrast includes texture peak and valley amplitude, edge sharpness, grayscale level, and local contrast coefficient; and the multi-source fusion processing quality assessment results include consistency deviation, processing offset, defect risk category, and judgment mark.

[0022] Specifically, such as Figure 2 , 3 As shown, the heterogeneous signal instantaneous synchronization module includes: The voltage acquisition submodule detects the transient voltage signal fed back by the piezoelectric sensor at the LED epitaxial wafer dicing station, reads the potential of continuous sampling points according to the sensing sampling clock, and performs normalization processing based on the potential change amplitude of adjacent sampling points to obtain the transient voltage amplitude. First, the voltage acquisition submodule calls the analog signal interface at the output of the piezoelectric sensor on the LED chip dicing machine to capture the electrical signal generated in real time at the moment of pressure applied at the epitaxial wafer dicing station. To achieve high-precision signal capture, its internal high-speed analog-to-digital converter and sample-and-hold circuit are used to sample the induced voltage on a continuous time axis at equal intervals according to a 500 MHz sensing sampling clock frequency. During this process, the weak piezoelectric response is converted into a discrete sequence of potential values. The module then extracts the potential readings from two adjacent sampling points and determines the transient fluctuation intensity of the potential by calculating the difference between the two values. To improve the accuracy of the acquisition, the voltage acquisition submodule incorporates a high-performance data acquisition control component, which takes over the underlying bus in real time through a hardware interrupt triggering mechanism. This ensures that the data acquisition control component can activate the voltage acquisition link with a nanosecond-level response speed in the extremely short time the dicing head touches the epitaxial wafer, enabling rigorous monitoring of the integrity of the piezoelectric signal waveform. To eliminate the influence of individual sensor differences and environmental noise on signal strength, normalization logic is executed, dividing the currently calculated potential change amplitude by a preset range reference voltage value. For example, when the potentials of adjacent sampling points fed back by the sensor are 120 mV and 185 mV respectively, the change amplitude is 65 mV. If the preset range reference voltage is 1000 mV, the ratio of the two is calculated, and the result is 0.065, thus obtaining the transient voltage amplitude.

[0023] The noise discrimination submodule extracts the noise trigger voltage reference based on the transient voltage amplitude, converts the transient voltage amplitude into a normalized amplitude, and compares the normalized amplitude with a preset threshold reference to obtain the noise exceeding the limit criterion. First, after acquiring the transient voltage amplitude, the noise discrimination submodule retrieves the average voltage value of the previous 500 stable operating conditions from the historical sampling sequence and defines it as the noise trigger voltage reference. Then, using its internal comparator unit, it scales the real-time input transient voltage amplitude proportionally to this reference, converting it into a normalized amplitude with uniform dimensions. During this process, the module calls a preset threshold reference value, typically set between 0.15 and 0.25, specifically determined through multiple dicing experiments. Experimental data shows that when the dicing tool contacts the edge of the epitaxial wafer, generating high-frequency fragmentation noise, the normalized amplitude significantly exceeds the critical value of 0.20. The data acquisition control logic performs bidirectional feedback adjustment here. Once the noise discrimination submodule detects an abnormal normalized amplitude, the data acquisition control component immediately adjusts the acquisition gain step, dynamically adapting to the signal-to-noise ratio range of the signal, thereby improving the robustness of the discrimination results. For example, if the calculated normalized amplitude of the currently input transient voltage is 0.28, it is logically compared with the preset threshold reference value of 0.20. Since 0.28 is greater than 0.20, the current signal is determined to have triggered a noise limit violation condition. This logic simplifies complex waveform analysis into a scalar comparison, ultimately yielding the noise limit violation criterion.

