Method and system for identifying alignment deviation of multi-layer board of PCB (Printed Circuit Board)

Abstract

The invention relates to the technical field of PCB production detection, and discloses a method and a system for PCB multilayer board alignment deviation identification. According to the method, the multi-layer target theoretical projection image is generated through physical modeling, frequency domain analysis is carried out in combination with the X-ray detection image and the multi-layer target theoretical projection image, and quantitative evaluation of the reliability of different areas of the image is achieved through the obtained pixel-level credibility weight matrix. Further, through a method of deeply combining credibility evaluation and image separation, noise in the image is effectively suppressed, structural information of a high-credibility region is reserved, and accurate and reliable interlayer alignment deviation data can be provided for alignment deviation recognition through the obtained single-layer target separation image and target design parameters; the problem that the alignment recognition result is inaccurate due to aliasing distortion when an X-ray detection image is directly analyzed by a traditional method is solved, and the recognition accuracy of the alignment deviation condition of the PCB is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of PCB production and inspection technology, and in particular to a method and system for identifying alignment deviations in multilayer PCB boards. Background Technology

[0002] In modern printed circuit board (PCB) manufacturing, multilayer board structures have become mainstream as electronic products evolve towards high performance, miniaturization, and high density. Interlayer alignment accuracy, a key indicator affecting PCB reliability and electrical performance, directly relates to the accuracy of circuit connections, the stability of impedance control, and the yield of the final product. In the PCB manufacturing process, the lamination process is a crucial step in molding multiple inner core boards, prepregs, and other materials into a single multilayer board under high temperature and pressure. Due to complex physical changes such as thermal expansion, resin flow, and interlayer slippage during lamination, micron-level alignment deviations inevitably occur between the circuit layers. If these interlayer deviations exceed the tolerance range, they can lead to serious quality defects such as deterioration of high-frequency signal integrity, uncontrolled characteristic impedance, and even interlayer short circuits or open circuits. Therefore, identifying interlayer alignment deviations after lamination is particularly important. Accurate deviation detection can not only promptly screen out defective products, avoiding waste of subsequent processing resources, but also provide data feedback for optimizing lamination process parameters, improving manufacturing process levels from the source.

[0003] In the current PCB manufacturing industry, the identification of interlayer alignment deviations after PCB lamination generally adopts the X-ray-based target visual inspection method. The implementation of this standard method includes three core steps: First, in the design stage, copper alignment targets are preset in the invalid areas of each PCB layer. Then, grayscale images containing the superimposed projections of all target layers are obtained through X-ray equipment. Finally, the offset of the center point of each target is directly observed and analyzed in the superimposed image. The traditional method is based on the mixed superposition of target images of each layer. Its fundamental defect is that it treats the complex multi-layer projection signal as a single image that can be directly analyzed, and cannot effectively separate the real position information of each independent layer. Due to interference factors such as material rheology and target deformation caused by the lamination process, the alignment deviation identification results will be inaccurate. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for identifying alignment deviations in multilayer PCB boards, aiming to solve the technical problems in the prior art.

[0005] This invention proposes a method for identifying alignment deviations in multilayer PCB boards, comprising: Obtain X-ray inspection images of the PCB board after multilayer board lamination, and obtain the layered structure design parameters and target design parameters of the PCB board; Based on the layered structure design parameters and the target design parameters, multi-layer target projection synthesis is performed to obtain a theoretical projection image of the multi-layer target. The frequency domain residual spectrum is obtained based on the X-ray detection image and the multi-layer target theoretical projection image, and frequency domain feature analysis is performed on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix. Based on the pixel-level confidence weight matrix and the layered structure design parameters, the X-ray detection image and the multi-layer target theoretical projection image are fused and deconvolved to obtain the deconvolution result image. Then, based on the target design parameters and the deconvolution result image, the single-layer target separation image of each target layer is obtained. Elliptical model fitting is performed on the single-layer target separation image of each target layer to obtain the corresponding optimal elliptical model, and deformation analysis is performed based on the optimal elliptical model to obtain the actual center coordinates of each target layer. The fusion weights corresponding to each adjacent layer pair are obtained according to the layered structure design parameters, and inter-layer deviation fusion is performed based on the fusion weights, target design parameters and actual center coordinates to obtain the comprehensive alignment deviation index. Determine whether the comprehensive alignment deviation index is greater than the deviation threshold; If the comprehensive alignment deviation index is greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be substandard. If the comprehensive alignment deviation index is not greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be up to standard.

[0006] This application also provides a system for identifying alignment deviations in multilayer PCB boards, comprising multiple modules for implementing the steps of the above-described method for identifying alignment deviations in multilayer PCB boards.

[0007] Preferably, the module includes multiple units, which are used to implement the steps of the above-described method for identifying alignment deviations in multilayer PCB boards.

[0008] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for identifying alignment deviations in multilayer PCB boards described above.

[0009] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for identifying alignment deviations in multilayer PCB boards described above.

