Integrated Moe structure large model optical communication carrier plate detection intelligent analysis system

Abstract

The invention relates to the technical field of photoelectron detection, in particular to an integrated Moe structure large model optical communication carrier plate detection intelligent analysis system which comprises an optical signal dynamic acquisition module, a multi-dimensional feature preprocessing module, a Moe structure dynamic analysis module, a defect map reconstruction module and an intelligent diagnosis output module. Wherein the optical signal dynamic acquisition module is used for acquiring original optical signal data on the surface of a carrier plate; the multi-dimensional feature preprocessing module is used for outputting a multi-dimensional feature tensor; the Moe structure dynamic analysis module is used for outputting a space thermodynamic diagram; the defect map reconstruction module is used for generating a three-dimensional defect distribution map; and the intelligent diagnosis output module is used for outputting a structured diagnosis report. According to the invention, by constructing an intelligent analysis system fusing multi-dimensional optical feature analysis and expert dynamic modeling, linkage of high-precision three-dimensional reconstruction and repair decision of the defects of the optical communication carrier plate is realized, and the detection accuracy and the diagnosis execution efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optoelectronic detection, and particularly to an intelligent analysis system for detecting optical communication carrier boards integrated with a large model of Moe structure. Background Art

[0002] With the development of high-frequency and high-speed optical communication technology, as a key component of the core optoelectronic interconnection structure, the microscopic defects of optical communication carrier boards have a significant impact on signal integrity, optical path consistency, and the overall system reliability; during the manufacturing and assembly processes of carrier boards, typical defects such as scratches, bubbles, and delamination frequently occur, with complex morphologies and three-dimensional distributions, making it difficult to achieve precise identification and in-depth positioning through traditional two-dimensional imaging or surface vision detection means.

[0003] Current mainstream detection methods mostly rely on manual experience for auxiliary judgment or simple image feature extraction, which are neither capable of identifying deep structural defects nor providing traceable repair decision-making bases, seriously restricting the high-quality delivery and long-term stable operation of optical communication systems. Therefore, there is an urgent need for an intelligent analysis system for detecting optical communication carrier boards integrated with a large model of Moe structure to solve the above problems. Summary of the Invention

[0004] Based on the above purpose, the present invention provides an intelligent analysis system for detecting optical communication carrier boards integrated with a large model of Moe structure.

[0005] An intelligent analysis system for detecting optical communication carrier boards integrated with a large model of Moe structure includes an optical signal dynamic acquisition module, a multi-dimensional feature preprocessing module, a Moe structure dynamic analysis module, a defect map reconstruction module, and an intelligent diagnosis output module; where:

[0006] The optical signal dynamic acquisition module: is used to obtain the original optical signal data on the surface of the carrier board through a confocal laser scanning device, and output a three-dimensional optical field matrix including scattering intensity, phase offset, and polarization angle;

[0007] The multi-dimensional feature preprocessing module: is used to receive the three-dimensional optical field matrix, generate an enhanced optical field matrix through a wavelet domain dynamic noise suppression algorithm, and extract frequency domain energy distribution features and spatial gradient features, and output a multi-dimensional feature tensor;

[0008] The Moe structure dynamic analysis module: is used to input the multi-dimensional feature tensor into a large model of Moe structure composed of N expert sub-networks, and each expert sub-network dynamically assigns weights for scratch, bubble, and delamination defect types respectively, and outputs a defect probability distribution vector and a spatial heat map;

[0009] The defect map reconstruction module: based on the spatial heat map, combines the refractive index parameters of the carrier board material, and reconstructs the three-dimensional spatial coordinates of the defect through an inverse ray tracing algorithm to generate a three-dimensional defect distribution map with defect type labels;

[0010] Intelligent diagnosis output module: used to match a preset process parameter library according to the three-dimensional defect distribution map and output a structured diagnosis report including defect location, type, and repair suggestions.

[0011] Optionally, the optical signal dynamic acquisition module includes a laser emission unit, a confocal scanning unit, an optical demodulation unit, and a data conversion unit; wherein:

[0012] Laser emission unit: used to emit a linearly polarized laser beam with continuous intensity at a central wavelength of 532 nm, and the output power of the laser beam is controlled within the range of 10 mW to 50 mW;

[0013] Confocal scanning unit: used to perform point-by-point laser focusing scanning on the surface of the carrier board with an optical axial resolution better than 1 μm under the support of an XYZ three-dimensional adjustable platform, and obtain the reflected and scattered return signals at each scanning point;

[0014] Optical demodulation unit: used to perform interference measurement analysis on the scanned return optical signal, extract the scattering intensity, phase shift amount, and polarization angle parameters of each point, and eliminate the non-linear interference components;

[0015] Data conversion unit: used to convert the three types of optical measurement parameters extracted into a three-dimensional light field matrix form with a unified structure.

[0016] Optionally, the multi-dimensional feature preprocessing module includes a wavelet decomposition unit, a threshold noise suppression unit, a reconstruction generation unit, a frequency domain feature extraction unit, a spatial gradient feature extraction unit, and a feature tensor assembly unit; wherein:

[0017] Wavelet decomposition unit: receives the three-dimensional light field matrix output by the optical signal dynamic acquisition module, performs three-level two-dimensional discrete wavelet transform on each two-dimensional slice, and obtains high-frequency coefficients of three scales and low-frequency coefficients of one scale;

[0018] Threshold noise suppression unit: used to denoise the high-frequency coefficients of each scale by the fixed hard threshold method. The hard threshold of the first scale is set to 0.018, the hard threshold of the second scale is set to 0.011, and the hard threshold of the third scale is set to 0.006. The coefficients with amplitudes lower than the corresponding thresholds are directly set to zero, and the coefficients with amplitudes higher than the thresholds remain unchanged;