[0024] The pixel acquisition submodule calls the noise limit criterion, maps the positive criterion interval to the charge transfer gating timing, calibrates the timing start point according to the sensor sampling clock, drives the image sensor to perform charge transfer acquisition, receives the pixel charge in the exposure window and maps it to grayscale data according to the row and column address, and obtains the audio-visual synchronized pixel matrix. First, upon receiving the noise exceeding the limit criterion, the pixel acquisition submodule immediately activates its internal timing generator. Based on the received positive criterion interval—the duration of the noise exceeding the limit—it converts it into a charge transfer gating signal for the image sensor. This signal determines the on / off timing of the photosensitive element transferring photoelectrons to the register. To ensure strict spatial correspondence between acoustic wave acquisition and optical imaging, the module calls the phase information of the sensor sampling clock to calibrate the rising edge trigger point of the gating timing. Subsequently, a drive command is sent to the image sensor's control bus to execute the charge transfer acquisition action. Under the action of the gating pulse, the photogenerated charges in the image sensor's photosensitive array are read row by row according to their column addresses and mapped to grayscale data between levels 0 and 255 via an analog-to-digital converter. During this stage, the data acquisition control unit precisely locks the image acquisition board using a synchronization pulse and performs flow shaping on the pixel data stream according to preset charge transfer depth parameters to ensure that data overflow or packet loss does not occur under high-speed pulse triggering. For example, when the dicing tool passes through a specific coordinate, the noise discrimination submodule outputs a high-level pulse. The pulse was captured to start at 12.55 microseconds, and an exposure window was opened from this point for 50 microseconds. Finally, a grayscale matrix of size 512*512 pixels was obtained, resulting in the sound and light synchronized pixel matrix.

[0025] Specifically, such as Figure 2 , 4 As shown, the thermal field optical deviation correction module includes: The temperature acquisition submodule reads the real-time temperature reading from the platinum resistance sensor on the inner wall of the objective lens barrel in the LED scribing machine. Combined with the audio-visual synchronized pixel matrix, the real-time temperature reading is compared with the ambient reference temperature reading point by point according to the sampling clock to obtain the temperature deviation range. First, during operation, the temperature acquisition submodule polls the four-wire platinum resistance sensor mounted on the inner wall of the objective lens barrel in real time via the industrial bus interface. This sensor converts temperature fluctuations inside the objective lens into changes in resistance, and then converts these changes back into a real-time temperature reading in degrees Celsius based on a standard resistance-to-temperature conversion characteristic curve. Simultaneously, the module reads the ambient reference temperature reading from memory, which is typically provided by an ambient sensor during the initial stable state of the equipment. Within each sampling clock cycle, the module calculates the difference between the real-time temperature reading and the ambient reference temperature reading. The data acquisition control protocol dynamically weights the temperature sensor reading frequency. When an increase in processing load is detected, the sampling polling interval is automatically shortened to ensure higher temporal resolution of the heat acquisition data, and data cleaning is performed on the acquired raw temperature stream. For example, if the real-time temperature reading is 24.85 degrees Celsius and the ambient reference temperature is 22.10 degrees Celsius, the difference between the two is 2.75 degrees Celsius. This difference is then compared with a preset fluctuation threshold range. If the difference continues to fluctuate between 2.5 degrees Celsius and 3.0 degrees Celsius, the current temperature deviation range is determined, and the final temperature deviation range is obtained.

