Multi-modal data analysis system and method for solder paste printing process

Abstract

The invention discloses a multi-modal data analysis system and method for a solder paste printing technology, and relates to the technical field of solder paste printing, and the method comprises the following steps: collecting transient Raman spectrum response data of each space measuring point of solder paste in a high-speed shearing state in real time; extracting a spectral intensity ratio of each spatial measuring point in the characteristic peak position interval, and constructing a spectral ratio representing a proportional relation between the soldering flux and the tin powder at the measuring point; and constructing a spectrum ratio difference matrix among the multiple measuring points, and generating a shear rate mapping model according to the scraper operation parameters. Through multi-channel Raman spectrum and shear field coupling analysis, high-precision identification of solder paste component segregation is achieved, micro-scale segregation indexes are constructed to quantify risks, a vibration response model and a dynamic regulation and control mechanism are combined, the system can adjust vibration parameters in real time, segregation is effectively restrained, and the reliability of the system is improved. And intelligent closed-loop control and welding quality improvement in the printing process are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tin paste printing, in particular to a multi-modal data analysis system and method for tin paste printing process. BACKGROUND

[0002] The multi-modal data analysis for tin paste printing process refers to comprehensively collecting and fusing heterogeneous data sources from multiple dimensions in the tin paste printing process, such as visual image data (printing pattern edge definition, offset degree, thickness uniformity), sensor data (squeegee pressure, printing speed, stencil temperature), machine parameter logs (device running state, process setting parameters), and environmental monitoring data (humidity, temperature, dust concentration), etc., to perform joint modeling and deep correlation analysis. By constructing a multi-modal fusion feature space, the coupling relationship and abnormal patterns between data sources are identified, which can systematically reveal the root cause mechanism of tin paste printing defects (such as offset, less tin, continuous tin, and edge collapse, etc.), realize fine prediction of printing quality, intelligent optimization of process parameters, and forward-looking maintenance of device operation, and provide data-driven decision support for tin paste printing process quality control under the intelligent manufacturing system.

[0003] The prior art has the following disadvantages: In the high-speed printing process of tin paste, due to the uneven distribution of the squeegee shear field, the active flux in the tin paste and the tin powder particles may be micro-segregated, which may lead to uneven distribution of internal functional components on the surface of the pad area. The existing tin paste detection method mainly relies on visual image recognition, focuses on two-dimensional visual features such as thickness profile and edge morphology for quality judgment, but fails to effectively identify the microstructure and segregation state of the material, lacks fusion analysis of spectral information or component distribution. This makes it possible for the pad area to be misjudged as qualified in the presence of flux shortage or tin powder accumulation abnormalities. This problem is particularly serious in high-pin-density chip applications (such as BGA, QFN), which can easily cause functional failures such as virtual welding and poor solder joint contact, and is difficult to detect early through conventional electrical performance testing methods, which poses a significant risk to product quality and reliability.

[0004] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a multimodal data analysis system and method for a solder paste printing process, which realizes real-time spectroscopy acquisition of solder paste in a high-speed printing process through a multi-channel micro-Raman spectrum array, accurately characterizes the micro-component fluctuation of solder paste by combining spectroscopy ratios, spectroscopy ratio difference matrices, and drift intensity factors, and constructs a micro-scale segregation index, significantly improving the spatial recognition and quantification ability of segregation risk. At the same time, by constructing a mapping model of the segregation index and the vibration response frequency band, and combining the sigmoid function to realize dynamic adjustment of the vibration amplitude, an intelligent feedback closed-loop system for segregation monitoring and control is established, effectively reducing the degree of component segregation, improving printing uniformity and welding reliability, to solve the problems in the above background technology.

[0006] To achieve the above purpose, the present application provides the following technical scheme: a multimodal data analysis method for a solder paste printing process, comprising the following steps: S101, arranging a multi-channel micro-Raman spectrum array in the doctor blade action area of a solder paste printing device, for real-time acquisition of transient Raman spectrum response data of each spatial measurement point of solder paste in a high-speed shearing state, to obtain spectroscopy characteristic data corresponding to the component distribution of active flux and tin powder particles in the solder paste; S102, based on the characteristic peak interval of active flux and tin powder particles in the solder paste in the Raman spectrum, extracting the spectral intensity ratio of each spatial measurement point in the characteristic peak interval, and constructing a spectroscopy ratio representing the proportion relationship between the flux and the tin powder at the measurement point; S103, constructing a spectroscopy ratio difference matrix between multiple measurement points, for reflecting the component difference of solder paste at different spatial positions; and generating a shearing rate mapping model according to the doctor blade operation parameters, to spatially correlate the spectroscopy ratio corresponding to each measurement point in the spectroscopy ratio difference matrix with the shearing rate information at its position; S104, defining a component deviation vector across spatial measurement points according to the characteristic peak shift trend and spectral intensity change characteristic of the flux and tin powder in the Raman spectrum, and calculating a component drift intensity factor representing the overall component fluctuation degree based on the component deviation vector; S105, performing tensor fusion operation on the component drift intensity factor and the shearing gradient tensor corresponding to each measurement point in the shearing rate mapping model, to obtain a micro-scale segregation index comprehensively reflecting the component change and shearing field distribution of the solder paste, for quantifying the component segregation degree of the solder paste printing area; S106, judging whether the micro-scale segregation index exceeds a preset segregation threshold; if so, real-time acquisition of vibration response spectrum data of the solder paste printing area is performed, and a low-frequency coupled vibration mode highly correlated with the micro-scale segregation index is identified therefrom, to construct a mapping model between the vibration frequency band and the segregation intensity; S107, determine the best response vibration frequency interval corresponding to the current segregation state according to the mapping model, and dynamically adjust the vibration amplitude and fine adjustment step length by combining the micro-scale segregation index with the sigmoid function, so as to enhance the mixing effect of the solder paste in the local doctor blade shear area, thereby reducing the micro-scale component segregation degree and improving the printing uniformity and welding reliability.

[0007] Preferably, the step of arranging the multi-channel micro-Raman spectrum array in the doctor blade action area comprises the following: According to the modeling results of the rheological behavior of the solder paste under different printing speeds, the spatial distribution density of the Raman spectrum measurement points in the doctor blade area is preset; A plurality of micro-Raman spectrum acquisition units are arranged in the front edge, middle and rear edge areas of the doctor blade respectively to cover the key stress field change sections where component segregation may occur in the solder paste under high-speed shearing state; A shutter control mechanism with nanosecond-level response capability is used to synchronously trigger all the arranged micro-Raman spectrum acquisition units, so that each measurement point accurately captures the corresponding transient Raman spectrum at different time slices in the solder paste shearing process, thereby realizing high-precision spectrum data acquisition combining time resolution and spatial coverage.

[0008] Preferably, the step of extracting the spectral intensity ratio of each spatial measurement point in the characteristic peak interval and constructing the spectral ratio comprises the following sub-steps: Based on the molecular structure information of the known active flux and tin powder particles in the solder paste, the corresponding reference characteristic peak interval is determined in the Raman spectrum database respectively, and the peak position drift tolerance range in actual measurement is confirmed through pre-experiment calibration; The transient Raman spectrum collected by each spatial measurement point is subjected to background noise reduction and normalization processing to eliminate the interference of system noise, temperature drift effect and laser drift, and to ensure the stability and comparability of the peak intensity contrast result; The maximum peak intensity of the flux peak and the tin powder peak in their corresponding intervals is extracted from the processed Raman spectrum, the integral intensity is calculated respectively, and the spectral intensity ratio corresponding to the measurement point is established; The spectral intensity ratio is combined with the peak shape change factor (such as half-height width, peak area ratio, etc.) to construct the spectral ratio, which is used as a quantitative characteristic index of the active flux and tin powder ratio at the measurement point, and is used to represent the component uniformity state of the solder paste at the point.