[0019] Reconstruction generation unit: used to perform inverse wavelet transform on the high-frequency coefficients after threshold noise suppression processing and the original low-frequency coefficients to generate an enhanced light field matrix;

[0020] Frequency domain feature extraction unit: It is used to perform three-dimensional fast Fourier transform on the enhanced light field matrix, and statistically calculate the energy density in three wavenumber segments of 0–1 / 3Nyquist, 1 / 3–2 / 3 Nyquist, and 2 / 3–Nyquist to form a frequency domain energy distribution feature vector;

[0021] Spatial gradient feature extraction unit: It is used to calculate the first-order differential in the spatial domain along the x, y, and z directions, summarize to obtain a three-dimensional gradient amplitude array, and perform max pooling on the array to compress it into a fixed-length gradient feature vector;

[0022] Feature tensor assembly unit: It is used to splice the frequency domain energy distribution feature vector and the spatial gradient feature vector in the index order, and output a multi-dimensional feature tensor in a unified format.

[0023] Optionally, the Moe structure dynamic analysis module includes a feature distribution unit, a group of expert sub-networks, a dynamic weighted fusion unit, a defect probability decoding unit, and a spatial heat map unit; among them:

[0024] Feature distribution unit: It is used to receive the multi-dimensional feature tensor output by the multi-dimensional feature preprocessing module, calculate the correlation weight coefficients of N expert sub-networks through a fully connected gating network, and broadcast the complete feature tensor to all expert sub-networks in the same clock cycle;

[0025] Expert sub-network group unit: It includes a first expert sub-network for scratch defects, a second expert sub-network for bubble defects, and a third expert sub-network for delamination defects, and is used to output a confidence score vector and a spatial attention map corresponding to the defect type;

[0026] Dynamic weighted fusion unit: It is used to perform weighted summation on the confidence score vector and the spatial attention map output by each expert sub-network according to the correlation weight coefficients generated by the feature distribution unit to obtain a global defect probability intermediate result and a fused spatial attention matrix;

[0027] Defect probability decoding unit: It is used to perform normalization processing on the global defect probability intermediate result to form a one-dimensional defect probability distribution vector, where each component corresponds to the occurrence probability of three types of defects: scratches, bubbles, and delamination;

[0028] Spatial heat map unit: It is used to perform channel-level superposition on the fused spatial attention matrix and the enhanced light field matrix to generate a spatial heat map corresponding one-to-one to the carrier board size, and the heat value ranges from 0 to 1.

[0029] Optionally, the feature distribution unit includes:

[0030] Input normalization subunit: used to flatten a multi-dimensional feature tensor into a one-dimensional vector, calculate the mean and standard deviation of the vector, and then complete the linear normalization of the vector;

[0031] Gated activation subunit: used to input the normalized feature vector into a fully-connected gated network, calculate the activation vector through a weight matrix and a bias vector, and the formula is; , where, is the weight matrix; is the normalized feature vector; is the bias vector; is the activation vector;

[0032] Probability mapping subunit: used to apply a Soft-Max mapping to the activation vector to generate the correlation weight coefficients corresponding to each expert sub-network, and the formula is; , where, is the weight coefficient of the th expert sub-network; is the th element of the activation vector; is the th element of the activation vector.

[0033] Optionally, the dynamic weighted fusion unit includes:

[0034] Score fusion subunit: used to receive the correlation weight coefficients output by the feature distribution unit and the confidence score vectors output by each expert sub-network, and perform weighted superposition on the scores corresponding to each expert sub-network according to the weight coefficients to obtain an intermediate result of the global defect probability; the formula is: , where, is the intermediate result vector of the global defect probability; is the confidence score vector of the th expert sub-network; is the number of expert sub-networks;

[0035] Thermal fusion subunit: used to perform pixel-by-pixel weighted summation on the spatial attention maps output by each expert sub-network according to the corresponding weight coefficients to obtain a fused spatial attention matrix, which is used to reflect the global defect spatial distribution trend; the formula is: , where, is the value of the fused spatial attention matrix at position ; is the th expert sub-network's spatial attention map at position value.

[0036] Optionally, the defect map reconstruction module includes a thermal threshold segmentation unit, a refractive index correction unit, a reverse ray tracing unit, a coordinate fusion unit, a defect annotation unit, and a map output unit; specifically:

[0037] The thermal threshold segmentation unit: is configured to receive the spatial thermal map generated by the dynamic weighted fusion unit, divide the surface of the carrier plate into a defect candidate area and a background area according to the thermal value threshold, and record the thermal peak position for each candidate area;

[0038] The refractive index correction unit: is configured to read the pre-stored refractive index parameters of the carrier plate material and perform optical path refraction correction on the thermal peak position of the candidate area;

[0039] The reverse ray tracing unit: is configured to start from the surface incident coordinates after refraction correction, perform reverse ray tracing along the incident light direction with a pixel step size, search for the thermal value increasing path in the three-dimensional voxel space until the thermal gradient drops to zero, and determine the defect voxel center position;

[0040] The coordinate fusion unit: is configured to perform spatial clustering on the voxel center positions output by the reverse ray tracing unit, calculate the geometric centroid of each cluster, and generate a list of three-dimensional spatial coordinates of the defects;

[0041] The defect annotation unit: calls the defect probability distribution vector output by the defect probability decoding unit, matches the defect type with the highest probability with the corresponding three-dimensional coordinates, and forms a defect entity record with type annotation;

[0042] The map output unit: is configured to map the defect entity records to a unified three-dimensional model coordinate system according to the three-dimensional coordinates, and generate a three-dimensional defect distribution map with defect type annotation.