[0026] The refraction mapping submodule calls the temperature deviation range, associates it with the thermal coefficient of the refractive index of the optical glass, and derives the thermal drift state of the lens barrel according to the positive correspondence between the temperature deviation range and the thermal coefficient of the refractive index, thus obtaining the refractive offset coefficient. First, after obtaining the temperature deviation range, the refractive mapping submodule retrieves the refractive index thermal coefficient of the corresponding optical glass from the optical material database. This coefficient describes the change in the glass's refractive index with each 1-degree Celsius temperature change. The module executes mapping logic, multiplying the temperature deviation range obtained in the previous steps with this thermal coefficient. For example, if the refractive index thermal coefficient of the optical glass is known to be 0.000005 per degree Celsius, and the median value of the current temperature deviation range is 2.75 degrees Celsius, multiplying the two yields the actual variable of the refractive index. Subsequently, based on the geometric parameters of the lens group, this variable is substituted into the lens barrel thermal drift state evaluation model to derive the changes in lens focal length and image position shift. This derivation logic is calibrated online using real-time thermal parameters output from the data acquisition control layer, coupling static physical property parameters with dynamically acquired lens barrel deformation data, improving the model's prediction accuracy under different temperature rise gradients. The final calculated result is defined as the refractive shift coefficient. This coefficient is a dimensionless correction factor used to quantify the degree of influence of thermal effects on the deflection of the imaging optical path. For example, the calculated refractive offset coefficient is 1.00001375, which represents a slight outward expansion and offset of the optical path due to the temperature rise, and finally the refractive offset coefficient is obtained.

[0027] The coordinate correction submodule maps the image plane geometric transformation matrix according to the refraction offset coefficient, reverses the acousto-optic synchronized pixel matrix coordinates, and synchronously writes the horizontal and vertical coordinates into the matrix row and column addresses to obtain the thermally induced geometric correction coordinates. First, the coordinate correction submodule uses the acquired refraction offset coefficient to geometrically reconstruct the original image coordinate system, thereby constructing a 3x3 image plane geometric transformation matrix. The diagonal elements of this matrix are composed of the reciprocals of the refraction offset coefficient, used to perform inverse scaling to compensate for pixel position shifts caused by thermal expansion and contraction. Subsequently, the module iterates through each original coordinate point in the acousto-optic synchronized pixel matrix, performing matrix multiplication operations on its horizontal and vertical coordinates with the transformation matrix. To ensure that the corrected coordinate data can be instantly accessed by subsequent algorithms, a dedicated high-bandwidth storage channel is established on the data acquisition and control bus. The results of the transformation matrix calculation are timestamped with the physical displacement data acquired by the sensor, achieving seamless stitching of image coordinates and physical positions. For example, the original horizontal coordinate of a feature point is 256, the vertical coordinate is 256, and the refraction offset coefficient is 1.00013. Through inverse scaling calculation, dividing the original coordinates by the offset coefficient yields a corrected horizontal coordinate of approximately 255.967 and a vertical coordinate of approximately 255.967. These new coordinates, corrected by floating-point operations, are rewritten into the row and column addresses of the matrix to form a new matrix that is perfectly aligned with the theoretical cutting trajectory in geometric space, thus obtaining the thermally induced geometrically corrected coordinates.

[0028] Specifically, such as Figure 2 , 5 As shown, the mechanical vibration feature mapping module includes: The amplitude monitoring submodule monitors the radial runout amplitude fed back by the diamond dicing cutter spindle accelerometer. Combined with the thermally induced geometric correction coordinates, it reads the radial acceleration waveform according to the spindle sampling clock and performs dimensional normalization processing on the amplitude of the peak and trough intervals to obtain the radial runout amplitude spectrum. First, the amplitude monitoring submodule captures the radial runout signal in real time through an accelerometer on the spindle bearing housing, reading the voltage waveform of the radial acceleration according to the spindle sampling clock. To convert the acceleration signal into an amplitude index that can intuitively measure displacement runout, dimensional normalization is performed. This process extracts the difference between each peak value and the adjacent trough value in the waveform sequence, and calculates the difference by combining it with the sensor's sensitivity coefficient. The data acquisition and control architecture allocates a dedicated differential acquisition channel for the acceleration signal, using a hardware-level low-pass filter to remove high-frequency electromagnetic harmonics, ensuring that the monitored vibration data truly reflects the radial runout of the diamond dicing cutter spindle. For example, the peak acceleration fed back by the accelerometer is 0.8 gravitational acceleration, the trough value is -0.7 gravitational acceleration, and its peak-to-peak value is 1.5 gravitational acceleration. If the sensitivity coefficient corresponds to 10 micrometers of displacement per gravitational acceleration, the initial runout amount is obtained. These amplitudes at different frequency components are aggregated and statistically analyzed to form an energy distribution map with frequency as the horizontal axis and amplitude as the vertical axis. This spectral line can clearly show the asynchronous runout energy distribution of the main shaft when it rotates at high speed, and finally obtains the radial runout amplitude spectrum.