[0009] Preferably, the step of constructing the spectral ratio difference matrix and spatially correlating with the shearing rate mapping model comprises the following sub-steps: According to the spatial coordinate sequence of each Raman spectrum measurement point in the solder paste printing area, the spectral ratio corresponding to each measurement point is extracted and arranged to form an initial spectral ratio distribution matrix; Based on the spectral ratio distribution matrix, the spectral ratio difference between each pair of adjacent measuring points is calculated, a spectral ratio difference matrix for characterizing the local component change amplitude is constructed, and a component mutation region is identified by combining a spatial gradient algorithm; According to the real-time running parameters of the scraper, including speed, pressure and motion trajectory information, a spatial mapping model of shear rate field is constructed, and each measuring point is assigned a local shear rate value in its region; Each data unit in the spectral ratio difference matrix is paired with its corresponding shear rate, a spatial correlation matrix for describing the relationship between shear conditions and component changes is established, and the quantitative modeling of the component segregation trend induced by shear stress is realized.

[0010] Preferably, the component mutation region is identified by combining a spatial gradient algorithm, and the specific implementation steps can be performed as follows: The spectral ratio corresponding to each Raman measuring point is rearranged according to its spatial coordinates (such as X-Y grid position), and a two-dimensional component distribution matrix is constructed. The local derivative of the spectral ratio difference matrix is calculated by using a spatial gradient algorithm (such as Sobel operator, Prewitt operator or self-defined Laplace edge enhancement detection algorithm), the spectral ratio change rate of each measuring point in its neighborhood is extracted, and the spatial gradient field of the spectral ratio is obtained. By setting a threshold judgment rule, the measuring points with gradient amplitude exceeding the set threshold are marked as component mutation regions, which usually represent the positions where the proportion of flux to tin powder in the solder paste changes significantly.

[0011] Preferably, the local shear rate value of each Raman spectrum measuring point is assigned by the following specific steps: Based on the real-time running parameters of the scraper in the solder paste printing equipment, including instantaneous linear speed, applied pressure and motion trajectory coordinates, and combined with the relative sliding characteristics of the solder paste between the scraper and the substrate surface, a continuum mechanics model is constructed to establish the shear rate distribution function of the scraper action area. The real-time acquired scraper speed vector field and pressure distribution data are used as input, combined with the actual thickness, viscosity and non-Newtonian rheological characteristics of the solder paste, and substituted into the continuum mechanics model to solve, and the shear rate field distribution map in the three-dimensional space of the entire printing area is obtained. For each Raman spectrum measuring point arranged under the scraper, its two-dimensional spatial coordinate position in the scraper area is extracted, and the coordinate is mapped to the established shear rate field. The shear rate value in the micro region where the measuring point is located is accurately calculated by local interpolation algorithm (such as bilinear interpolation, triangular subdivision interpolation or finite element grid interpolation), which is used as the quantitative description result of the local shear environment of the measuring point.

[0012] Preferably, the steps of tensor fusion operation of the component drift intensity factor with the shear gradient tensor in the shear rate mapping model and calculation of the micro-scale segregation index include the following sub-steps: For each Raman spectrum measurement point, obtain its composition drift intensity factor, and extract the shear gradient tensor corresponding to the position of the measurement point from the shear rate mapping model, which describes the directional variation and gradient amplitude distribution of the local shear rate in space; The composition drift intensity factor and the shear gradient tensor are dimensionally standardized to ensure comparability in numerical scale, and the shear gradient tensor is converted to a matrix form that can participate in fusion calculation through eigenvector expansion; The composition drift intensity factor and the shear gradient tensor are fused at the corresponding measurement point by using tensor inner product or weight mapping method, to obtain the tensor response intensity value corresponding to the measurement point, reflecting the response ability of composition fluctuation in the local shear field structure; The tensor response intensity values of all measurement points are integrated, and the micro-scale segregation index representing the coupling degree of composition non-uniformity and shear force field in the entire doctor blade region is calculated based on the weighted average or variance enhancement aggregation strategy, which is used to quantify the overall risk level and spatial trend distribution of tin paste composition segregation during printing.

[0013] Preferably, the step of constructing a mapping model between vibration frequency band and segregation intensity includes the following sub-steps: Set the allowed segregation threshold in the tin paste printing process, and monitor the micro-scale segregation index of each measurement point in real time during printing to determine whether it exceeds the segregation threshold; When the micro-scale segregation index of a certain region is detected to exceed the segregation threshold, immediately start the high-sensitivity vibration sensor array to collect the vibration response spectrum of the region at the printing moment, and obtain the original vibration response spectrum data containing amplitude-frequency characteristics; Perform Fourier transform and frequency band feature extraction on the collected vibration spectrum data, select low-frequency vibration components with high linear correlation with the change trend of the micro-scale segregation index, and mark them as low-frequency coupled vibration modes; Based on the mapping relationship between different micro-scale segregation indexes and corresponding low-frequency vibration energy distribution, a function model between vibration frequency band and segregation intensity is constructed as a reference for the subsequent vibration regulation response mechanism, which is used to intelligently adjust the local mixing state of tin paste under shear environment.

[0014] Preferably, the step of dynamically adjusting the vibration amplitude and fine adjustment step size in combination with the micro-scale segregation index includes the following sub-steps: Based on the mapping model between the vibration frequency band and the segregation intensity, find the best response vibration frequency interval corresponding to the micro-scale segregation index calculated at present, and set the frequency interval as the target working frequency band of the current control system; According to the interval division strategy implemented according to the numerical range of the micro-scale segregation index, the micro-scale segregation index is divided into multiple adjustment levels, and each level corresponds to a set of preset vibration initial amplitude and fine adjustment step length parameter table for guiding the response rhythm under different segregation intensities; A real-time feedback control mechanism is started, and the vibration control strategy is dynamically corrected in combination with the change rate of the current micro-scale segregation index in the time sequence window, specifically: when the micro-scale segregation index shows a rapid upward trend, the vibration amplitude is significantly increased to quickly disperse the local segregation structure; when the micro-scale segregation index slowly decreases, the step length is appropriately reduced to achieve fine adjustment of the mixing degree; Within the optimal response vibration frequency interval, a nonlinear amplitude modulation model is constructed in combination with the sigmoid function to form a smooth response relationship between the vibration amplitude and the micro-scale segregation index, and the vibration amplitude and the fine adjustment step length are iteratively adjusted to enhance the mixing effect of the solder paste in the local doctor blade shear region, thereby effectively reducing the micro-scale component segregation degree and improving the printing uniformity and welding reliability.

[0015] A multi-modal data analysis system for a solder paste printing process includes a spectroscopy data acquisition module, a spectroscopy ratio construction module, a component-shear correlation modeling module, a component drift evaluation module, a segregation calculation module, a vibration mode identification module, and an intelligent vibration control module. The spectroscopy data acquisition module acquires real-time transient Raman spectrum response data of each spatial measurement point of the solder paste under high-speed shearing state. The spectroscopy ratio construction module extracts the spectral intensity ratio of each spatial measurement point in the characteristic peak interval to construct a spectroscopy ratio representing the proportion relationship between the flux and the tin powder at the measurement point. The component-shear correlation modeling module constructs a spectroscopy ratio difference matrix among multiple measurement points, and generates a shear rate mapping model according to the doctor blade running parameters to spatially correlate the spectroscopy ratio of each measurement point in the spectroscopy ratio difference matrix with the shear rate information at its location. The component drift evaluation module defines a component deviation vector across spatial measurement points, and calculates a component drift intensity factor representing the overall component fluctuation degree based on the component deviation vector. The segregation calculation module performs tensor fusion operation on the component drift intensity factor and the shear gradient tensor corresponding to each measurement point in the shear rate mapping model to obtain the micro-scale segregation index. The vibration mode identification module determines whether the micro-scale segregation index exceeds a preset segregation threshold; if so, it acquires real-time vibration response spectrum data of the solder paste printing area, identifies a low-frequency coupled vibration mode highly correlated with the micro-scale segregation index from the vibration response spectrum data, and constructs a mapping model between the vibration frequency band and the segregation intensity. The intelligent vibration regulation module determines the optimal response vibration frequency interval corresponding to the current segregation state according to the mapping model, and dynamically adjusts the vibration amplitude and the fine adjustment step length by using a sigmoid function in combination with the micro-scale segregation index.