[0010] The beneficial effects of this invention are as follows: This invention generates multi-layer target theoretical projection images through physical modeling, thereby establishing an accurate theoretical benchmark for identifying alignment deviations in multi-layer boards; and combines X-ray detection images and multi-layer target theoretical projection images for frequency domain analysis, achieving quantitative evaluation of the reliability of different regions of the image through a pixel-level confidence weight matrix; this invention effectively suppresses noise in the image and retains the structural information of high-confidence regions by deeply combining confidence evaluation with image separation, and accurately extracts the true position information of each independent layer from the mixed projection; at the same time, it introduces shape deviation as an evaluation index for adaptive center positioning to solve the problem of target center positioning deviation in real scenarios such as compression deformation caused by the lack of deformation perception capability in traditional methods; in the field of high-precision PCB manufacturing and quality control, especially in application scenarios with stringent requirements for inter-layer alignment accuracy, this invention can provide accurate and reliable inter-layer alignment deviation data, achieving accurate evaluation of PCB alignment deviation. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0014] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0016] like Figure 1 As shown, this application provides a method for identifying alignment deviations in multilayer PCB boards, comprising: S1. Obtain X-ray inspection images of the PCB board after multilayer board lamination, and obtain the layered structure design parameters and target design parameters of the PCB board; S2. Based on the layered structure design parameters and the target design parameters, perform multi-layer target projection synthesis to obtain a multi-layer target theoretical projection image; S3. Obtain the frequency domain residual spectrum based on the X-ray detection image and the multi-layer target theoretical projection image, and perform frequency domain feature analysis on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix. S4. Based on the pixel-level confidence weight matrix and the layered structure design parameters, the X-ray detection image and the multi-layer target theoretical projection image are fused and deconvolved to obtain the deconvolution result image, and the single-layer target separation image of each target is obtained according to the target design parameters and the deconvolution result image. S5. Fit an elliptical model to the single-layer target separation image of each target layer to obtain the corresponding optimal elliptical model, and perform deformation analysis based on the optimal elliptical model to obtain the actual center coordinates of each target layer. S6. Obtain the fusion weight corresponding to each adjacent layer pair according to the layered structure design parameters, and perform inter-layer deviation fusion based on the fusion weight, target design parameters and actual center coordinates to obtain the comprehensive alignment deviation index. S7. Determine whether the comprehensive alignment deviation index is greater than the deviation threshold; If the comprehensive alignment deviation index is greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be substandard. If the comprehensive alignment deviation index is not greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be up to standard.

[0017] As described in steps S1-S7 above, this invention establishes an observational basis for the actual manufacturing state by acquiring X-ray inspection images of the PCB multilayer board after lamination. The X-ray inspection image refers to a grayscale image acquired by an industrial CT system, containing the superimposed projections of all target layers. Simultaneously, layered structure design parameters and target design parameters are extracted from the PCB design documents. The layered structure design parameters are a set of parameters describing the PCB's stack-up structure and material physical properties, including the material type and design thickness of each dielectric layer and copper foil layer, and the corresponding attenuation coefficient obtained based on the material type using a pre-established material-X-ray attenuation relationship mapping table. The target design parameters are a set of parameters defining the target's geometric properties and spatial position, including the target design thickness, target attenuation coefficient, and target shape. The parameters, such as shape parameters and target design coordinates, are used to describe the ideal contour and spatial distribution of the target in the image. By acquiring multi-dimensional parameters, a complete understanding of the PCB physical structure and imaging characteristics can be established. Traditional detection methods directly analyze X-ray inspection images, but since X-ray inspection images are superimposed projections of all layers of targets, lines, and noise, direct analysis of X-ray inspection images will lead to inherent errors in the calculation of interlayer deviations. This invention synthesizes multi-layer target projections based on layered structure design parameters and target design parameters, and generates multi-layer target theoretical projection images through physical modeling. The multi-layer target theoretical projection image refers to the projection performance that each layer of target should have under ideal alignment. The acquired multi-layer target theoretical projection image can establish an accurate theoretical benchmark for the identification of alignment deviations in multi-layer boards.

[0018] This invention obtains the frequency domain residual spectrum from X-ray detection images and multi-layer target theoretical projection images. By performing radial energy analysis and main coupling band identification on the frequency domain residual spectrum, a pixel-level reliability weight matrix is ​​obtained. This invention achieves quantitative evaluation of the reliability of different regions of the image through the pixel-level reliability weight matrix, thus overcoming the limitations of the uniformity assumption in traditional image processing. Based on the pixel-level reliability weight matrix and layered structure design parameters, the X-ray detection image and multi-layer target theoretical projection image are fused and deconvolved. Iterative optimization guided by the pixel-level reliability weight matrix yields the deconvolution result image, thereby obtaining single-layer target separation images of each target layer. A single-layer target separation image refers to a clear image reconstructed from the mixed projection that contains only single-layer target information. This invention, by deeply combining reliability evaluation and image separation, effectively suppresses noise in the image while preserving structural information in high-reliability regions, and accurately extracts the true position information of each independent layer from the mixed projection.