[0029] The displacement definition submodule calls the radial runout amplitude spectrum, matches the exposure duration parameter and the rotation frequency parameter, and derives the instantaneous displacement deviation of imaging according to the correspondence between the exposure window span and the rotation period, thus obtaining the imaging displacement deviation trajectory. First, the displacement definition submodule, after acquiring the radial runout amplitude spectrum, calls the exposure time parameters of the image sensor and the spindle rotation frequency parameters. By executing the corresponding logic, it determines the proportion of the spindle rotation period within a single exposure window. The module couples the main vibration frequencies in the amplitude spectrum with the rotation period for analysis, calculating the instantaneous displacement trajectory of the scriber tip relative to the epitaxial wafer surface within the extremely short time from exposure start to exposure end. In this stage, the data acquisition control logic enforces a multi-sensor synchronization protocol, hard-wired and synchronizing the electronic shutter signal of the image sensor with the master clock of the vibration acquisition card, thereby locking the sampling error of the displacement trajectory to the sub-microsecond level. For example, if the exposure time is 50 microseconds and the spindle speed is 60,000 revolutions per minute, that is, each revolution takes 1,000 microseconds, the exposure time accounts for only 5% of the rotation period. Based on the acceleration vector integral in the amplitude spectrum within this time scale, the minute displacement deviation at the instant of imaging is derived. If it is calculated that within the exposure window, the blade tip moves 0.2 micrometers laterally and 0.1 micrometers longitudinally, the final imaging displacement deviation trajectory can be obtained.

[0030] The blur compensation submodule, based on the imaging displacement deviation trajectory, associates the spatial orientation offset information under the thermally induced geometric correction coordinates, maps the horizontal and vertical components of the coordinates to the exposure direction respectively, and obtains the vibration motion blur compensation factor. First, the blur compensation submodule further correlates the spatial orientation offset information under the thermally induced geometric correction coordinates based on the imaging displacement deviation trajectory. It performs projection operations to decompose the motion vector of the displacement deviation trajectory in three-dimensional space and map it onto the pixel plane of the current image sensor. Then, the module reverse-engineers the convolution calculation logic of the lateral and longitudinal components of the coordinates with the exposure direction, aiming to determine the charge diffusion range of each pixel due to motion during exposure. A vibration motion blur compensation factor is introduced during the calculation. The data acquisition control logic injects parameters into the convolution kernel of the image preprocessor through the instruction set, dynamically loading the compensation template based on the real-time feedback amplitude variable to ensure that the compensation factor maintains real-time mapping accuracy even when the tool speed fluctuates. For example, if the pixel size is 3.5 micrometers and the lateral displacement deviation is 0.7 micrometers, the lateral compensation factor is 0.2. This factor acts as an inverse filtering operator to quantify the proportion of grayscale leakage between adjacent pixels, ultimately yielding the vibration motion blur compensation factor.