[0016] In the above technical solution, the present application provides technical effects and advantages: By introducing a multi-channel micro-Raman spectrum array, the present application realizes real-time spectral acquisition of the high-speed shearing state of the solder paste during printing, so that not only the surface morphology can be obtained, but also the relative proportion and distribution state of the flux and tin powder particles in the material can be deeply detected. By constructing a spectral ratio, a spectral ratio difference matrix between multiple measurement points, and a composition drift intensity factor based on peak position shift, high-resolution and multi-dimensional characterization of the micro-component fluctuation of the solder paste is realized, and the technical bottleneck that the existing two-dimensional image recognition technology cannot detect internal segregation structure is overcome. In addition, after spatial coupling of the composition change information and the shearing rate field, the generated micro-scale segregation index has the advantages of quantifiability, distributability and traceability, which provides theoretical support and data basis for accurately identifying the local segregation risk area, and greatly improves the accuracy and scientificity of the printing quality control.

[0017] By constructing a micro-scale segregation index-vibration response frequency band mapping model, the present application establishes a physical and data double-layer coupling relationship between the composition segregation state and the equipment vibration characteristics, and further designs a nonlinear amplitude adjustment mechanism based on a sigmoid function, which can dynamically adjust the vibration parameters according to the numerical level and change trend of the micro-scale segregation index. When the segregation risk occurs, the system can enter the feedback control mode in real time, and adaptively enhance the mixing effect of the local doctor blade shearing area in the optimal response frequency interval, so as to realize immediate intervention and trend reversal of segregation. This mechanism not only avoids the continuous accumulation of material defects, but also realizes the closed-loop transformation from "detecting segregation" to "regulating segregation", greatly improves the intelligent level, stability and welding reliability of the solder paste printing process, and is especially suitable for high-pin-density packaging products in high-consistency scenarios of solder paste. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0019] Figure 1 The method flowchart of the multi-modal data analysis method of the present application for the solder paste printing process.

[0020] Figure 2A module schematic diagram of the multi-modal data analysis system for the solder paste printing process of the present application. DETAILED DESCRIPTION

[0021] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art.

[0022] The present application provides a multi-modal data analysis method for the solder paste printing process as shown in the accompanying drawings, comprising the following steps: Figure 1 The multi-modal data analysis method for the solder paste printing process as shown in the accompanying drawings comprises the following steps: S101, arranging a multi-channel micro-Raman spectrum array in the squeegee action area of the solder paste printing equipment, for real-time acquisition of transient Raman spectrum response data of each spatial measurement point of the solder paste in a high-speed shearing state, to obtain spectral characteristic data corresponding to the composition distribution of active flux and tin powder particles in the solder paste; The step of arranging a multi-channel micro-Raman spectrum array in the squeegee action area comprises the following: According to the modeling results of the rheological behavior of the solder paste at different printing speeds, the spatial distribution density of the Raman spectrum measurement points in the squeegee area is preset; A plurality of micro-Raman spectrum acquisition units are arranged in the leading edge, middle and trailing edge areas of the squeegee, to cover the key stress field change sections where composition segregation may occur inside the solder paste in a high-speed shearing state; A shutter control mechanism with nanosecond-level response capability is used to synchronously trigger all the arranged micro-Raman spectrum acquisition units, so that each measurement point accurately captures the corresponding transient Raman spectrum at different time slices in the solder paste shearing process, thereby realizing high-precision spectrum data acquisition combining time sequence resolution and spatial coverage.

[0023] To achieve the spatial distribution density of preset Raman spectrum measurement points according to the modeling results of the rheological behavior of the solder paste at different printing speeds, the following methods can be used: first, the shear rate, shear stress and viscosity response relationship of the solder paste at various printing speeds (such as 100 mm / s, 200 mm / s, 300 mm / s, etc.) is tested, and the rheological model (such as Carreau-Yasuda or Bingham model) is used to fit the rheological characteristics, and the stress field distribution trend of the solder paste at different speeds is obtained; then, based on the simulation results, the key positions of stress concentration, shear gradient mutation or material flow direction change in the blade action area are set as high-priority Raman collection points; finally, according to the differences in stress field distribution at different printing speeds, the spatial density of the measurement points is dynamically adjusted, and the measurement point layout is densified in the area with sharp stress changes, and the measurement point layout is appropriately sparse in the area with relatively stable stress, so as to realize an adaptive measurement point distribution strategy that matches the rheological behavior of the solder paste and covers the key shear behavior area. This strategy ensures that the spectrum array can collect data at the most representative physical state, improving the detection sensitivity and discrimination accuracy of micro-scale component segregation.

[0024] The multi-channel micro-Raman spectrum array is arranged in the blade action area of the solder paste printing equipment, and its core function is to realize real-time, non-contact detection and accurate identification of the micro-component distribution change of the solder paste under high-speed shear state. Solder paste, as a typical non-Newtonian paste, will exhibit complex rheological behavior during high-speed shear by the blade, such as relative slip between the flux and tin powder particles, micro-scale separation, and even physical segregation, especially under the condition of high printing speed and sharp change of blade pressure. Traditional solder paste detection methods, such as two-dimensional visual image recognition or thickness profile analysis, can only capture the appearance geometric features and cannot obtain the real-time distribution change of the internal functional components (such as active flux and solder metal powder) of the solder paste in space, so they cannot effectively characterize the micro-segregation behavior.

[0025] By arranging a multi-channel micro-Raman spectrum array, the transient Raman spectrum response of different measurement points of the solder paste in the shearing state can be captured in real time without destroying the material structure. The Raman spectrum technology has the chemical composition recognition ability at the molecular level, can identify the organic group peak position in the flux and the lattice vibration characteristics of the tin powder, and thus realizes the quantitative analysis of the distribution intensity and variation trend of the two types of key ingredients. By arranging multiple Raman measurement points at the key shearing field regions such as the front edge, the middle and the rear edge of the scraper, the stress evolution process of the solder paste in the whole shearing path can be comprehensively covered, and the evolution law of the ingredients in time and space can be captured, thereby providing data support for subsequent construction of the ingredient segregation model, the ingredient drift index and the correlation with the process parameters. This step not only significantly improves the early identification ability of the micro-quality problems of the solder paste such as the flux shortage and the tin powder accumulation, but also lays a foundation for realizing the intelligent quality control and online segregation compensation in the high-speed printing process.

[0026] In S102, based on the characteristic peak position intervals of the active flux and the tin powder particles in the Raman spectrum of the solder paste, the spectral intensity ratio of each spatial measurement point in the characteristic peak position interval is extracted, and a spectral ratio representing the proportion relationship of the flux and the tin powder at the measurement point is constructed. The step of extracting the spectral intensity ratio of each spatial measurement point in the characteristic peak position interval and constructing the spectral ratio includes the following sub-steps: Based on the molecular structure information of the known active flux and the tin powder particles in the solder paste, the corresponding reference characteristic peak position intervals are determined in the Raman spectrum database respectively, and the peak position drift tolerance range in the actual measurement is confirmed through pre-experiment calibration. The transient Raman spectrum collected at each spatial measurement point is subjected to background noise reduction and normalization processing, so as to eliminate the interference of system noise, temperature drift effect and laser drift, and ensure the stability and comparability of the peak intensity contrast result. The maximum peak intensity of the flux peak and the tin powder peak in its corresponding interval is extracted from the processed Raman spectrum, the integral intensity thereof is calculated respectively, and the spectral intensity ratio corresponding to the measurement point is established; The spectral intensity ratio is combined with the peak shape change factor (such as half-height width, peak area ratio, etc.) to construct the spectral ratio, which is used as a quantitative characteristic index of the proportion relationship of the active flux and the tin powder at the measurement point, and is used to represent the composition uniformity state of the solder paste at the point.