[0019] This invention obtains the corresponding optimal elliptical model by fitting an elliptical model and performing deformation analysis on the single-layer target separation image of each target layer. The optimal elliptical model refers to the elliptical geometric model that best represents the spatial distribution law of the target contour. Deformation analysis is then performed on the optimal elliptical model, and based on the deformation analysis results, an adaptive center positioning strategy is used to obtain the actual center coordinates of each target layer. This mechanism, which introduces shape deviation as an evaluation index for adaptive center positioning, intelligently selects the optimal solution strategy between the ellipse center and the geometric centroid, solving the problem of target center positioning deviation in real-world scenarios such as compression deformation caused by the lack of deformation perception capability in traditional methods. Furthermore, the fusion weights of each adjacent layer pair are obtained according to the layered structure design parameters, and based on these fusion weights and the target... The design parameters and the actual center coordinates are fused to obtain a comprehensive alignment deviation index. This electrical performance-oriented deviation identification method transforms pure geometric deviation analysis into a deviation assessment directly related to the actual function of the PCB. This allows the identification results to focus on interlayer regions that have a more significant impact on circuit signal integrity. By comparing the comprehensive alignment deviation index with a deviation threshold, the alignment quality of the multilayer PCB board can be accurately determined. In the field of high-precision PCB manufacturing and quality control, especially in applications with stringent requirements for interlayer alignment accuracy, such as first article verification, periodic quality sampling, and quality arbitration, this invention can provide accurate and reliable interlayer alignment deviation data, enabling precise assessment of PCB alignment deviation.

[0020] In one embodiment, step S2, which involves synthesizing multi-layer target projections based on the layered structure design parameters and the target design parameters to obtain a theoretical projection image of the multi-layer target, includes: S21. Obtain the layered structure design thickness and layered structure attenuation coefficient according to the layered structure design parameters, and perform layer-by-layer penetration intensity simulation based on the Lambert-Beer law according to the layered structure design thickness and layered structure attenuation coefficient of each layer of the PCB board to obtain the theoretical irradiation intensity of X-rays reaching the target position of each layer. S22. Obtain the target design thickness, target attenuation coefficient, target shape parameters and target design coordinates according to the target design parameters, and simulate the target absorption intensity based on the Lambert-Beer law, the target design thickness, the target attenuation coefficient and the theoretical irradiation intensity, to obtain the X-ray emission intensity after penetrating the target. S23. Obtain the theoretical attenuation intensity of each target layer based on the theoretical irradiation intensity and the emission intensity, and construct an ideal target image based on the theoretical attenuation intensity, target shape parameters and target design coordinates to obtain multiple single-layer theoretical projection images; S24. Pixel-level grayscale superposition is performed on multiple single-layer theoretical projection images to obtain multi-layer target theoretical projection images.

[0021] As described in steps S11-S17 above, this invention obtains the layered structure design thickness and layered structure attenuation coefficient based on the layered structure design parameters. The layered structure attenuation coefficient is a coefficient characterizing the inherent blocking ability of the material against X-rays of a specific energy, including the dielectric attenuation coefficient and the copper foil attenuation coefficient. The layered structure design thickness includes the dielectric design thickness and the copper foil design thickness. Traditional methods directly analyze the acquired X-ray detection image, which is a superimposed projection of all layer targets, circuits, and noise. The fundamental flaw of this mixed image is that it assumes that the target features in the image can be directly used for interlayer alignment calculations. However, due to the severe overlap and mutual interference of target images between layers, as well as the material rheology and target deformation that may occur during the lamination process, there are inherent errors in the calculation of interlayer deviations. Since the intensity attenuation of X-rays when penetrating a multilayer PCB follows the Lambert-Beer law, i.e., the intensity of the penetrated ray is exponentially related to the incident intensity, the material attenuation coefficient, and the penetration thickness, this invention sets the initial incident intensity of the X-rays as normalized. Value 1: For the image region corresponding to the target, starting from the top layer, the attenuation of X-rays penetrating all dielectric layers and copper foil layers above the target layer is calculated sequentially. Based on the theoretical thickness and attenuation coefficient of each material layer, the Lambert-Beer law is applied layer by layer for calculation, and the output intensity of the previous layer is used as the incident intensity of the next layer. The calculation is iterated until the bottom layer, thus obtaining the theoretical irradiation intensity of X-rays when they are about to be incident on the copper target layer after penetrating all materials above the target layer. This value quantifies the softening effect of the medium itself on X-rays without considering the copper target. The target design thickness, target attenuation coefficient, target shape parameters, and target design coordinates are obtained according to the target design parameters. By taking the theoretical irradiation intensity reaching the target layer as the incident intensity, and substituting the target design thickness and target attenuation coefficient of each target layer into the Lambert-Beer law, the output intensity after penetrating the target is calculated, and the difference between the incident intensity and the output intensity is calculated to obtain the theoretical attenuation intensity corresponding to the target layer. This value accurately characterizes the amount of X-ray intensity absorbed by the copper target layer itself.

[0022] By determining the ideal position of each target layer on a blank digital canvas with the same dimensions as a real X-ray inspection image, based on the target design coordinates, and filling the region centered on this ideal position and conforming to the target shape parameters with the calculated theoretical attenuation intensity as a constant grayscale value, a single-layer theoretical projection image of each aligned target layer is generated. This output image is a grayscale image containing only the ideal aligned target information for that layer. Following the actual stacking order of the PCB, from top to bottom, all single-layer theoretical projection images are pixel-level superimposed in grayscale, with the superposition rule following the imaging principle of X-rays. The grayscale value of the final imaging point is determined by the combined intensity after penetrating all layers. The goal is to synthesize a multi-layer theoretical projection image of the target that contains only the ideal alignment projections of all layers and has the same dimension as the X-ray detection image. This invention obtains a multi-layer theoretical projection image of the target and uses it as a precise reference benchmark for comparison and analysis with the measured X-ray image. This transforms the process of identifying interlayer alignment deviations from the traditional method of blindly estimating in mixed signals to a quantifiable and precise comparison guided by a clear theoretical benchmark. This fundamentally overcomes the problems of low identification accuracy and poor anti-interference ability caused by the lack of a single-layer benchmark in traditional methods.