[0043] Optionally, the reverse ray tracing unit includes:

[0044] The ray initialization sub-unit: is configured to use the surface incident coordinates after refraction correction as the starting position vector, and read the unit direction vector of the incident light , and set the pixel-level step size scalar ;

[0045] The step iteration sub-unit: is configured to perform reverse stepping along the direction vector in the three-dimensional voxel space with a fixed step size, update the current position, and the update formula is: , where is the position information after the th iteration; is the position information at the th iteration;

[0046] The gradient determination sub-unit: is configured to obtain the thermal value of the current position after each step , while calculating the local thermal gradient , when , a stop flag is triggered;

[0047] Center positioning subunit: used to record the position vector of the maximum value in the thermal value sequence , and output it as the center coordinates of the defective voxel.

[0048] Optionally, the coordinate fusion unit includes:

[0049] Initial clustering subunit: used to perform a single-round density clustering on the set of voxel center coordinates output by the reverse ray tracing subunit according to the Euclidean distance threshold, and calculate the Euclidean distance between any two points and ; when , when is less than or equal to the preset distance threshold , then these two points are classified into the same initial cluster;

[0050] Cluster merging subunit: used to compare all initial clusters pairwise. If the minimum distance between clusters is not greater than , then the corresponding clusters are merged until the minimum distance between any two clusters is greater than ;

[0051] Centroid calculation subunit: used to calculate the geometric centroid of the voxel center coordinates within the th cluster after merging. The formula is: , where is the geometric centroid coordinates of the th cluster; is the number of voxel center coordinates within the th cluster; is the voxel center coordinate of the th individual within the cluster; ;

[0052] Inventory output subunit: used to construct a list of three-dimensional defect space coordinates according to the order generated by the centroid calculation subunit, and assign a cluster identification number to each record.

[0053] Optionally, the intelligent diagnosis output module includes a defect attribute extraction unit, a parameter matching unit, a repair suggestion generation unit, and a report structured output unit; among them:

[0054] Defect attribute extraction unit: used to receive the three-dimensional defect distribution map output by the defect map reconstruction module, extract the spatial coordinates, type identification, and distribution scale information of each defect, and convert them into a structured defect attribute tuple; each attribute tuple includes the corresponding position coordinates, defect type label, volume estimation value, and local thermal significance level;

[0055] Parameter matching unit: It is used to retrieve the optimal matching item in the preset process parameter library according to the type tag and volume estimation value in the defect attribute tuple. The process parameter library is indexed by defect type + size range and stores the corresponding processing process code, process feasibility level, and recommended processing path. The successfully matched item outputs the recommended processing process entry corresponding to the defect.

[0056] Repair suggestion generation unit: It is used to combine the processing process entry returned by the parameter matching unit and formulate repair suggestions for the corresponding defects according to the processing equipment resources, repair priority rules, and defect spatial distribution density, including processing methods, recommended tool models, recommended time windows, and precautions.

[0057] Report structured output unit: It is used to classify, code, and format the spatial coordinates, defect types, matching processing processes, and repair suggestions of all defects, and generate a structured diagnostic report according to the unified report template.

[0058] Advantages of the present invention:

[0059] In the present invention, by constructing an intelligent analysis system integrating the Moe structure large model, introducing confocal laser scanning, multi-dimensional feature preprocessing, expert network dynamic fusion, defect map reconstruction, and diagnostic output modules, high-precision identification and three-dimensional visualization positioning of surface and subsurface defects of optical communication carrier boards are realized. The system can automatically classify and judge different types of defects, output probability distributions and spatial heat maps, significantly improving the accuracy and spatial analysis ability of detection results.

[0060] In the present invention, by combining the defect map and material refractive index parameters, the three-dimensional coordinates of the defect are accurately reconstructed using the reverse ray tracing algorithm, and it is linked with the preset process parameter library to automatically generate a structured diagnostic report including defect location, type, and processing suggestions, forming an intelligent closed-loop from detection to repair suggestion output, greatly improving the fault handling efficiency and process response accuracy. Description of the drawings

[0061] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0062] Figure 1 Schematic diagram of the intelligent analysis system for optical communication carrier board detection according to the embodiment of the present invention;

[0063] Figure 2 Schematic diagram of the Moe structure dynamic analysis module according to the embodiment of the present invention. Detailed implementation mode

[0064] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments, and are not intended to specifically limit the present invention.

[0065] As Figure 1 - Figure 2 shown, an intelligent analysis system for detecting an optical communication carrier board of an integrated Moe structure large model includes an optical signal dynamic acquisition module, a multi-dimensional feature preprocessing module, a Moe structure dynamic analysis module, a defect map reconstruction module, and an intelligent diagnosis output module; among them:

[0066] Optical signal dynamic acquisition module: used to obtain the original optical signal data on the surface of the carrier board through a confocal laser scanning device, and output a three-dimensional optical field matrix including scattering intensity, phase offset, and polarization angle;

[0067] Multi-dimensional feature preprocessing module: used to receive the three-dimensional optical field matrix, generate an enhanced optical field matrix through a wavelet domain dynamic noise suppression algorithm, and extract frequency domain energy distribution features and spatial gradient features, and output a multi-dimensional feature tensor;

[0068] Moe structure dynamic analysis module: used to input the multi-dimensional feature tensor into a Moe structure large model composed of N expert sub-networks, and each expert sub-network dynamically assigns weights for scratch, bubble, and delamination defect types respectively, and outputs a defect probability distribution vector and a spatial heat map;

[0069] Defect map reconstruction module: based on the spatial heat map, combined with the refractive index parameter of the carrier board material, reconstruct the three-dimensional spatial coordinates of the defect through the reverse ray tracing algorithm, and generate a three-dimensional defect distribution map with defect type labels;

[0070] Intelligent diagnosis output module: used to match the preset process parameter library according to the three-dimensional defect distribution map, and output a structured diagnosis report including defect location, type, and repair suggestions.