[0031] Specifically, such as Figure 2 , 6 As shown, the processing morphology feature analysis module includes: The extreme value extraction submodule extracts the extreme value features of the gray value distribution of the pixel set based on the vibration motion fuzzy compensation factor. It reads the gray value peak and valley readings according to the pixel row and column address, compares the gray value span with the gray value discrete sequence of the pixel neighborhood, and obtains the gray value extreme value spectrum. First, the extreme value extraction submodule calls the generated vibration motion blur compensation factor to perform deep feature mining on the gray-level distribution of the pixel set. It traverses the thermally geometrically corrected pixel matrix and reads the maximum and minimum gray-level values ​​within each local region by row and column address. To eliminate gray-level equalization caused by motion blur, these extreme values ​​are enhanced using the compensation factor. This involves statistically comparing the gray-level span (the difference between peak and valley readings) with the discrete gray-level sequence within the pixel's neighborhood. The data acquisition control logic is responsible for reading the pixel buffer in blocks and performing streaming enhancement on the gray-level bitstream of sensitive areas using the compensation factor. This allows the extreme value extraction process to bypass low-contrast noise interference and improve the contrast resolution of feature points. For example, within a 3x3 pixel window, the highest gray-level value is 210, the lowest is 45, and the gray-level span is 165. The average neighborhood dispersion of this region is 12. The compensation factor corrects these extreme values ​​and eliminates false extreme points caused by vibration. All gray-level extremes that meet the saliency characteristics are arranged according to spatial frequency to form a data map reflecting the local contrast limit distribution of the image, and finally the gray-level extreme value spectrum is obtained.

[0032] The benchmark matching submodule calls the gray-level extreme value spectrum to match the contrast benchmark value, and converts the gray-level extreme value spectrum and the contrast benchmark value according to the proportional relationship to obtain the contrast offset domain. First, the benchmark matching submodule retrieves the grayscale extreme value spectrum generated in the previous steps and matches a contrast benchmark value from a preset process parameter library. This benchmark value is typically set to a fixed range, such as 180 to 220 grayscale levels. The module executes conversion logic, performing a proportional division operation between each energy peak in the real-time calculated grayscale extreme value spectrum and the contrast benchmark value to obtain a set of coefficients reflecting the current degree of contrast attenuation or enhancement. The data acquisition and control component acts as a data gateway here, performing cross-correlation analysis between the preset process contrast threshold and the real-time ambient light sensor data to ensure that the benchmark matching logic can adapt to fluctuations in light intensity under different external environments. For example, if the average span of the current extreme value spectrum is 150, and the contrast benchmark value is 200, then the ratio value is 0.75. This value intuitively depicts the degree to which the imaging quality deviates from the ideal state due to vibration, temperature rise, or tool wear. The ratio values ​​across the entire image range are aggregated and processed to identify the deviation areas in the contrast distribution, ultimately obtaining the contrast offset domain.

[0033] The degradation correction submodule, based on the contrast offset domain, applies the vibration motion blur compensation factor to correct the visual degradation component in the contrast baseline value, and subtracts the degradation ratio according to the compensation factor and the contrast offset domain to obtain the static feature texture contrast. First, losses due to non-processing factors such as equipment vibration and thermal drift are deducted from the overall contrast offset. The module executes a combination of subtraction and multiplication logic, subtracting the degradation ratio according to the compensation factor and the contrast offset domain. The data acquisition and control algorithm monitors this correction logic throughout its entire lifecycle. By modeling the acquired multi-source interference noise, the feedback link of the correction coefficient is connected to the image reconstruction engine to ensure the scientific validity of the degradation reduction ratio and the reproducibility of experimental data. For example, if the contrast offset domain shows a total loss ratio of 25%, and the compensation factor calculates that degradation due to spindle radial runout accounts for 10% and thermal drift accounts for 5%, then this 15% interference is subtracted from the total offset. The remaining value is defined as the static feature texture contrast. This result represents the true physical contrast attribute of the epitaxial wafer surface scribing marks after excluding all dynamic environmental interference, ensuring that the judgment is based solely on the processing process itself, ultimately yielding the static feature texture contrast.