[0023] In one embodiment, step S3, which involves obtaining the frequency domain residual spectrum based on the X-ray detection image and the multi-layer target theoretical projection image, and performing frequency domain feature analysis on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix, includes: S31. Perform pixel-level difference operation on the X-ray detection image and the multi-layer target theoretical projection image to generate an initial residual image, and perform two-dimensional fast Fourier transform on the initial residual image to obtain the frequency domain residual spectrum. S32. Perform radial energy distribution analysis on the frequency domain residual spectrum to obtain a radial energy distribution vector, wherein the radial energy distribution vector includes multiple frequency bands and their corresponding frequency band energy values, and based on the frequency band energy values, screen the multiple frequency bands by energy peak detection to obtain the main coupling frequency band; S33. Generate a frequency domain filtering matrix based on the main coupling frequency band, and obtain the filtered enhanced frequency domain spectrum according to the frequency domain filtering matrix and the frequency domain residual spectrum; S34. Perform inverse Fourier transform on the filtered enhanced frequency domain spectrum to obtain the enhanced residual image, and use the sliding window statistical method to perform local variance analysis on the enhanced residual image to obtain the local variance map. S35. The local variance map is mapped with an inverse Gaussian function to obtain a pixel-level confidence weight matrix.

[0024] As described in steps S31-S35 above, this invention generates an initial residual image by performing pixel-level difference operations on the X-ray detection image and the multi-layer target theoretical projection image. This image quantifies the degree of difference between the real and ideal imaging at all spatial locations; the higher the pixel value, the greater the deviation between reality and ideal at that location. Traditional methods cannot distinguish between high-confidence reliable regions and low-confidence noise / distortion regions, making it impossible to focus on effective information when identifying deviations. This invention transforms the initial residual image from the spatial domain defined by pixel coordinates to the frequency domain defined by frequency and direction by performing a two-dimensional fast Fourier transform on the initial residual image. In the frequency domain, the image information is decomposed into a superposition of a series of sine waves (i.e., frequency components) with different frequencies and directions, thereby obtaining the frequency domain residual spectrum. The frequency domain residual spectrum refers to a... A complex matrix, where each element corresponds to a specific frequency component, contains the amplitude and phase information of that component. Its amplitude characterizes the intensity of that frequency component in the original image. In the image frequency domain, low-frequency components correspond to large areas of slowly changing background and contour information in the image, while high-frequency components correspond to differences in edges, noise, and fine structures. By dividing the two-dimensional frequency domain into multiple continuous concentric rings centered at the zero frequency point of the frequency domain coordinate system, multiple radial frequency bands are obtained. The sum of the squares of the amplitudes of all frequency components within each band is calculated, which is used as the energy within the center frequency range represented by that band, thus obtaining a radial energy distribution vector. Each element of this vector corresponds to the residual energy of a band, and its index corresponds to the center frequency of that band. This vector quantifies the distribution of residual energy at different center frequency scales.

[0025] Due to the edge ghosting caused by minute misalignments in the multi-layered circular target, an approximately periodic interference pattern is formed in space. This pattern concentrates in a specific mid-to-high frequency band in the frequency domain, manifesting as energy peaks. Furthermore, due to the characteristic properties of the target edge details, these energy peaks are precisely concentrated in the mid-to-high frequency range. Based on this physical imaging principle, this invention analyzes the radial energy distribution vector to select the frequency band corresponding to the energy peak. This frequency band is identified as the main coupling frequency band. If multiple frequency bands correspond to energy peaks, the band with the higher center frequency is identified as the main coupling frequency band. Because frequency bands are discretely divided, the probability that the center frequencies of two different frequency bands are strictly equal is extremely low. Based on the center frequency of the main coupling frequency band and... A frequency domain filtering matrix is ​​generated, which is a matrix with the same size as the frequency domain residual spectrum. Its value is close to 1 in the target frequency band and close to 0 in other frequency bands. The frequency domain filtering matrix and the frequency domain residual spectrum are multiplied in the frequency domain to obtain the filtered enhanced frequency domain spectrum. This process allows frequency components within the main coupling frequency band to pass without loss, while significantly attenuating frequency components outside the band. By performing an inverse Fourier transform on the filtered enhanced frequency domain spectrum, it is reconstructed from the frequency domain back to the spatial domain to obtain the enhanced residual image. This invention selectively enhances the characteristic signal generated by the core physical phenomenon of interlayer target misalignment through frequency domain filtering, while actively suppressing low-frequency background fluctuations and high-frequency random noise unrelated to misalignment, thereby obtaining a difference image with a significantly improved signal-to-noise ratio.