[0071] The optical signal dynamic acquisition module includes a laser emission unit, a confocal scanning unit, an optical demodulation unit, and a data conversion unit; among them:

[0072] Laser emission unit: used to emit a linearly polarized laser beam with a continuous intensity at a central wavelength of 532 nm, and the output power of the laser beam is controlled within the range of 10 mW to 50 mW to ensure sufficient irradiation penetration for the surface structure of the carrier board;

[0073] Confocal scanning unit: Used to perform point-by-point laser focusing scanning on the surface of the carrier plate with an optical axial resolution better than 1μm under the support of an XYZ three-dimensional adjustable platform, and obtain reflected and scattered return signals at each scanning point;

[0074] Optical demodulation unit: Used to perform interference measurement and analysis on the optical signals returned by scanning, extract the scattering intensity, phase offset and polarization angle parameters of each point, and eliminate non-linear interference components;

[0075] Data conversion unit: Used to convert the three types of optical measurement parameters extracted into a three-dimensional light field matrix form with a unified structure; The above units can organically integrate the functions of laser emission, spatial scanning, optical demodulation and data matrix conversion, and can accurately obtain three types of optical response characteristics on the surface of the carrier plate at the micron-level resolution, form a three-dimensional light field matrix in a unified format, and provide a structured and high signal-to-noise ratio input basis for subsequent defect identification and modeling analysis.

[0076] The multi-dimensional feature preprocessing module includes a wavelet decomposition unit, a threshold denoising unit, a reconstruction generation unit, a frequency domain feature extraction unit, a spatial gradient feature extraction unit, and a feature tensor assembly unit; Among them:

[0077] Wavelet decomposition unit: Receives the three-dimensional light field matrix output by the optical signal dynamic acquisition module, performs three-level two-dimensional discrete wavelet transform on each two-dimensional slice, and obtains high-frequency coefficients of three scales and low-frequency coefficients of one scale;

[0078] Threshold denoising unit: Used to denoise the high-frequency coefficients of each scale by the fixed hard threshold method. The hard threshold of the first scale is set to 0.018, the hard threshold of the second scale is set to 0.011, and the hard threshold of the third scale is set to 0.006. The coefficients with amplitudes lower than the corresponding thresholds are directly set to zero, and the coefficients with amplitudes higher than the thresholds remain unchanged;

[0079] Reconstruction generation unit: Used to perform inverse wavelet transform on the high-frequency coefficients after threshold denoising processing and the original low-frequency coefficients to generate an enhanced light field matrix with improved signal-to-noise ratio;

[0080] Frequency domain feature extraction unit: Used to perform three-dimensional fast Fourier transform on the enhanced light field matrix, and statistically calculate the energy density in the three wavenumber segments of 0–1 / 3Nyquist, 1 / 3–2 / 3 Nyquist, and 2 / 3–Nyquist to form a frequency domain energy distribution feature vector;

[0081] Spatial gradient feature extraction unit: Used to calculate the first-order differential in the spatial domain along the x, y, and z directions, summarize to obtain a three-dimensional gradient amplitude array, and perform max pooling on the array to compress it into a fixed-length gradient feature vector;

[0082] Feature Tensor Assembly Unit: It is used to splice the frequency-domain energy distribution feature vectors and spatial gradient feature vectors in the index order and output a multi-dimensional feature tensor in a unified format for the Moe structure dynamic analysis module to call; the above unit performs hierarchical wavelet denoising by setting definite numerical hard thresholds for three layers of high-frequency coefficients. This multi-dimensional feature preprocessing module removes random noise and periodic stripe noise while maintaining the defect edge details, increasing the signal-to-noise ratio of the enhanced light field matrix by about 9 dB, thereby improving the accuracy of subsequent defect recognition.

[0083] The Moe structure dynamic analysis module includes a feature distribution unit, a group of expert sub-networks, a dynamic weighted fusion unit, a defect probability decoding unit, and a spatial heat map unit; among them:

[0084] Feature Distribution Unit: It is used to receive the multi-dimensional feature tensor output by the multi-dimensional feature preprocessing module, calculate the correlation weight coefficients of N expert sub-networks through a fully connected gating network, and broadcast the complete feature tensor to all expert sub-networks simultaneously within the same clock cycle;

[0085] Group of Expert Sub-network Units: It includes a first expert sub-network for scratch defects, a second expert sub-network for bubble defects, and a third expert sub-network for delamination defects. Each expert sub-network adopts a convolutional attention structure and completes end-to-end training on its exclusive defect dataset, and is used to output a confidence score vector and a spatial attention map corresponding to the defect type;

[0086] Dynamic Weighted Fusion Unit: It is used to perform weighted summation on the confidence score vectors and spatial attention maps output by each expert sub-network according to the correlation weight coefficients generated by the feature distribution unit to obtain the global defect probability intermediate result and the fused spatial attention matrix;

[0087] Defect Probability Decoding Unit: It is used to perform normalization processing on the global defect probability intermediate result to form a one-dimensional defect probability distribution vector, where each component corresponds to the occurrence probabilities of three types of defects: scratches, bubbles, and delamination;

[0088] Spatial Heat Map Unit: It is used to perform channel-level stacking on the fused spatial attention matrix and the enhanced light field matrix to generate a spatial heat map corresponding one-to-one to the carrier board size, and the heat value ranges from 0 to 1, which is used to reflect the spatial saliency of the defect appearance; through the multi-expert collaborative mechanism driven by gating weights, the Moe structure dynamic analysis module can simultaneously achieve defect probability quantification and spatial saliency localization in one forward inference process, which not only improves the defect recognition accuracy but also provides a visualized heat map result, enhancing the spatial accuracy of subsequent defect map reconstruction.