[0027] The spectral ratio is constructed based on the respective characteristic peak shape parameters of the active flux and the tin powder particles in the Raman spectrum, and specifically includes the following contents: the integral intensity of the flux characteristic peak and the integral intensity of the tin powder characteristic peak wherein the integral intensity is the spectral area of each Raman peak in the preset peak position interval, representing the relative abundance of the ingredient at the measurement point; and the half-height width parameters of the flux and tin powder characteristic peaks are obtained With , for reflecting the molecular structure broadening degree of different components in the shearing state, and then assisting in evaluating their mixing uniformity and physical state change; extracting the symmetry parameters of the characteristic peaks of flux and tin powder With , the symmetry parameter is defined as the ratio of the left and right half-widths of the peak shape, which is used to represent the asymmetry change caused by the structure deviation or stress stretching due to shear disturbance; the parameters of the above three spectral dimensions are fused to construct the following spectral ratio B, and the specific expression is: , wherein: , , are the integral intensity ratio , the peak width ratio and the symmetry ratio The weighting coefficients in the spectral ratio expression are experimentally calibrated according to the actual tin paste material system and the Raman collection conditions. Through the above expression, the local proportion relationship and structure state of the active flux and tin powder particles in the tin paste can be comprehensively quantitatively evaluated, providing accurate spectral feature index support for subsequent segregation identification and shear compensation.

[0028] In the high-speed printing process of tin paste, the micro-proportion change of active flux and tin powder particles directly affects the wetting, filling and welding strength of the solder joint, especially in high pin density packaging (such as BGA, QFN). Therefore, real-time quantitative detection of these two core components in tin paste has become a key technical means to ensure printing quality and functional reliability. The role of this step is to extract the spectral intensity ratio of each spatial measurement point in the characteristic peak interval, and to construct a spectral ratio representing the proportion relationship of flux and tin powder at the measurement point, that is, by introducing the Raman spectrum technology with molecular recognition ability, real-time, non-contact and quantitative analysis of the local component proportion of tin paste at the microscale is realized. Specifically, the crystal structure vibration of the organic formula components (such as carboxylate, alcohol) in the active flux and the tin powder corresponds to different Raman characteristic peaks, by extracting the integral intensity in the characteristic waveband and calculating the ratio, the relative content relationship of the two at the measurement point can be intuitively reflected. At the same time, this ratio can be further expressed and optimized in combination with peak width, peak shape symmetry and other structural characteristics, to enhance its sensitivity to component changes under shear state. By constructing this spectral ratio, not only can local segregation phenomena caused by uneven shear or material accumulation be quickly identified, but also it provides basic data support for subsequent spatial difference modeling, component drift factor calculation and segregation risk assessment, which is a key intermediate step to realize intelligent monitoring of tin paste component quality.

[0029] S103, construct a spectral ratio difference matrix between multiple measurement points, for reflecting the component difference of the solder paste at different spatial positions; and generate a shear rate mapping model according to the doctor blade running parameters, to spatially correlate the spectral ratio corresponding to each measurement point in the spectral ratio difference matrix with the shear rate information at the position of the measurement point; The step of constructing the spectral ratio difference matrix and spatially correlating the spectral ratio difference matrix with the shear rate mapping model includes the following sub-steps: According to the spatial coordinate sequence of each Raman spectrum measurement point in the solder paste printing area, extract and arrange the spectral ratio corresponding to each measurement point to form an initial spectral ratio distribution matrix; Based on the spectral ratio distribution matrix, calculate the spectral ratio difference between each pair of adjacent measurement points, construct a spectral ratio difference matrix for characterizing the local component change amplitude, and identify the component mutation region in combination with a spatial gradient algorithm; According to the real-time running parameters of the doctor blade, including speed, pressure and motion trajectory information, construct a spatial mapping model of the shear rate field, and assign a local shear rate value of the region to each measurement point; Pair each data unit in the spectral ratio difference matrix with the corresponding shear rate one by one to establish a spatial correlation matrix for describing the relationship between the shear condition and the component change, thereby realizing the quantitative modeling of the component segregation trend caused by the shear stress.

[0030] In combination with the spatial gradient algorithm, the component mutation region is identified, and the specific implementation steps can be performed in the following manner: Re-arrange the spectral ratio corresponding to each Raman measurement point according to its spatial coordinates (such as X-Y grid position) to construct a two-dimensional component distribution matrix; use a spatial gradient algorithm (such as Sobel operator, Prewitt operator or a self-defined Laplace edge enhancement detection algorithm) to calculate the local derivative of the spectral ratio difference matrix, extract the spectral ratio change rate of each measurement point in its neighborhood, and thereby obtain the spatial gradient field of the spectral ratio; by setting a threshold judgment rule, mark the measurement points with a gradient amplitude exceeding the set threshold as the component mutation region, and these points usually represent positions where the proportion of the flux to the solder powder in the solder paste changes significantly.

[0031] The local shear rate value of the region is assigned to each Raman spectrum measurement point, which can be realized by the following specific steps: Based on the real-time running parameters of the squeegee in the solder paste printing equipment, including the instantaneous linear speed, the applied pressure and the motion trajectory coordinates, combined with the relative slip characteristics of the solder paste between the squeegee and the surface of the substrate, a continuous medium mechanics model is constructed, and a shear rate distribution function of the squeegee action area is established; the real-time collected squeegee speed vector field and pressure distribution data are taken as inputs, combined with the actual thickness, viscosity and non-Newtonian rheological characteristics of the solder paste, substituted into the medium mechanics model to solve, and the shear rate field distribution diagram in the entire printing area three-dimensional space is obtained; for each Raman spectrum measurement point arranged below the squeegee, the two-dimensional spatial coordinate position of the measurement point in the squeegee area is extracted, and the coordinate is mapped to the established shear rate field, and the shear rate value of the micro area where the measurement point is located is accurately calculated through a local interpolation algorithm (such as bilinear interpolation, triangular subdivision interpolation or finite element grid interpolation), which is taken as the quantitative description result of the local shear environment of the measurement point; The process can realize real-time spatial coupling between the printing process composition distribution and the shear behavior, and lay the foundation for identifying composition segregation caused by uneven shear.

[0032] The main role of this step is to realize fine spatial coupling between the solder paste composition change information and the shear mechanical behavior, so as to reveal the root cause of composition segregation in the high-speed printing process, and provide a data basis for subsequent segregation identification and intelligent compensation. By constructing a spectral ratio difference matrix among multiple measurement points, the composition proportion change trend of the solder paste at different spatial positions in the squeegee area can be intuitively reflected at the Raman spectrum level, and the non-uniform distribution of the flux and the tin powder at the micro scale is quantified. The difference matrix not only reveals the composition distribution gradient in space, but also can be used to identify composition mutation areas or potential segregation bands, solving the problem that traditional image detection cannot penetrate into the material. At the same time, in order to understand the physical causes of composition change, the running parameters of the squeegee are further introduced, including the linear speed of the squeegee, the applied pressure and the motion trajectory varying with time, and a spatial mapping model of the shear rate is established through a continuous medium mechanics model. The model reflects the shear intensity difference experienced by the solder paste in different areas of the squeegee, which is a key physical mechanism affecting the composition segregation behavior. By spatially pairing the spectral ratio of each spectrum measurement point with the shear rate corresponding to its position, a one-to-one mapping between the composition gradient and the shear behavior is realized. This mapping relationship can reveal whether high shear rate areas are more likely to induce flux migration or tin powder accumulation, thereby systematically quantifying the coupling effect between "shear-segregation", which is an important basic step for building a composition drift model, predicting printing quality fluctuations and optimizing process parameters.