[0026] The enhanced residual image contains two main components: first, the structural difference signal generated by interlayer misalignment, which manifests as a regular edge ghosting pattern; and second, residual random noise and irrelevant interference. This invention uses a sliding window to calculate the gray-level variance in the neighborhood of each pixel in the enhanced residual image, obtaining a local variance map. In the local variance analysis, the structural difference signal generated by interlayer misalignment exhibits a relatively uniform and continuous spatial distribution, with a relatively small local variance. In contrast, residual random noise, image artifacts, and other interference factors exhibit irregular and drastic fluctuations, leading to a significant increase in local variance. Therefore, the smaller the local variance, the more likely the signal in that region is a reliable feature generated by the target interlayer misalignment; conversely, the larger the local variance, the more severe the noise interference in that region, and the less reliable the information. Furthermore, the local variance value of each pixel is input into an inverse Gaussian function for mapping; that is, the larger the variance, the closer the output weight value is to 0. The smaller the difference, the closer the output weight value is to 1, thus obtaining a pixel-level confidence weight matrix. This pixel-level confidence weight matrix is ​​a two-dimensional matrix with the same dimensions as the original X-ray detection image. Each element is a floating-point number between 0 and 1, quantitatively representing the reliability of the information provided by the multi-layer target theoretical projection image at the corresponding pixel location. This matrix has the same dimensions as the original image, and each pixel value represents the reliability of the corresponding location in subsequent image separation processing. This invention combines frequency domain feature extraction with spatial domain uncertainty assessment to obtain a pixel-level confidence weight matrix. This matrix provides a basis for data confidence decisions in the subsequent image separation process, enabling the separation algorithm to adaptively adjust the confidence level of data in different regions based on this matrix. This helps solve the problem of low separation accuracy caused by the inability of traditional methods to distinguish between useful information and interference noise in the residual signal.

[0027] In one embodiment, step S4, which involves fusing and deconvolving the X-ray detection image and the multi-layer target theoretical projection image based on the pixel-level confidence weight matrix and the layered structure design parameters to obtain a deconvolution result image, and then obtaining single-layer target separation images of each target layer based on the target design parameters and the deconvolution result image, includes: S41. Based on the pixel-level confidence weight matrix, the X-ray detection image and the multi-layer target theoretical projection image are fused to generate a fused guidance image, and the point diffusion characteristics of the multi-path propagation and attenuation of the X-rays are simulated based on the layered structure design parameters to generate a point diffusion matrix. S42. Obtain initial deconvolution iteration parameters based on the fused guide image, X-ray detection image and the point spread matrix, and perform deconvolution iteration optimization on the fused guide image based on the initial deconvolution iteration parameters and the pixel-level confidence weight matrix to obtain multiple intermediate iteration images; S43. Obtain adjacent iteration difference sequences based on multiple intermediate images of the iteration, and determine the iteration termination based on the difference sequences to obtain the deconvolution result image; S44. Based on the target design parameters and the deconvolution result image, target region extraction is performed to obtain single-layer target separation images of each target layer.

[0028] As described in steps S41-S44 above, this invention performs pixel-level weighted fusion of the X-ray detection image and the multi-layer target theoretical projection image based on a pixel-level confidence weight matrix: The weight value corresponding to each pixel position is obtained from the pixel-level confidence weight matrix; the actual gray value corresponding to each pixel position is obtained from the X-ray detection image; and the theoretical gray value corresponding to each pixel position is obtained from the multi-layer target theoretical projection image. This is achieved through "I..." f(i, j) =w (i, j) *I t(i, j) +[1-w (i, j) ]*I m(i, j) "Calculate the final grayscale value, where I..." f(i, j) w represents the final grayscale value corresponding to the pixel with coordinates (i, j). (i, j) I represents the weight value corresponding to the pixel with coordinates (i, j). t(i, j) I represents the theoretical gray value corresponding to the pixel with coordinates (i, j). m(i, j) The actual gray value corresponding to the pixel with coordinates (i, j) is represented by (i, j), where (i, j) represents the position coordinates of the pixel. Based on the final gray value, a fused guiding image that completely corresponds to the size and spatial coordinates of the X-ray detection image is generated. In areas with high confidence, this image mainly presents the theoretically predicted target structure, while in areas with low confidence, it retains the original information of the measured image, thus forming an initial image driven by confidence that combines the advantages of theoretical prior and real observation.

[0029] Point diffusion simulation was performed using a fast convolution-based simulation method based on PCB layered structure design parameters. First, a two-dimensional digital grid parallel to the detector plane was constructed in the computer. The grid's precision was much higher than the detector's pixel size. Then, based on the PCB layered structure design parameters, a three-dimensional linear attenuation coefficient distribution model was constructed on this grid along the X-ray penetration direction (defined as the x-axis). The model construction rules were as follows: The spatial position of each material layer on the x-axis was determined according to the stacking order and thickness in the layered structure design parameters. When a point in the model was located in the k-th dielectric layer, its linear attenuation coefficient was assigned the design attenuation coefficient of that dielectric layer. For example, when located in the k-th copper foil layer, it was assigned the design attenuation coefficient of the copper foil. Subsequently, a line integral was performed along the x-axis on this three-dimensional distribution. "Calculate the equivalent attenuation distribution, where A(p, q) represents the equivalent attenuation distribution experienced by the X-ray signal received at the plane coordinates (p, q), a represents the coordinates of the incident surface of the PCB board in the X-ray penetration direction, b represents the coordinates of the exit surface of the PCB board in the X-ray penetration direction, f(x) represents the integrand, the output value of which is the design attenuation coefficient, and x represents the integration variable, which represents the position coordinates in the X-ray penetration direction. It is a function that is in the interval [a, q]." b] A continuously changing scalar value; finally, based on the physical model that small-angle forward scattering of X-rays in matter is the main factor causing imaging blur, a preset point spread function kernel, used to characterize the energy distribution caused by scattering, is convolved with the equivalent attenuation distribution to obtain the point spread matrix. This matrix accurately describes how a PCB stack-up structure will blur an ideal sharp point into a two-dimensional distribution during imaging. The above processing method replaces ray tracing with a parameterized physical model, which significantly reduces computational complexity while ensuring accuracy. Traditional methods usually use general, hypothetical blur kernels, such as Gaussian blur, which cannot accurately reflect the unique blur caused by a specific PCB stack-up structure. However, this invention obtains this accurate blur model, which enables the subsequent image separation to accurately cancel the imaging blur introduced by the specific PCB stack-up structure.