[0089] The feature distribution unit includes:

[0090] Input normalization subunit: used to flatten a multi-dimensional feature tensor into a one-dimensional vector, calculate the mean and standard deviation of the vector, and then complete the linear normalization of the vector; the normalization formula is: , where is the flattened one-dimensional feature vector; is the mean of the one-dimensional vector; is the standard deviation of the one-dimensional vector; is the normalized feature vector;

[0091] Gated activation subunit: used to input the normalized feature vector into a fully connected gated network, calculate the activation vector through a weight matrix and a bias vector, and the formula is; , where is the weight matrix; is the normalized feature vector; is the bias vector; is the activation vector;

[0092] Probability mapping subunit: used to apply a Soft-Max mapping to the activation vector to generate a correlation weight coefficient corresponding to each expert sub-network, and the formula is; , where is the weight coefficient of the th expert sub-network; is the th element of the activation vector; is the th element of the activation vector; Through the step-by-step processing of normalization, gated activation, and probability mapping by the above subunits, the feature distribution unit can output stable and interpretable weight coefficients within a single inference cycle, ensuring that each expert sub-network obtains accurate dynamic weights according to the correlation of the current sample, thereby improving the discrimination accuracy of defect probability fusion.

[0093] The dynamic weighted fusion unit includes:

[0094] Score fusion subunit: used to receive the correlation weight coefficient output by the feature distribution unit and the confidence score vector output by each expert sub-network, and weighted and superimpose the scores corresponding to each expert sub-network according to the weight coefficient to obtain an intermediate result of the global defect probability; the formula is: , where is the intermediate result vector of the global defect probability; is the th confidence score vector of the expert sub-network; is the number of expert sub-networks;

[0095] Thermal fusion subunit: It is used to perform pixel-by-pixel weighted summation on the spatial attention maps output by each expert sub-network according to the corresponding weight coefficients to obtain a fused spatial attention matrix, which is used to reflect the global defect spatial distribution trend; the formula is: , where is the value of the fused spatial attention matrix at position ; is the value of the spatial attention map of the -th expert sub-network at position ; Through the weighted calculation of the score fusion subunit and the thermal fusion subunit, the dynamic weighted fusion unit can accurately superimpose various defect judgments and spatial response results according to the feature correlation of each expert sub-network, generate an intermediate output that is both discriminative and has clear spatial distribution, and provide a consistent and high-quality input basis for defect probability decoding and thermal map generation.

[0096] The defect map reconstruction module includes a thermal threshold segmentation unit, a refractive index correction unit, a reverse ray tracing unit, a coordinate fusion unit, a defect annotation unit, and a map output unit; among them:

[0097] Thermal threshold segmentation unit: It is used to receive the spatial thermal map generated by the dynamic weighted fusion unit, divide the surface of the carrier board into a defect candidate area and a background area according to the thermal value threshold, and record the thermal peak position for each candidate area;

[0098] Refractive index correction unit: It is used to read the pre-stored refractive index parameters of the carrier board material, perform optical path refraction correction on the thermal peak position of the candidate area, and output a set of surface incident coordinates adjusted by the influence of material refraction;

[0099] Reverse ray tracing unit: It is used to start from the refraction-corrected surface incident coordinates, perform reverse ray tracing along the incident light direction at pixel step sizes, search for the thermal value increasing path in the three-dimensional voxel space until the thermal gradient drops to zero, and determine the defect voxel center position;

[0100] Coordinate fusion unit: It is used to perform spatial clustering on the voxel center positions output by the reverse ray tracing unit, calculate the geometric centroid of each clustering body, and generate a list of defect three-dimensional space coordinates;

[0101] Defect annotation unit: It calls the defect probability distribution vector output by the defect probability decoding unit, matches the defect type with the highest probability with the corresponding three-dimensional coordinates, and forms a defect entity record with type annotation;

[0102] Map Output Unit: It is used to map the defect entity records to the unified three-dimensional model coordinate system according to the three-dimensional coordinates, generate a three-dimensional defect distribution map with defect type annotations, and transfer the map to the intelligent diagnosis output module; through the hierarchical process of threshold segmentation, refractive index correction, reverse ray tracing and type annotation, the defect map reconstruction module can accurately locate the three-dimensional coordinates of the defects inside the carrier board and attach defect type information, significantly improving the resolution and classification accuracy of defect space reconstruction, and providing a reliable three-dimensional visualization basis for the subsequent structured diagnosis report.

[0103] The reverse ray tracing unit includes:

[0104] Ray initialization sub-unit: It is used to take the surface incident coordinates after refractive correction as the starting position vector, and read the unit direction vector of the incident light , and set the pixel-level step scalar ;

[0105] Stepping iteration sub-unit: It is used to perform reverse stepping along the direction vector in the three-dimensional voxel space according to a fixed step size, update the current position, and the update formula is: , where is the position information after the th iteration; is the position information at the th iteration;

[0106] Gradient determination sub-unit: It is used to obtain the thermal value of the current position after each step , and calculate the local thermal gradient , and its expression is: , where, - is the thermal gradient in the th step; correspond to the thermal values of the positions respectively, and when , the stop flag is triggered;

[0107] Center positioning sub-unit: It is used to record the maximum value position vector in the thermal value sequence , and output it as the defect voxel center coordinate; through the hierarchical process of ray initialization, step-by-step tracking, gradient monitoring and extreme value positioning, the reverse ray tracing sub-unit can determine the three-dimensional coordinates of the defect at the voxel level, avoid the positioning error caused by material refraction, and significantly improve the accuracy and stability of defect space reconstruction.