[0033] S104、According to the characteristic peak position shift trend and spectral intensity change characteristics of the flux and the tin powder in the Raman spectrum, a composition deviation vector across the spatial measurement points is defined, and a composition drift intensity factor representing the overall composition fluctuation degree is calculated based on the composition deviation vector; The specific steps for calculating the component drift intensity factor representing the overall component fluctuation degree based on the deviation vector are as follows: Extract the spectral ratios and corresponding characteristic peak positions of all Raman measurement points in the scraping area of the scraper, and establish a multi-dimensional spectral state vector set between the measurement points; Select a group of reference measurement points as the reference vector in the ideal component state, and calculate the vector difference between each measurement point and the reference point in turn, to build a component deviation vector field reflecting the component change trend of each point; Based on the component deviation vector field, the modulus of all deviation vectors is calculated using the Euclidean norm (L2 norm), and statistical normalization processing is performed to obtain a numerical index representing the fluctuation amplitude of the components in the entire printing area; The average modulus after statistical processing is taken as the component drift intensity factor to quantify the overall stability of the active flux and tin powder particle component distribution in the high-speed shearing process of the solder paste, and to identify the component abnormal fluctuation area and assess the segregation risk level.

[0034] The reference measurement points as the ideal component state need to be determined comprehensively in combination with the stable working condition section and spectral data distribution characteristics in the solder paste printing process. First, select the area with stable shearing rate and small scraper pressure fluctuation in the entire printing area as the candidate reference area, which usually corresponds to the working condition with uniform material flow and minimum component segregation. Second, select a number of measurement points with stable spectral intensity ratio, no obvious peak shift, and good peak symmetry as reference samples by statistically analyzing the spectral ratios and characteristic peak positions of all Raman measurement points in the area. Third, perform cluster analysis on the spectral state vectors (including intensity ratio, peak position, half-width, etc.) of these sample measurement points, and select the cluster center or mean vector as the reference vector. This vector represents the component characteristic distribution of the solder paste in the ideal mixing state, and can be used as a unified reference standard for calculating the component deviation degree of other measurement points, thereby ensuring the consistency and representativeness of the component deviation vector calculation. This method can effectively avoid the deviation of the overall drift index caused by selecting the reference from local abnormal measurement points.

[0035] This step quantitatively analyzes the component fluctuation behavior of the solder paste in the high-speed printing and shearing process by constructing the component deviation vector and calculating the component drift intensity factor, providing a mathematical basis and discrimination basis for identifying potential component segregation risks. Solder paste is a complex multiphase material composed of active flux and tin powder particles, which is prone to microstructure disturbance under the action of the scraper, resulting in uneven distribution of the active flux and tin powder particles. These changes are not easily detected macroscopically, but they directly affect the subsequent welding quality and electrical performance. Traditional image visual detection methods cannot obtain real-time change information of the internal components, so it is difficult to effectively identify such "hidden defects".

[0036] By analyzing the change trend (such as peak shift, peak width change) of the characteristic peak position corresponding to the flux and tin powder in the Raman spectrum and the spectral intensity ratio (such as intensity ratio decrease or enhancement), the spectral state vector of each spatial measurement point can be extracted. The vector not only contains composition ratio information, but also contains physical state evolution characteristics, such as molecular stretching or agglomeration effect caused by shearing. Then, the vectors are compared with the preset reference vector (representing the ideal mixing state) to construct a composition deviation vector to represent the deviation between the composition state of each measurement point and the ideal state.

[0037] Further, by statistically analyzing the modulus length distribution of the entire deviation vector field and calculating the mean or standard deviation, a comprehensive quantitative indicator, the composition drift intensity factor, can be obtained. The composition drift intensity factor truly reflects the consistency and stability of the mixing state of the flux and tin powder in the entire printing area. If the composition drift intensity factor is large, it indicates that there is extensive and significant composition deviation in the area, and there may be abnormal composition distribution, which is a high segregation risk area. Conversely, it indicates that the composition fluctuation is small and the overall mixing is uniform. Through this step, not only can the multi-dimensional quantification of the composition distribution be realized, but also the core data support for subsequent micro-scale segregation index calculation, quality warning and process adjustment can be provided, which is an important bridge connecting spectral data and quality decision.

[0038] S105, tensor fusion operation is performed on the composition drift intensity factor and the shear gradient tensor corresponding to each measurement point in the shear rate mapping model to obtain a micro-scale segregation index that comprehensively reflects the composition change of the solder paste and the distribution of the shear field, which is used to quantify the composition segregation degree of the solder paste printing area; The step of performing tensor fusion operation on the composition drift intensity factor and the shear gradient tensor in the shear rate mapping model and calculating the micro-scale segregation index includes the following sub-steps: For each Raman spectrum measurement point, the composition drift intensity factor is obtained, and the shear gradient tensor corresponding to the position of the measurement point is extracted from the shear rate mapping model. The shear gradient tensor describes the directional change and gradient amplitude distribution of the local shear rate in space; The composition drift intensity factor and the shear gradient tensor are subjected to dimension standardization processing to ensure comparability in numerical scale, and the shear gradient tensor is converted into a matrix form that can participate in fusion calculation through feature vector expansion; The composition drift intensity factor and the shear gradient tensor are fused at the corresponding measurement point by using tensor inner product or weight mapping method to obtain the tensor response intensity value corresponding to the measurement point, which reflects the response ability of composition fluctuation in the local shear field structure; The microscale segregation index is calculated based on the weighted average or variance-enhanced aggregation strategy to represent the degree of composition heterogeneity and shear field coupling in the entire squeegee area, which is used to quantify the overall risk level and spatial trend distribution of composition segregation during printing.

[0039] Before the fusion operation of the composition drift intensity factor and the shear gradient tensor, there are significant differences in physical meaning, dimension unit, and numerical scale. The composition drift intensity factor is usually a dimensionless statistical value, reflecting the degree of composition change; while the shear gradient tensor is derived from physical field modeling, and its element value usually has a unit and a large numerical interval. In order to enable these two types of data to be mathematically fused, dimension standardization must be performed, that is, different dimensions and different dimensional data are mapped to a unified numerical scale space through mathematical methods. Common methods include normalization (mapping numerical values to the 0-1 interval), Z-score standardization (subtracting the mean and dividing by the standard deviation), or maximum and minimum scaling. Standardization not only avoids the dominance of a data dimension with a large factor value in fusion calculation, but also improves the calculation stability and interpretability, which is an important preprocessing step before cross-modal tensor fusion.

[0040] The shear gradient tensor is a multi-dimensional structure, usually in the form of a second or third order tensor, representing the shear rate change of the local area of the squeegee in each direction in space. In order to enable it to participate in fusion calculation, it must be converted to a form compatible with the composition drift intensity factor. Eigenvalue expansion refers to the linear algebraic processing (such as tensor expansion, eigenvalue decomposition, or singular value decomposition) of the shear gradient tensor to convert it into a set of vector or matrix representations. Specifically, the shear gradient tensor can be projected onto its principal characteristic direction to extract the most representative directional shear change pattern, thereby forming a characteristic vector set that reflects the "dominant trend" of shear behavior. This matrix structure formed after expansion facilitates subsequent inner product, weighted fusion, or correlation coefficient calculation, allowing the composition drift index to be response-weighted in each shear direction, enabling more fine-grained composition-mechanical coupling modeling.

[0041] The step constructs a comprehensive evaluation index, i.e., a micro-scale segregation index, by tensor fusion of information on composition fluctuation (expressed by a composition drift intensity factor) generated by the solder paste material in the printing process and information on the physical distribution of the shear field (expressed by a shear gradient tensor), for comprehensively and quantitatively describing the composition non-uniformity and potential segregation risk caused by the shear action in the solder paste printing area. Under the high-speed movement of the squeegee, the microstructure of the solder paste and the flux is easily disturbed by the non-uniform shear field, causing local separation and segregation of the two-phase material, and further affecting the quality of the solder joint. Single reliance on composition detection data (such as spectrum) can only reflect the result state, and it is difficult to track the physical causes of segregation; and relying only on shear field modeling cannot reflect the real response inside the material. Therefore, the step couples the two by tensor fusion, not only revealing whether the composition drift occurs, but also revealing why and where it occurs.