[0030] Traditional image separation methods typically involve direct deconvolution of the image. A fundamental flaw is that this approach assumes the entire image possesses uniform statistical properties. However, due to noise, deformation, and other non-uniform interference present in actual imaging, direct deconvolution amplifies noise significantly in low signal-to-noise ratio regions and fails to fully utilize prior information in high-reliability regions, leading to distorted separation results. Therefore, this invention constructs an image reconstruction model based on the Richardson-Lucy deconvolution algorithm. This model uses a fused guide image as the initial image estimate for iterative calculations, an X-ray detection image as the observation target to be approximated, and a point spread matrix as the convolution kernel describing the system's ambiguity characteristics. An objective function is established to minimize the difference between the estimated image after convolution with the point spread matrix and the observed image, thus completing the initial configuration of the deconvolution solution system and obtaining the initial deconvolution iteration parameters. In each iteration, an improved Richardson-Lucy deconvolution iteration calculation is performed. The core improvement lies in integrating the pixel-level confidence weight matrix into the iterative update mechanism. The specific implementation is as follows: In each iteration, firstly, according to the standard Rich... The Ardson-Lucy algorithm calculates the initial update value for the entire image. Then, this initial update value is multiplied element-wise with the pixel-level confidence weight matrix to achieve spatially adaptive modulation: in confident regions with high weight values, the update value is fully preserved to ensure convergence speed; in unconfident regions with low weight values, the update value is significantly suppressed to avoid noise amplification. Finally, the modulated update value is added to the image estimate of the current iteration to obtain a new round of intermediate images. Through this confidence-modulated iterative update mechanism, the image estimate is optimized in a targeted manner. This invention uses the pixel-level confidence weight matrix as a regularization constraint to dynamically control the iteration process, thereby achieving intelligent and differentiated processing of different image regions. This iterative process is repeated. The system calculates the difference norm of the pixel gray values ​​between the intermediate images obtained from two adjacent iterations to obtain the adjacent iteration difference sequence, and compares the adjacent iteration differences with a preset convergence threshold in real time. When the difference value is continuously less than the convergence threshold, it is determined that the image reconstruction result has reached a stable state. At the same time, a maximum number of iterations is set as a termination guarantee. When the number of iterations reaches the preset upper limit, the calculation is forcibly terminated.The resulting image estimate is the deconvolution result image, a panoramic image with a significantly improved signal-to-noise ratio after multiple rounds of confidence-guided optimization. Its reconstruction process is strongly constrained by the PCB multilayer structure physical model and the target design coordinates. This process is equivalent to spatially separating and rearranging the target information, originally superimposed in the projection, on a two-dimensional plane according to their physical layer attributes. Therefore, the deconvolution result image is an image where each layer of targets is precisely arranged in space according to its design coordinates and does not overlap. Using each center point as a reference, square regions of interest of fixed size are extracted according to the target shape parameters corresponding to each layer, resulting in single-layer target separation images for each target layer. This achieves the transformation from mixed projection to a single-layer clear target image.

[0031] In one embodiment, step S5, which involves fitting an elliptical model to the single-layer target separation image of each target layer to obtain the corresponding optimal elliptical model, and performing deformation analysis based on the optimal elliptical model to obtain the actual center coordinates of each target layer, includes: S51. The Zernike moment orthogonal polynomial algorithm is used to perform sub-pixel edge localization on the single-layer target separation image of each target layer to obtain the sub-pixel precision edge point set of each target layer, and multiple ellipse equations are constructed based on the sub-pixel precision edge point set. S52. Based on the sub-pixel precision edge point set, perform interior point statistical verification on each ellipse equation to obtain the number of interior points corresponding to each ellipse equation, and filter multiple ellipse equations based on the number of interior points to obtain the optimal ellipse model corresponding to each target layer. S53. Obtain the initial center point coordinates and ellipse semi-axis parameters of each target layer according to the optimal ellipse model, and obtain the target design shape and design aspect ratio of each target layer according to the target design parameters. S54. Obtain the shape deviation based on the target design shape, the design aspect ratio, and the elliptical semi-axis parameter; S55. Obtain the deformation tolerance and determine whether the shape deviation is less than the deformation tolerance; If the shape deviation is less than the deformation tolerance, then the initial center point coordinates are used as the actual center coordinates of the target in that layer. If the shape deviation is not less than the deformation tolerance, then a smooth fitting based on the B-spline curve is performed based on the sub-pixel precision edge point set to obtain the geometric centroid coordinates of the polygon, and the geometric centroid coordinates of the polygon are used as the actual center coordinates of the target in this layer.