[0108] The coordinate fusion unit includes:

[0109] Initial clustering sub-unit: It is used to perform a single-round density clustering on the set of voxel center coordinates output by the reverse ray tracing sub-unit according to the Euclidean distance threshold, and calculate any two points and Euclidean distance between When is less than or equal to a preset distance threshold , then these two points are classified into the same initial cluster; the threshold takes ; The calculation formula of Euclidean distance is , where is the Euclidean distance between point and point , are the three-dimensional coordinate components of the corresponding points respectively;

[0110] Cluster merging subunit: used to compare all initial clusters pairwise. If the minimum distance between clusters is not greater than , then the corresponding clusters are merged until the minimum distance between any clusters is greater than ;

[0111] Centroid calculation subunit: used to calculate the geometric centroid of the voxel center coordinates within the rd cluster after merging. The formula is: , where is the geometric centroid coordinate of the th cluster; is the number of voxel center coordinates within the th cluster; is the voxel center coordinate of the th individual within the cluster; ;

[0112] Inventory output subunit: used to construct a list of three-dimensional coordinates of defect spaces in the order generated by the centroid calculation subunit, assign a cluster identification number to each record, and output it to the defect annotation subunit; through the hierarchical processing of Euclidean distance threshold clustering, inter-cluster merging, and centroid precise calculation by the above subunits, the coordinate fusion unit can merge discrete voxel centers into physically consistent defect entities and generate a unique and reusable three-dimensional coordinate list, significantly improving the spatial consistency and data manageability of defect location. The intelligent diagnosis output module includes a defect attribute extraction unit, a parameter matching unit, a repair suggestion generation unit, and a report structured output unit; among them:

[0113] Defect attribute extraction unit: used to receive the three-dimensional defect distribution map output by the defect map reconstruction module, extract the spatial coordinates, type identification, and distribution scale information of each defect, and convert them into structured defect attribute tuples; each attribute tuple includes the corresponding position coordinates, defect type label, volume estimation value, and local thermal significance level;

[0114]

[0115] ​Parameter matching unit: It is used to retrieve the optimal matching item in the preset process parameter library according to the type label and volume estimation value in the defect attribute tuple. The process parameter library is indexed by defect type + size range and stores the corresponding processing process code, process feasibility level, and recommended processing path. The successfully matched item outputs the recommended processing process entry corresponding to the defect.

[0116] Table 1 Example of Process Parameter Library

[0117] Number Defect type Volume range (mm³) Process code Recommended treatment process Feasibility level Treatment tool model Suggested repair time window Remarks D001 Scratch 0–0.5 SC-P1 Micro-polishing + Transparent protective layer coating Grade A (High) MP-03 Micro-polishing unit Within 6 hours Slight surface scratch, optical function not affected after repair D002 Scratch 0.5–2.0 SC-P2 Laser melting reconstruction + Coating restoration Grade B (Medium) LM-20 Laser module Within 12 hours There is a risk of light scattering interference in the moderately scratched area D003 Bubble 0–1.0 BP-F1 Thermal pressing to guide exhaust + Transparent resin backfill Grade A (High) TP-01 Heating press head Within 8 hours Local micro-bubbles, thermal pressing repair is preferred D004 Delamination 1.0–5.0 DL-S2 Ultrasonic curing agent injection + Repackaging Grade B (Medium) US-CUR Ultrasonic needle Within 24 hours The delamination structure may affect the mechanical strength of the carrier board D005 Delamination >5.0 DL-S3 Laser stripping + Structure reconstruction Grade C (Low) LS-50 Laser head Within 48 hours The structure is severely damaged, the repair cost is high, and scrapping assessment is recommended

[0118] In the above Table 1, the number represents the unique identifier for each parameter record; the defect type corresponds one-to-one with the type field output by the defect annotation subunit; the volume range represents the defect volume range adapted by the current process; the process code is a concise code for system internal calls and is bound to the process control program; the recommended processing process represents the specific defect repair method and must be a standardized process path; the feasibility level represents the processing priority given by the system based on experience and resource evaluation; the processing tool model represents the main hardware device model required to execute the process; the recommended repair time window represents the recommended optimal processing time starting from the detection of the defect; the remarks indicate supplementary explanations such as risks and priorities after processing.

[0119] Repair recommendation generation unit: It is used to combine the processing process entries returned by the parameter matching unit and formulate repair recommendations for the corresponding defects based on the processing equipment resources, repair priority rules, and defect spatial distribution density, including processing methods, recommended tool models, recommended time windows, and precautions.

[0120] Report structured output unit: It is used to classify, code, and format the spatial coordinates, defect types, matching processing processes, and repair recommendations of all defects, and generate a structured diagnostic report according to a unified report template. Each defect entry in the report contains five fields: number - location - type - repair process - processing recommendation, and provides a report export interface. Through the structured attribute extraction, standardized parameter matching, and rule-driven recommendation generation process, the intelligent diagnosis output module can not only accurately retrieve the process parameters matching the defect characteristics, but also automatically push the optimal repair plan based on the defect location and type, realizing the standardized and intelligent output of the diagnosis results, effectively supporting the rapid closed-loop processing of defects and maintenance decision-making.