[0042] Specifically, the process first characterizes the spectral state stability of the material at each position by the composition drift intensity factor, and then extracts the shear gradient tensor of each position from the shear rate mapping model to describe the local mechanical disturbance intensity and direction distribution. After dimension standardization, the tensor inner product, mapping weight or correlation matrix algorithm is used to perform fusion calculation on each measurement point to obtain the response intensity of the composition fluctuation under the actual stress field. Finally, by statistically summarizing the fusion results of all measurement points in the entire printing area, a global micro-scale segregation index is obtained. The micro-scale segregation index can be used to accurately evaluate the risk degree of uneven mixing, composition stratification, local accumulation and other defects of the solder paste in the entire printing area, and becomes an important quantitative basis for online judgment of printing quality, optimization of process control and early warning of segregation. Its introduction realizes the key transition from "data detection" to "physical modeling driven data fusion analysis", and is one of the core technical support points in intelligent solder paste printing quality monitoring.

[0043] S106, judging whether the micro-scale segregation index exceeds a preset segregation threshold; if yes, real-time acquisition of vibration response spectrum data of the solder paste printing area, and identification of a low-frequency coupled vibration mode highly correlated with the micro-scale segregation index from the vibration response spectrum data, to construct a mapping model between the vibration frequency band and the segregation intensity; The step of constructing the mapping model between the vibration frequency band and the segregation intensity includes the following sub-steps: Setting an allowed segregation threshold in the solder paste printing process, and real-time monitoring of the micro-scale segregation index of each measurement point in the printing process to judge whether it exceeds the segregation threshold; When it is detected that the micro-scale segregation index of a certain area exceeds the segregation threshold, a high-sensitivity vibration sensor array is immediately started to acquire the vibration response spectrum of the area at the printing moment, to obtain original vibration response spectrum data containing amplitude-frequency characteristics; The collected vibration spectrum data is subjected to Fourier transform and frequency band feature extraction, and low-frequency vibration components with high linear correlation with the change trend of the micro-scale segregation index are screened out and marked as low-frequency coupled vibration modes; Based on the mapping relationship between different micro-scale segregation indexes and corresponding low-frequency vibration energy distribution, a function model between vibration frequency band and segregation intensity is constructed as a reference for subsequent vibration regulation response mechanism, which is used for intelligent adjustment of local mixing state under the shear environment of tin paste.

[0044] The collected vibration spectrum data is subjected to Fourier transform and frequency band feature extraction, and low-frequency vibration components with high linear correlation with the change trend of the micro-scale segregation index are screened out, which can be carried out according to the following steps: The vibration response time series data is subjected to frequency domain transformation by using fast Fourier transform (FFT) algorithm, the original vibration signal is converted from time domain to frequency domain, the corresponding amplitude-frequency spectrum is obtained, and the distribution characteristics of vibration energy in different frequency intervals are extracted; the vibration spectrum data collected under multiple different segregation states is compared and analyzed, the average energy value of each frequency band (such as within 5 Hz bandwidth) is extracted, and the frequency band energy vector sequence is constructed; the above frequency band energy vector and the micro-scale segregation index value of the corresponding region are subjected to regression analysis, the Pearson correlation coefficient is calculated or the linear regression model is used, and the low-frequency band with significant positive or negative correlation with the change trend of the micro-scale segregation index is identified; the low-frequency band with significant correlation (such as the absolute value of the correlation coefficient is greater than 0.8, or the P value is less than 0.05) is marked as low-frequency coupled vibration mode, which is considered as the response representation of the segregation process at the mechanical level, and provides frequency basis for subsequent construction of vibration regulation model and precise intervention. This process realizes automatic extraction of feature components closely coupled with material state from physical vibration signal, which is a key link in intelligent diagnosis of segregation.

[0045] The purpose of this step is to establish a dynamic correlation mechanism between micro-scale component segregation and physical vibration response in the printing process, so as to realize active identification of segregation state and development of intervention adjustment strategy. In the tin paste printing process, active flux and tin powder particles may be separated or unevenly mixed under the action of high-speed shear force, forming a material internal component segregation area. Although the analysis of component drift and shear distribution can quantify this segregation trend, the physical feedback process is often accompanied by mechanical vibration changes during equipment operation, especially in the case of local material flow abnormality or blockage, the vibration mode will change significantly. By presetting the threshold value of micro-scale segregation index, early trigger monitoring of potential segregation risk can be realized; when the micro-scale segregation index exceeds the threshold value, the system automatically collects high-precision vibration response spectrum data of the current printing area, and converts the original time domain signal to frequency domain energy spectrum through Fourier transform and other signal processing methods, and identifies the low-frequency coupled vibration features highly related to the segregation state.

[0046] By comparing and analyzing the corresponding vibration spectrum under different segregation intensity conditions, a mapping relationship model between the micro-scale segregation index and the vibration energy of a specific frequency band is constructed. This model reveals the modulation effect of segregation degree on vibration mode, i.e., the more severe the segregation, the stronger or more frequent the vibration response in certain low-frequency intervals. This relationship not only provides a macroscopic feedback channel for the physical explanation of material segregation, but also serves as a basis for subsequent vibration control strategies, such as automatically adjusting the frequency to the minimum segregation response interval to enhance the mixing of the solder paste.

[0047] This step couples and analyzes the material spectroscopy data with physical vibration signals, breaking through the limitation of relying solely on spectra or images to determine composition changes. It has high intelligence and self-adaptive ability, and is a key link to realize closed-loop control of the printing process and real-time adjustment of material behavior.

[0048] S107, according to the mapping model, determine the best response vibration frequency interval corresponding to the current segregation state, and use the sigmoid function to dynamically adjust the vibration amplitude and fine-tuning step size in combination with the micro-scale segregation index, to enhance the mixing effect of the solder paste in the local doctor blade shear region, thereby reducing the micro-scale component segregation degree and improving the printing uniformity and welding reliability; The step of dynamically adjusting the vibration amplitude and fine-tuning step size in combination with the micro-scale segregation index includes the following sub-steps: Based on the mapping model between the vibration frequency band and the segregation intensity, find the best response vibration frequency interval corresponding to the micro-scale segregation index calculated at present, and set this frequency interval as the target working frequency band of the current control system; According to the interval division strategy based on the numerical range of the micro-scale segregation index, divide the micro-scale segregation index into multiple adjustment levels, each level corresponding to a set of vibration initial amplitude and fine-tuning step size parameter table, which is used to guide the response rhythm under different segregation intensities; Start the real-time feedback control mechanism, and dynamically correct the vibration control strategy in combination with the change rate of the current micro-scale segregation index in the time series window. Specifically, when the micro-scale segregation index shows a rapid upward trend, significantly increase the vibration amplitude to quickly disperse the local segregation structure; when the micro-scale segregation index slowly decreases, appropriately reduce the step size to achieve fine adjustment of the mixing degree; Within the best response vibration frequency interval, construct a nonlinear amplitude modulation model based on the sigmoid function to form a smooth response relationship between the vibration amplitude and the micro-scale segregation index, and iteratively adjust the vibration amplitude and fine-tuning step size to enhance the mixing effect of the solder paste in the local doctor blade shear region, thereby effectively reducing the micro-scale component segregation degree and improving the printing uniformity and welding reliability.

[0049] According to the specific numerical level of the micro-scale segregation index, an interval division strategy is adopted, which can be implemented through the following specific steps: based on historical experimental data and field process experience, the numerical distribution range of the micro-scale segregation index under different printing states is counted, and the upper and lower limit values of the variation interval are determined; the numerical range is divided into several engineering significance level intervals, for example, set as "low segregation level", "medium segregation level", "high segregation level" and "severe segregation level", each level corresponds to a specific numerical threshold interval (such as 0.00-0.25, 0.25-0.50, 0.50-0.75, 0.75-1.00); a set of vibration control parameters is preset for each segregation level, including the corresponding initial vibration amplitude, maximum adjustment amplitude range and fine adjustment step, to form a one-to-one mapping table of level and parameter; the change of the micro-scale segregation index is monitored in real time during the printing process, and the interval is determined, and the corresponding vibration parameter combination of the level is automatically called, so as to realize the self-adaptive adjustment mechanism of intelligent matching of intervention intensity according to the segregation degree. The interval division strategy not only enhances the flexibility of system response, but also ensures the stability, gradualness and predictability of the regulation process, and is one of the core foundations of vibration control closed-loop regulation.