[0032] As described in steps S51-S55 above, this invention uses the Zernike moment orthogonal polynomial algorithm to calculate the complex integral values ​​of multiple Zernike moments in the neighborhood of each pixel in the image for each single-layer target separation image. Based on the phase relationship of these moments, the precise sub-pixel position and direction of the edge crossing the neighborhood of the pixel are accurately determined, resulting in a sub-pixel precision edge point set. The sub-pixel precision edge point set refers to a set of ordered two-dimensional coordinate points in the image plane coordinate system, where the horizontal and vertical coordinates of each coordinate point are sub-pixel precision values ​​in floating-point form. This point set accurately describes the continuous geometric path of the target contour with a resolution higher than that of the inherent pixel grid of the image.

[0033] Traditional methods for target center localization typically employ pixel-level edge detection combined with least-squares fitting. However, pixel-level edge localization inherently suffers from quantization errors, limiting the accuracy of contour information. Furthermore, traditional fitting methods assign equal weight to all edge points, causing noise points and deformed regions to significantly impact the accuracy of center coordinate calculations, resulting in a systematic deviation of the localization result from the true geometric center. This invention, however, randomly selects the coordinates of five points from a sub-pixel precision edge point set. Using these five coordinates, the coefficients of an ellipse equation are calculated using the direct least-squares method, thus constructing an ellipse equation. All coordinates from the sub-pixel precision edge point set are then substituted into this ellipse equation for verification. The geometric distance from each point to the hypothetical ellipse is calculated. If the geometric distance is less than a preset error threshold, the point is considered an interior point conforming to the current ellipse model. This process of random sampling, ellipse equation construction, and model verification is repeated, and the ellipse equation with the most interior points is selected as the optimal ellipse model.

[0034] The initial center point coordinates and semi-axis parameters of the ellipse are obtained based on the optimal ellipse model. The center point coordinates refer to the preliminary estimate of the target center position based on the ellipse model assumptions. The semi-axis parameters include the semi-major axis and semi-minor axis. The target design shape and aspect ratio of each layer are obtained based on the target design parameters. If the design shape is circular, then according to… "Calculate the shape deviation, where C represents the shape deviation, a represents the semi-major axis, and b represents the semi-minor axis. For a perfect circle, its semi-major axis equals its semi-minor axis, therefore..." "When the value equals 1, the shape deviation is 0. The flatter the ellipse, the greater the difference between the semi-major axis and the semi-minor axis." The smaller the value of "", the closer the shape deviation is to 1. If the design shape is elliptical, the actual aspect ratio of the optimal elliptical model is calculated based on the semi-major axis and semi-minor axis, and the absolute value of the difference between the designed aspect ratio and the actual aspect ratio is calculated to obtain the shape deviation. The shape deviation characterizes the degree of geometric difference between the actual imaged target shape and the ideal circle. A preset deformation tolerance is obtained. The value of the deformation tolerance is determined based on the material rheological properties of the PCB board, the tolerance of the lamination process, and the detection accuracy requirements. It is determined by statistical analysis of known qualified samples. This invention compares the shape deviation with the deformation tolerance. If the shape deviation is less than the deformation tolerance, the target deformation is considered insignificant. In this case, the initial center point coordinates are directly used as the actual center coordinates of the target in this layer. If the shape deviation is greater than or equal to the deformation tolerance, the deformation is considered non-negligible, indicating that the elliptical model is no longer applicable. In this case, a smooth fitting based on a B-spline curve is performed on the coordinates of all sub-pixel precision edge point sets. First, the pixel coordinates of all pixels in the sub-pixel precision edge point set are sorted according to the contour order to ensure the continuity of the point set. Then, the optimal order and node vector of the B-spline curve are determined based on the cross-validation method. The sorted edge point set is used as the control vertices and input into the B-spline curve model. The curve parameters are solved by least squares optimization to minimize the average distance between the fitted curve and the edge point set. Based on the fitted continuous B-spline curve, sampling is performed at equal intervals according to a preset step size to obtain a series of ordered sampling points. These sampling points are connected in sequence to reconstruct a continuous and closed contour polygon, and the geometric centroid of this polygon is calculated. This geometric centroid is used as the actual center coordinate of the target in this layer. This invention introduces shape deviation as a deformation evaluation index and adaptively selects the most suitable center calculation algorithm accordingly. This fundamentally solves the problem of insufficient target center positioning accuracy in complex real-world scenarios caused by the lack of deformation perception capability in traditional methods, which is beneficial to improving the accuracy and reliability of target center recognition.

[0035] In one embodiment, step S6, which involves obtaining the fusion weights corresponding to each adjacent layer pair based on the layered structure design parameters, and performing inter-layer deviation fusion based on the fusion weights, target design parameters, and the actual center coordinates to obtain the comprehensive alignment deviation index, includes: S61. Obtain the characteristic impedance sensitivity coefficient of each adjacent layer pair according to the layered structure design parameters, and obtain the fusion weight corresponding to each adjacent layer pair according to the multiple characteristic impedance sensitivity coefficients. S62. Obtain the design center coordinates of each target layer according to the target design parameters, and obtain the translation deviation amount according to the actual center coordinates and the design center coordinates of each target layer, wherein the translation deviation amount includes the X-axis deviation amount and the Y-axis deviation amount. S63. Construct actual connection vectors and theoretical connection vectors based on a pair of actual center coordinates and a pair of design center coordinates corresponding to each pair of adjacent layers, and obtain the rotation deviation based on the actual connection vectors and the theoretical connection vectors. S64. Obtain the single-layer pair comprehensive deviation of each pair of adjacent layers based on the X-axis deviation, Y-axis deviation and rotation deviation, and fuse multiple single-layer pair comprehensive deviations based on the fusion weight to obtain the comprehensive alignment deviation index.