[0034] Specifically, such as Figure 2 , 7 As shown, the quality status fusion analysis module includes: The point retrieval submodule retrieves spatial point information corresponding to thermally induced geometric correction coordinates based on static feature texture contrast, processes trajectory sampling positions according to coordinate row and column address index, compares texture contrast readings of adjacent sampling positions, and obtains a spatial point index set. First, the processing trajectory sampling positions are indexed by coordinate row and column addresses, and a physical location is retrieved from the coordinate matrix every 5 micrometers according to a preset processing step size. By comparing the texture contrast readings of adjacent sampling positions, boundary points where contrast changes abruptly or coordinate points with abnormal fluctuations are identified. The data acquisition control layer assigns a globally unified timestamp to each texture contrast data item by synchronously reading the encoder values, establishing a precise indexing logic based on location services to ensure that the geometric accuracy of the sampling points is not lost during dynamic acquisition. For example, at 10.55 mm on the coordinate axis, if the texture contrast drops sharply from 0.90 to 0.50, this coordinate point and its corresponding contrast change are recorded as an anomaly index. This process achieves precise backtracking from image features to physical processing coordinates, enabling each quality defect to find its corresponding spatial location on the actual epitaxial wafer surface, forming a mapping chain between features and locations, and ultimately obtaining a set of spatial point indexes.

[0035] The trajectory mapping submodule calls the spatial point index set, associates texture features with the processing trajectory dimension and maps spatial points, and matches the texture comparison readings with the point coordinate sequence according to the trajectory sampling order to obtain the texture point mapping matrix; First, the module fills in the sampling gaps using interpolation logic to ensure that the texture information is continuous and complete along the entire dicing path. The data acquisition control protocol performs conflict checks on the multi-threaded concurrent data streams during the trajectory mapping process. By establishing a trajectory buffer pool, it sorts and smooths massive spatial points to ensure the topological consistency of the texture mapping results on the macroscopic trajectory. For example, a 50mm long dicing line is decomposed into 10,000 data nodes, each containing an x-coordinate, y-coordinate, and texture contrast value. This multidimensional matrix fully displays the quality fluctuation curve during processing, realizing an up-dimensional mapping from discrete pixels to a continuous processing path, ultimately yielding a texture point mapping matrix.

[0036] The quality judgment submodule compares the preset processing quality benchmark range with the texture point mapping matrix, makes a quantitative judgment on the processing consistency deviation, assigns a quality status code according to the position of the deviation falling into the benchmark range, and obtains the multi-source fusion processing quality evaluation result. First, the quality judgment submodule compares the texture point mapping matrix with a preset processing quality benchmark range. This range is set according to process requirements; for example, the texture contrast of a qualified product must be between 0.85 and 1.00. The module traverses all nodes in the mapping matrix, quantifies the processing consistency deviation, and assigns a quality status code based on the deviation falling within the benchmark range. Finally, the data acquisition and control logic physically interlocks the final processing result status code with the dicing machine's shutdown control logic by verifying the integrity of the acquired parameters throughout the process. Once a continuous quality status anomaly is detected, an emergency stop command is immediately sent to the actuator via the data acquisition and control link. For example, if the benchmark range requires a texture contrast of no less than 0.75, and the average value of a certain area in the matrix is ​​0.60, the calculated deviation is 0.15. Status code 1 is assigned for deviations within 0.05, representing a minor defect, while status code 2 is assigned for deviations exceeding 0.10, representing a serious non-conformity. The module summarizes the status distribution of the entire trajectory and statistically analyzes the proportion of qualified points to obtain a quantitative assessment of processing accuracy and consistency, ultimately yielding a multi-source fusion processing quality assessment result.