[0050] The core role of this step is to build an intelligent intervention mechanism with "component segregation perception-vibration response regulation" as a closed loop, and to realize the dynamic mixing optimization of the solder paste in the high-speed doctor blade shear region through the frequency-amplitude coordinated regulation strategy, so as to reduce the risk of micro-scale component segregation and improve the uniformity and welding reliability of the printing process. In the process of solder paste printing, due to uneven shear rate distribution, complex material rheological properties and other factors, the tin powder and active flux may occur local aggregation or migration, resulting in unbalanced component distribution. Although the previous step can extract the micro-scale segregation index as a quantitative index through spectral detection and shear modeling, if there is no effective feedback regulation mechanism, it is difficult to fundamentally control the segregation by relying only on data identification.

[0051] To this end, the step first determines the optimal response vibration frequency interval corresponding to the current segregation state based on the mapping model between the vibration frequency band and the segregation intensity, that is, within the frequency range, the solder paste shows the best mixing response ability to vibration excitation. Subsequently, according to the specific numerical value of the micro-scale segregation index, the micro-scale segregation index is divided into multiple adjustment levels, and each level corresponds to a group of preset vibration initial amplitude and fine adjustment step parameters. By starting the real-time feedback mechanism, the change trend of the micro-scale segregation index in the time window is continuously monitored, and a nonlinear response model is constructed by combining the sigmoid function to realize the smooth mapping of the vibration amplitude to the micro-scale segregation index. The introduction of the sigmoid function ensures the continuous change of the amplitude, avoiding the sudden change of the material disturbance. At the same time, according to the segregation rising or falling trend, the vibration fine adjustment step is automatically adjusted to realize the combination of rough response and fine control. Finally, the vibration strategy is iteratively optimized within the target frequency interval, effectively dispersing the local segregation structure and improving the distribution uniformity of the solder paste in the shear field. This method not only enhances the internal mixing state of the material, but also provides a guarantee for the subsequent printing layer quality consistency, and is a key link to build a high-robustness and high-intelligent solder paste printing system.

[0052] The present application realizes real-time spectral acquisition of the solder paste in the high-speed shear state in the printing process by introducing a multi-channel micro-Raman spectral array, so that not only the surface morphology can be obtained, but also the relative proportion and distribution state of the flux and tin powder particles in the material can be deeply detected. By constructing the spectral ratio, the spectral ratio difference matrix between multiple measurement points, and the composition drift intensity factor based on the peak position shift, high-resolution and multi-dimensional characterization of the micro-component fluctuation of the solder paste is realized, and the technical bottleneck that the existing two-dimensional image recognition technology cannot detect internal segregation structure is overcome. In addition, after spatial coupling of these composition change information and the shear rate field, the generated micro-scale segregation index has the advantages of quantifiability, distributability and traceability, which provides theoretical support and data basis for accurately identifying the local segregation risk area, and greatly improves the accuracy and scientificity of the printing quality control.

[0053] The application establishes a physical and data double-layer coupling relationship between the component segregation state and the equipment vibration characteristics by constructing a micro-scale segregation index-vibration response frequency band mapping model, and further designs a nonlinear amplitude adjustment mechanism based on a sigmoid function, which can dynamically adjust the vibration parameters according to the numerical level and trend of the micro-scale segregation index. When the segregation risk occurs, the system can enter the feedback control mode in real time, and adaptively enhance the mixing effect of the local doctor blade shear region in the optimal response frequency interval, so as to realize the immediate intervention and trend reversal of segregation. This mechanism not only avoids the continuous accumulation of material defects, but also realizes the closed-loop conversion from 'detecting segregation' to 'controlling segregation', greatly improves the intelligent level, stability and welding reliability of the solder paste printing process, and is especially suitable for high-pin-density packaging products in high-consistency scenarios of solder paste.

[0054] The application provides a multi-modal data analysis system for a solder paste printing process as shown in Figure 2 The application provides a multi-modal data analysis system for a solder paste printing process as shown in The spectral data acquisition module acquires the transient Raman spectrum response data of the solder paste at each spatial measurement point in a high-speed shearing state in real time. The spectral ratio construction module extracts the spectral intensity ratio of each spatial measurement point in the characteristic peak interval, and constructs the spectral ratio representing the proportion relationship between the flux and the tin powder at the measurement point. The component-shearing correlation modeling module constructs a spectral ratio difference matrix among multiple measurement points, and generates a shearing rate mapping model according to the doctor blade operation parameters, so as to spatially correlate the spectral ratio corresponding to each measurement point in the spectral ratio difference matrix with the shearing rate information at the position. The component drift evaluation module defines a component deviation vector across spatial measurement points, and calculates a component drift intensity factor representing the overall component fluctuation degree based on the component deviation vector. The segregation calculation module performs tensor fusion operation on the component drift intensity factor and the shearing gradient tensor corresponding to each measurement point in the shearing rate mapping model, to obtain a micro-scale segregation index. The vibration mode recognition module judges whether the micro-scale segregation index exceeds a preset segregation threshold; if yes, the vibration response spectrum data of the solder paste printing area are acquired in real time, and a low-frequency coupled vibration mode highly related to the micro-scale segregation index is recognized from the vibration response spectrum data, to construct a mapping model between the vibration frequency band and the segregation intensity. The intelligent vibration control module determines the optimal response vibration frequency interval corresponding to the current segregation state according to the mapping model, and dynamically adjusts the vibration amplitude and the fine adjustment step length by using a sigmoid function in combination with the micro-scale segregation index.

[0055] The multi-modal data analysis method for the solder paste printing process provided by the embodiment of the present application is realized through the multi-modal data analysis system for the solder paste printing process, and the specific method and process of the multi-modal data analysis system for the solder paste printing process are described in the above embodiment of the multi-modal data analysis method for the solder paste printing process, which will not be described here.

[0056] The above only describes certain exemplary embodiments of the present application by way of illustration, and it is needless to say that the described embodiments can be modified in various ways without departing from the spirit and scope of the present application for those skilled in the art. Therefore, the above drawings and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A method of multi-modal data analysis for a solder paste printing process, characterized in that, The method comprises the following steps: S101, real-time collection of transient Raman spectrum response data of each spatial measurement point of the solder paste in a high-speed shearing state; S102, extraction of spectral intensity ratio of each spatial measurement point in the characteristic peak interval, and construction of a spectral ratio representing the proportion relationship between the flux and the tin powder at the measurement point; S103, construction of a spectral ratio difference matrix between multiple measurement points, and generation of a shearing rate mapping model according to the scraper running parameters, so as to spatially correlate the spectral ratio of each measurement point in the spectral ratio difference matrix with the shearing rate information at the position of the measurement point; S104, definition of a component deviation vector across the spatial measurement points, and calculation of a component drift intensity factor representing the overall component fluctuation degree based on the component deviation vector; S105, tensor fusion operation of the component drift intensity factor and the shearing gradient tensor corresponding to each measurement point in the shearing rate mapping model, to obtain a micro-scale segregation index; S106, judgment of whether the micro-scale segregation index exceeds a preset segregation threshold; If yes, real-time collection of vibration response spectrum data of the solder paste printing area, identification of a low-frequency coupled vibration mode highly related to the micro-scale segregation index from the vibration response spectrum data, and construction of a mapping model between the vibration frequency band and the segregation intensity; S107, determination of a best response vibration frequency interval corresponding to a current segregation state according to the mapping model, and dynamic adjustment of the vibration amplitude and the fine adjustment step length by using a sigmoid function in combination with the micro-scale segregation index.

2. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of arranging the multi-channel micro Raman spectrum array in the scraper action area are as follows: According to the modeling results of the rheological behavior of the solder paste under different printing speeds, the spatial distribution density of the Raman spectrum measurement points in the scraper area is preset; A plurality of micro Raman spectrum acquisition units are arranged in the front edge, middle and rear edge areas of the scraper respectively, so as to cover the key stress field change sections inside the solder paste in the high-speed shearing state; All the arranged micro Raman spectrum acquisition units are synchronously triggered, so that each measurement point accurately captures the corresponding transient Raman spectrum in different time slices in the solder paste shearing process.

3. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of extracting the spectral intensity ratio of each spatial measurement point in the characteristic peak interval and constructing the spectral ratio are as follows: Based on the molecular structure information of the known active flux and tin powder particles in the solder paste, the corresponding reference characteristic peak interval is determined in the Raman spectrum database respectively, and the peak position drift tolerance range in actual measurement is confirmed through pre-experiment calibration; The transient Raman spectrum collected by each spatial measurement point is subjected to background noise reduction and normalization processing; The maximum peak intensity of the flux peak and the tin powder peak in the corresponding interval is extracted from the processed Raman spectrum, the integral intensity thereof is calculated respectively, and the spectral intensity ratio corresponding to the measurement point is established; The spectral intensity ratio is combined with the peak shape change factor to construct the spectral ratio as a quantitative characteristic index of the proportion relationship between the active flux and the tin powder at the measurement point.

4. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of constructing the spectral ratio difference matrix and spatially correlating the spectral ratio with the shearing rate mapping model are as follows: According to the spatial coordinate sequence of each Raman spectrum measurement point in the solder paste printing area, the spectral ratio corresponding to each measurement point is extracted and arranged to form an initial spectral ratio distribution matrix; Based on the spectral ratio distribution matrix, the spectral ratio difference between each pair of adjacent measuring points is calculated, a spectral ratio difference matrix for representing the local component change amplitude is constructed, and the component mutation region is identified by combining the spatial gradient algorithm; According to the real-time running parameters of the scraper, a spatial mapping model of the shear rate field is constructed, and each measuring point is assigned a local shear rate value in its region; Each data unit in the spectral ratio difference matrix is paired with its corresponding shear rate to establish a spatial correlation matrix for describing the relationship between shear conditions and component changes.

5. The method of multi-modal data analysis for a solder paste printing process of claim 4, wherein, The component mutation region is identified by combining the spatial gradient algorithm, and the specific implementation steps are as follows: The spectral ratio corresponding to each Raman measuring point is rearranged according to its spatial coordinates to construct a two-dimensional component distribution matrix; The local derivative of the spectral ratio difference matrix is calculated using the spatial gradient algorithm to extract the spectral ratio change rate of each measuring point in its neighborhood and obtain the spatial gradient field of the spectral ratio; By setting a threshold judgment rule, the measuring points with gradient amplitude exceeding the set threshold are marked as component mutation regions.

6. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The specific steps of calculating the component drift intensity factor representing the overall component fluctuation degree based on the component deviation vector are as follows: Extract the spectral ratio and its corresponding characteristic peak position of all Raman measuring points in the scraper action area to establish a multi-dimensional spectral state vector set between measuring points; Select a group of reference measuring points as the reference vector in the ideal component state, and calculate the vector difference between each measuring point and the reference point based on the reference vector to construct a component deviation vector field reflecting the component change trend of each point; Based on the component deviation vector field, the modulus of all deviation vectors is calculated using the Euclidean norm, and statistical normalization processing is performed; The mean value of the statistical processing modulus is taken as the component drift intensity factor.

7. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of tensor fusion operation and micro-scale segregation index calculation between the component drift intensity factor and the shear gradient tensor in the shear rate mapping model are as follows: For each Raman spectrum measuring point, obtain its component drift intensity factor, and extract the shear gradient tensor corresponding to the position of the measuring point from the shear rate mapping model; Perform dimension standardization processing on the component drift intensity factor and the shear gradient tensor, and expand the feature vectors; The component drift intensity factor and the shear gradient tensor are fused at the corresponding measuring point by tensor inner product to obtain the tensor response intensity value of the measuring point; Based on the weighted average enhancement aggregation strategy, the micro-scale segregation index is calculated by integrating the tensor response intensity values of all measuring points.

8. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of constructing the mapping model between the vibration frequency band and the segregation intensity are as follows: Set the allowed segregation threshold in the solder paste printing process, and monitor the micro-scale segregation index of each measuring point in real time during printing to determine whether it exceeds the segregation threshold; When the micro-scale segregation index of a certain region exceeds the segregation threshold, the vibration response spectrum data containing the amplitude-frequency characteristics are obtained by collecting the vibration response spectrum of the region at the printing moment; The collected vibration spectrum data is subjected to Fourier transform and frequency band feature extraction, low-frequency vibration components with high linear correlation with the change trend of the micro-scale segregation index are screened out, and are marked as low-frequency coupled vibration modes; Based on the mapping relationship between different micro-scale segregation indexes and corresponding low-frequency vibration energy distribution, a function model between vibration frequency band and segregation intensity is constructed.

9. The method of multi-modal data analysis for a solder paste printing process of claim 1, wherein, The steps of dynamically adjusting the vibration amplitude and the fine adjustment step length based on the numerical value of the micro-scale segregation index are as follows: Based on the mapping model between the vibration frequency band and the segregation intensity, the best response vibration frequency interval corresponding to the currently calculated micro-scale segregation index is found, and the frequency interval is set as the target working frequency band of the current control system; According to the interval division strategy based on the numerical range of the micro-scale segregation index, the micro-scale segregation index is divided into multiple adjustment levels, each level corresponding to a set of vibration initial amplitude and fine adjustment step length parameter table; The real-time feedback control mechanism is started, the change rate of the micro-scale segregation index in the time sequence window is combined, the vibration control strategy is dynamically corrected, the non-linear amplitude modulation model is constructed in the best response vibration frequency interval combined with the sigmoid function, the smooth response relationship between the vibration amplitude and the micro-scale segregation index is formed, and the vibration amplitude and the fine adjustment step length are iteratively adjusted.

10. A multi-modal data analysis system for a solder paste printing process for implementing the multi-modal data analysis method for a solder paste printing process according to any one of the preceding claims 1 to 9, characterized in that, It comprises a spectroscopy data acquisition module, a spectroscopy ratio construction module, a composition-shear correlation modeling module, a composition drift evaluation module, a segregation calculation module, a vibration mode recognition module, and an intelligent vibration regulation module. The spectroscopy data acquisition module acquires real-time transient Raman spectrum response data of the solder paste at each spatial measurement point under high-speed shearing state. The spectroscopy ratio construction module extracts the spectral intensity ratio of each spatial measurement point in the characteristic peak interval, and constructs a spectroscopy ratio representing the proportion relationship between the flux and the tin powder at the measurement point. The composition-shear correlation modeling module constructs a spectroscopy ratio difference matrix among multiple measurement points, and generates a shear rate mapping model according to the scraper running parameters, so as to spatially correlate the spectroscopy ratio corresponding to each measurement point in the spectroscopy ratio difference matrix with the shear rate information at its location. The composition drift evaluation module defines a composition deviation vector across the spatial measurement points, and calculates a composition drift intensity factor representing the overall composition fluctuation degree based on the composition deviation vector. The segregation calculation module performs tensor fusion operation on the composition drift intensity factor and the shear gradient tensor corresponding to each measurement point in the shear rate mapping model to obtain the micro-scale segregation index. The vibration mode recognition module determines whether the micro-scale segregation index exceeds a preset segregation threshold. If it exceeds, the vibration response spectrum data of the solder paste printing area is acquired in real time, and the low-frequency coupled vibration mode with high correlation with the micro-scale segregation index is recognized therefrom to construct a mapping model between the vibration frequency band and the segregation intensity. The intelligent vibration regulation module determines the best response vibration frequency interval corresponding to the current segregation state according to the mapping model, and dynamically adjusts the vibration amplitude and the fine adjustment step length using the sigmoid function based on the micro-scale segregation index.

Images