[0121] The present invention encompasses any alternatives, modifications, equivalent methods, and solutions that are made within the spirit and scope of the present invention. For the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention. However, those skilled in the art can fully understand the present invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc. are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0122] The above description is only a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent analysis system for detecting optical communication carrier boards of large models with an integrated Moe structure, characterized in that, It includes an optical signal dynamic acquisition module, a multi-dimensional feature preprocessing module, a Moe structure dynamic analysis module, a defect map reconstruction module, and an intelligent diagnosis output module; among which: Optical signal dynamic acquisition module: It is used to obtain the original optical signal data on the surface of the carrier board through a confocal laser scanning device, and output a three-dimensional optical field matrix including scattering intensity, phase offset, and polarization angle; Multi-dimensional feature preprocessing module: It is used to receive the three-dimensional optical field matrix, generate an enhanced optical field matrix through a wavelet domain dynamic noise suppression algorithm, extract frequency domain energy distribution features and spatial gradient features, and output a multi-dimensional feature tensor; Moe structure dynamic analysis module: It is used to input the multi-dimensional feature tensor into a large Moe structure model composed of N expert sub-networks. Each expert sub-network dynamically assigns weights for scratch, bubble, and delamination defect types respectively, and outputs a defect probability distribution vector and a spatial heat map; Defect map reconstruction module: Based on the spatial heat map, combined with the refractive index parameter of the carrier board material, it reconstructs the three-dimensional spatial coordinates of the defect through the reverse ray tracing algorithm, and generates a three-dimensional defect distribution map with defect type annotations; Intelligent diagnosis output module: It is used to match the preset process parameter library according to the three-dimensional defect distribution map, and output a structured diagnosis report including defect location, type, and repair suggestions.

2. The intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 1, wherein The optical signal dynamic acquisition module includes a laser emission unit, a confocal scanning unit, an optical demodulation unit, and a data conversion unit; among which: Laser emission unit: It is used to emit a linearly polarized laser beam with continuous intensity at a central wavelength of 532nm, and the output power of the laser beam is controlled within the range of 10mW to 50mW; Confocal scanning unit: It is used to perform point-by-point laser focusing scanning on the surface of the carrier board with an optical axial resolution better than 1μm under the support of an XYZ three-dimensional adjustable platform, and obtain the reflected and scattered return signals at each scanning point; Optical demodulation unit: It is used to perform interference measurement analysis on the scanned return optical signal, extract the scattering intensity, phase offset, and polarization angle parameters of each point, and eliminate non-linear interference components; Data conversion unit: It is used to convert the three types of optical measurement parameters extracted into a three-dimensional optical field matrix form with a unified structure.

3. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 2, characterized in that, The multi-dimensional feature preprocessing module includes a wavelet decomposition unit, a threshold noise suppression unit, a reconstruction generation unit, a frequency domain feature extraction unit, a spatial gradient feature extraction unit, and a feature tensor assembly unit; among which: Wavelet decomposition unit: It receives the three-dimensional optical field matrix output by the optical signal dynamic acquisition module, and performs three-level two-dimensional discrete wavelet transform on each two-dimensional slice to obtain high-frequency coefficients of three scales and low-frequency coefficients of one scale; Threshold noise suppression unit: It is used to denoise the high-frequency coefficients of each scale by using the fixed hard threshold method. The hard threshold of the first scale is set to 0.018, the hard threshold of the second scale is set to 0.011, and the hard threshold of the third scale is set to 0.

006. The coefficients with amplitudes lower than the corresponding thresholds are directly set to zero, and the coefficients with amplitudes higher than the thresholds remain unchanged; Reconstruction generation unit: It is used to perform inverse wavelet transform on the high-frequency coefficients after threshold noise suppression processing and the original low-frequency coefficients to generate an enhanced optical field matrix; Frequency domain feature extraction unit: It is used to perform three-dimensional fast Fourier transform on the enhanced light field matrix, and statistically calculate the energy density in three wavenumber segments of 0–1 / 3Nyquist, 1 / 3–2 / 3 Nyquist, and 2 / 3–Nyquist to form a frequency domain energy distribution feature vector; Spatial gradient feature extraction unit: It is used to calculate the first-order differential along the x, y, and z directions in the spatial domain, summarize to obtain a three-dimensional gradient amplitude array, and perform max pooling on the array to compress it into a fixed-length gradient feature vector; Feature tensor assembly unit: It is used to splice the frequency domain energy distribution feature vector and the spatial gradient feature vector in the index order and output a multi-dimensional feature tensor in a unified format.

4. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 3, characterized in that, The Moe structure dynamic analysis module includes a feature distribution unit, an expert sub-network group, a dynamic weighted fusion unit, a defect probability decoding unit, and a spatial heat map unit; among them: Feature distribution unit: It is used to receive the multi-dimensional feature tensor output by the multi-dimensional feature preprocessing module, calculate the correlation weight coefficients of N expert sub-networks through a fully connected gating network, and broadcast the complete feature tensor to all expert sub-networks in the same clock cycle; Expert sub-network group unit: It includes a first expert sub-network for scratch defects, a second expert sub-network for bubble defects, and a third expert sub-network for delamination defects, and is used to output a confidence score vector and a spatial attention map corresponding to the defect type; Dynamic weighted fusion unit: It is used to perform weighted summation on the confidence score vector and the spatial attention map output by each expert sub-network according to the correlation weight coefficients generated by the feature distribution unit to obtain a global defect probability intermediate result and a fused spatial attention matrix; Defect probability decoding unit: It is used to perform normalization processing on the global defect probability intermediate result to form a one-dimensional defect probability distribution vector, where each component corresponds to the occurrence probability of three types of defects: scratches, bubbles, and delamination; Spatial heat map unit: It is used to perform channel-level stacking on the fused spatial attention matrix and the enhanced light field matrix to generate a spatial heat map corresponding one-to-one to the size of the carrier board, and the heat value ranges from 0 to 1.

5. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 4, characterized in that The feature distribution unit includes: Input normalization sub-unit: It is used to flatten the multi-dimensional feature tensor into a one-dimensional vector, calculate the mean and standard deviation of the vector, and then complete the linear normalization of the vector; Gating activation subunit: It is used to input the normalized feature vector into the fully connected gating network, and calculate the activation vector through the weight matrix and the bias vector. The formula is: , where is the weight matrix; is the normalized feature vector; is the bias vector; is the activation vector; Probability mapping subunit: used to apply Soft-Max mapping to the activation vector to generate correlation weight coefficients corresponding to each expert subnet, with the formula; , where is the weight coefficient of the th expert subnet; is the th element of the activation vector; is the th element of the activation vector.