[0037] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-source data fusion acquisition and analysis system for LED chip processing, characterized in that, include: The heterogeneous signal instantaneous synchronization module detects the transient voltage signal fed back by the piezoelectric sensor, compares the transient voltage signal with the noise trigger voltage threshold, and triggers the image sensor to acquire data when the threshold is exceeded, thus obtaining an audio-visual synchronized pixel matrix. The thermal field optical deviation correction module reads the temperature value of the platinum resistance thermometer and extracts the temperature difference change. Combined with the acousto-optic synchronized pixel matrix, it corrects the image plane geometric transformation matrix according to the temperature difference mapping to obtain the thermally induced geometric correction coordinates. The mechanical vibration feature mapping module monitors the radial runout amplitude of the spindle, combines the thermally induced geometric correction coordinates, matches the exposure time and rotational frequency, extracts the imaging displacement deviation and associates it with the spatial offset orientation, and obtains the vibration motion blur compensation factor. The processing morphology feature analysis module acquires image pixel features and extracts grayscale distribution extreme value features. Based on the vibration motion blur compensation factor, it corrects visual degradation components and obtains static feature texture contrast. The quality status fusion analysis module, based on the static feature texture contrast, retrieves the spatial point information corresponding to the thermally induced geometric correction coordinates, associates the texture features with the spatial point mapping, compares the quality benchmark interval and quantifies the consistency deviation, and obtains the multi-source fusion processing quality evaluation result.

2. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that: The audio-visual synchronization pixel matrix includes synchronization frame number, pixel grayscale, trigger time stamp, and array index; the thermally induced geometric correction coordinates include horizontal correction coordinates, vertical correction coordinates, thermal drift calibration terms, and image plane geometric residuals; the vibration motion blur compensation factor includes radial blur weight, tangential trailing weight, displacement sensitivity coefficient, and phase correction coefficient; the static feature texture contrast includes texture peak and valley amplitude, edge sharpness, grayscale level, and local contrast coefficient; and the multi-source fusion processing quality evaluation result includes consistency deviation, processing offset, defect risk category, and judgment mark.

3. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that: The image plane geometric transformation matrix refers to the coordinate transformation matrix that performs geometric correction on the pixel coordinates of the original image based on thermal drift and refraction offset; The imaging displacement deviation refers to the offset of the image target position caused by the combined effects of spindle radial runout, rotational frequency, and exposure time at the moment of exposure acquisition.

4. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that: The visual degradation components refer to the image contrast reduction, ghosting, and feature blur caused by non-processing factors such as vibration blur, thermal drift, and imaging offset. The quantification consistency deviation refers to the degree of numerical deviation obtained by mapping texture features to spatial points and comparing them with a preset processing quality benchmark range.

5. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that, The heterogeneous signal instantaneous synchronization module includes: The voltage acquisition submodule detects the transient voltage signal fed back by the piezoelectric sensor at the LED epitaxial wafer dicing station, reads the potential of continuous sampling points according to the sensing sampling clock, and performs normalization processing based on the potential change amplitude of adjacent sampling points to obtain the transient voltage amplitude. The noise discrimination submodule extracts the noise trigger voltage reference based on the transient voltage amplitude, converts the transient voltage amplitude into a normalized amplitude, and compares the normalized amplitude with a preset threshold reference to obtain the noise exceeding the limit criterion. The pixel acquisition submodule calls the noise limit criterion, maps the positive criterion interval to charge transfer gating timing, calibrates the timing start point according to the sensor sampling clock, drives the image sensor to perform charge transfer acquisition, receives the pixel charge in the exposure window and maps it to grayscale data according to row and column addresses, and obtains the audio-visual synchronized pixel matrix.

6. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that, The thermal field optical deviation correction module includes: The temperature acquisition submodule reads the real-time temperature reading from the platinum resistance sensor on the inner wall of the objective lens barrel in the LED scribing machine. Combined with the aforementioned acoustic-optical synchronized pixel matrix, the real-time temperature reading is compared with the ambient reference temperature reading point by point according to the sampling clock to obtain the temperature deviation range. The refractive mapping submodule calls the temperature deviation range, associates it with the thermal coefficient of the refractive index of the optical glass, and derives the thermal drift state of the lens barrel according to the positive correspondence between the temperature deviation range and the thermal coefficient of the refractive index to obtain the refractive offset coefficient. The coordinate correction submodule maps the image plane geometric transformation matrix according to the refraction offset coefficient, reverses the acousto-optic synchronized pixel matrix coordinates, and synchronously writes the horizontal and vertical coordinates into the matrix row and column addresses to obtain the thermally induced geometric correction coordinates.

7. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that, The mechanical vibration feature mapping module includes: The amplitude monitoring submodule monitors the radial runout amplitude fed back by the diamond dicing cutter spindle accelerometer. Combined with the thermally induced geometric correction coordinates, it reads the radial acceleration waveform according to the spindle sampling clock and performs dimensional normalization processing on the amplitude of the peak and trough intervals to obtain the radial runout amplitude spectrum. The displacement definition submodule calls the radial runout amplitude spectrum, matches the exposure duration parameter and the rotation speed frequency parameter, and derives the instantaneous displacement deviation of imaging according to the correspondence between the exposure window span and the rotation speed period, thus obtaining the imaging displacement deviation trajectory. The blur compensation submodule, based on the imaging displacement deviation trajectory, associates the spatial orientation offset information under the thermally induced geometric correction coordinates, maps the horizontal and vertical components of the coordinates to the exposure direction respectively, and obtains the vibration motion blur compensation factor.

8. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that, The processing morphology feature analysis module includes: The extreme value extraction submodule extracts the extreme value features of the gray value distribution of the pixel set based on the vibration motion fuzzy compensation factor, reads the gray value peak reading and gray value valley reading according to the pixel row and column address, compares the gray value span with the gray value discrete sequence of the pixel neighborhood, and obtains the gray value extreme value spectrum. The benchmark matching submodule calls the gray-level extreme value spectrum, matches the contrast benchmark value, and converts the gray-level extreme value spectrum and the contrast benchmark value according to the proportional relationship to obtain the contrast offset domain. The degradation correction submodule, based on the contrast offset domain, applies a vibration motion blur compensation factor to correct the visual degradation component in the contrast reference value, and subtracts the degradation ratio according to the compensation factor and the contrast offset domain to obtain the static feature texture contrast.

9. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 1, characterized in that, The quality status fusion analysis module includes: The point retrieval submodule retrieves spatial point information corresponding to the thermally induced geometric correction coordinates based on the static feature texture contrast, processes the trajectory sampling position according to the coordinate row and column address index, compares the texture contrast readings of adjacent sampling positions, and obtains a spatial point index set. The trajectory mapping submodule calls the spatial point index set, associates texture features with the processing trajectory dimension and maps spatial points, and matches the texture comparison readings with the point coordinate sequence according to the trajectory sampling order to obtain the texture point mapping matrix; The quality judgment submodule compares the texture point mapping matrix with the preset processing quality benchmark range, makes a quantitative judgment on the processing consistency deviation, assigns a quality status code according to the position of the deviation falling into the benchmark range, and obtains the multi-source fusion processing quality evaluation result.

10. The LED chip processing multi-source data fusion acquisition and analysis system according to claim 9, characterized in that: During the process of comparing the texture contrast readings of adjacent sampling positions, the static feature texture contrast carried by two adjacent sampling points in the processing trajectory sampling position is obtained. The static feature texture contrast is subjected to differential subtraction calculation and the absolute value deviation is extracted. When the absolute value deviation reaches the preset texture mutation limit value, the two adjacent sampling points are stored in the spatial point index set. In the process of quantifying the processing consistency deviation, the spatial geometric straight-line distance between each coordinate point in the texture point mapping matrix and the center position of the preset processing quality benchmark interval is measured. The spatial geometric straight-line distance is defined as the processing consistency deviation. The processing consistency deviation is compared with the coverage area of ​​the preset processing quality benchmark interval. The target position of the processing consistency deviation falling into the preset processing quality benchmark interval is mapped to the corresponding quality status code to obtain the multi-source fusion processing quality evaluation result.

Images