6. The intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 5, wherein, The dynamic weighted fusion unit includes: Score fusion subunit: It is used to receive the correlation weight coefficients output by the feature distribution unit and the confidence score vectors output by each expert sub-network, and perform weighted superposition on the scores corresponding to each expert sub-network according to the weight coefficients to obtain the intermediate result of the global defect probability; the formula is: , where is the intermediate result vector of the global defect probability; is the confidence score vector of the th expert sub-network; is the number of expert sub-networks; Thermal fusion subunit: It is used to perform pixel-by-pixel weighted summation on the spatial attention maps output by each expert sub-network according to the corresponding weight coefficients to obtain a fused spatial attention matrix, which is used to reflect the global defect spatial distribution trend; the formula is: , where is the value of the fused spatial attention matrix at position ; is the value of the spatial attention map of the th expert sub-network at position .

7. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 4, characterized in that, The defect map reconstruction module includes a thermal threshold segmentation unit, a refractive index correction unit, a reverse ray tracing unit, a coordinate fusion unit, a defect annotation unit, and a map output unit; among them: Thermal threshold segmentation unit: It is used to receive the spatial heat map generated by the dynamic weighted fusion unit, divide the surface of the carrier board into a defect candidate area and a background area according to the thermal value threshold, and record the position of the thermal peak for each candidate area; Refractive index correction unit: It is used to read the pre-stored refractive index parameters of the carrier board material and perform optical path refraction correction on the position of the thermal peak in the candidate area; Reverse ray tracing unit: It is used to start from the surface incident coordinates after refraction correction, perform reverse ray tracing along the incident light direction with a pixel step size, search for the increasing path of the thermal value in the three-dimensional voxel space until the thermal gradient drops to zero, and determine the center position of the defective voxel; Coordinate fusion unit: It is used to perform spatial clustering on the voxel center positions output by the reverse ray tracing unit, calculate the geometric centroid of each cluster, and generate a list of three-dimensional space coordinates of defects; Defect annotation unit: It calls the defect probability distribution vector output by the defect probability decoding unit, matches the defect type with the highest probability with the corresponding three-dimensional coordinates, and forms a defect entity record with type annotation; Atlas output unit: It is used to map the defect entity records to the unified three-dimensional model coordinate system according to the three-dimensional coordinates, and generate a three-dimensional defect distribution atlas with defect type annotation.

8. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 7, characterized in that The reverse ray tracing unit includes: Ray initialization sub-unit: used to take the surface incident coordinates after refraction correction as the starting position vector, and read the unit direction vector of the incident light , and set the pixel-level step scalar ; Stepping iteration sub-unit: used to perform reverse stepping along the direction vector in the three-dimensional voxel space according to a fixed step size, update the current position, and the update formula is: , where is the position information after the th iteration; is the position information at the th iteration; Gradient determination subunit: used to obtain the thermal value of the current position after each step , and calculate the local thermal gradient , when , trigger the stop flag; Central positioning subunit: used to record the position vector of the maximum value in the thermal value sequence , and output it as the central coordinates of the defective voxel.

9. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 8, characterized in that, The coordinate fusion unit includes: Initial clustering subunit: It is used to perform a single-round density clustering on the set of voxel center coordinates output by the reverse ray tracing subunit according to the Euclidean distance threshold, and calculate the Euclidean distance between any two points and the Euclidean distance between When is less than or equal to the preset distance threshold then these two points are classified into the same initial cluster; Cluster merging subunit: used to pairwise compare all initial clusters. If the minimum distance between clusters is not greater than , the corresponding clusters are merged until the minimum distance between any clusters is greater than ; Centroid calculation subunit: used to calculate the geometric centroid of the center coordinates of voxels within the th merged cluster, and the formula is: , where , is the geometric centroid coordinate of the th cluster; is the number of voxel center coordinates within the th cluster; is the center coordinate of the th voxel within the cluster; List output subunit: used to construct a list of three-dimensional spatial coordinates of defects in the order generated by the centroid calculation subunit and assign a cluster identification number to each record.

10. An intelligent analysis system for detecting an integrated Moe structure large model optical communication carrier board according to claim 1, characterized in that, The intelligent diagnosis output module includes a defect attribute extraction unit, a parameter matching unit, a repair suggestion generation unit, and a report structured output unit; among them: Defect attribute extraction unit: It is used to receive the three-dimensional defect distribution atlas output by the defect atlas reconstruction module, extract the spatial coordinates, type identification, and distribution scale information of each defect, and convert them into a structured defect attribute tuple; each attribute tuple includes the corresponding position coordinates, defect type label, volume estimation value, and local thermal significance level; Parameter matching unit: It is used to retrieve the optimal matching item in the preset process parameter library according to the type label and volume estimation value in the defect attribute tuple. The process parameter library is indexed by defect type + size range, and stores the corresponding processing process code, process feasibility level, and recommended processing path; the output of the successful matching item is the recommended processing process entry corresponding to the defect; Repair suggestion generation unit: It is used to combine the processing process entry returned by the parameter matching unit, and formulate repair suggestions for the corresponding defects according to the processing equipment resources, repair priority rules, and defect space distribution density, including processing methods, recommended tool models, recommended time windows, and precautions; Report structured output unit: It is used to classify, code, and format the spatial coordinates, defect types, matching processing processes, and repair suggestions of all defects, and generate a structured diagnosis report according to the unified report template.

Images