[0036] As described in steps S61-S64 above, this invention obtains the dielectric design thickness, dielectric constant, transmission line width, line spacing, and corresponding target characteristic impedance value of each interlayer dielectric based on the layered structure design parameters. The dielectric design thickness refers to the nominal thickness of the insulating dielectric between two adjacent conductive layers specified in the PCB design document. The target characteristic impedance value refers to the ideal impedance value set in the PCB design for a specific transmission line, which is required to be achieved in actual manufacturing. Based on three-dimensional electromagnetic simulation software, such as HFSS and CST, a transmission line simulation model for each adjacent layer pair is built according to the layered structure design parameters. The model includes key structures such as conductive layers, interlayer dielectrics, and vias, ensuring consistency with the actual PCB design. Then, in the simulation model, a series of multi-dimensional alignment deviation combinations are preset according to the possible deviation range in actual production. Each alignment deviation combination includes X-axis deviation, Y-axis deviation, and rotational deviation to cover the coupling effects of translation and rotation. Twenty groups are selected using a uniform gradient. For each deviation combination, electromagnetic simulation is run to obtain the corresponding simulated characteristic impedance value. The difference between the simulated characteristic impedance value and the target characteristic impedance value is calculated, and the ratio of this difference to the target characteristic impedance value is calculated to obtain the impedance change rate, thereby quantifying the degree of impedance deviation from the target value. A comprehensive deviation factor is obtained by normalizing the X-axis deviation, Y-axis deviation, and rotational deviation, and calculating using weighted Euclidean distance. The weights are set according to the transmission line type: for microstrip lines, the deviation weights corresponding to X-axis and Y-axis deviations are both 0.4, and the deviation weight corresponding to rotational deviation is 0.2; for striplines, the deviation weights corresponding to X-axis and Y-axis deviations are both 0.3, and the deviation weight corresponding to rotational deviation is 0.4; and using the comprehensive deviation factor as the independent variable and the impedance change rate as the dependent variable, a nonlinear fitting model, such as Gaussian kernel function fitting or cubic polynomial fitting, is used to fit the data to obtain a nonlinear mapping model of the deviation and impedance change of the adjacent layer pair; and the slope of the model at the maximum allowable comprehensive deviation is obtained to obtain the characteristic impedance sensitivity coefficient of the adjacent layer pair. The characteristic impedance sensitivity coefficient is used to characterize the degree of influence of the alignment accuracy of the adjacent layer on maintaining the overall characteristic impedance stability of the PCB board. By normalizing the characteristic impedance sensitivity coefficient corresponding to each adjacent layer pair, the fusion weight of the deviation of each adjacent layer pair is obtained; when evaluating the overall alignment quality of multilayer boards, traditional methods usually only focus on the maximum single-layer deviation or perform simple deviation averaging. Its fundamental defect is that it assumes that all interlayer deviations While the impact of deviations on the final PCB electrical performance is uniform, in actual high-speed circuit design, high-speed signal lines on adjacent layers are connected vias. If misalignment occurs between these layers, it can lead to minute changes in signal line width, dielectric thickness, etc., resulting in variations in the characteristic impedance of the signal line. Different interlayer structures will cause different impedance changes for the same amount of misalignment. Furthermore, differences in dielectric thickness and dielectric constant between different layers make some interlayer structures more sensitive to changes in characteristic impedance. Even minor interlayer misalignments in these sensitive areas can cause significant signal integrity degradation. This invention transforms the PCB electrical performance design objective into a quantifiable weighting factor to guide manufacturing quality assessment, providing a crucial electrical performance perspective for subsequent deviation fusion.

[0037] This invention obtains the design center coordinates of the targets for each layer of a PCB board through target design parameters. For each pair of adjacent layers in a multilayer PCB board, the translational deviation between the actual center coordinates and the design center coordinates of the corresponding targets in the X-axis and Y-axis directions is calculated. The translational deviation includes the X-axis deviation and the Y-axis deviation, and this data quantifies the relative misalignment between adjacent layers. For each pair of adjacent layers in a multilayer PCB board, an actual connection vector is constructed based on the actual center coordinates of the two targets in each pair of adjacent layers, and a theoretical connection vector is constructed based on the design center coordinates of the two targets in each pair of adjacent layers. Furthermore, based on… Calculate the rotational deviation, where θ represents the rotational deviation, and V a V represents the actual connection vector. t Representing the theoretical connection vector, the rotational deviation is expressed in radians, quantifying the relative rotational misalignment between adjacent layers; through "M" i =sqrt[(Δx i ) 2 +(Δy i ) 2 +α*θ i 2]"Calculate the overall deviation of a single layer, where M i The single-layer pair composite deviation represents the i-th adjacent layer pair, sqrt represents the mathematical operation symbol for the square root, and Δx i Δy represents the X-axis deviation of the i-th adjacent layer pair. i Let θ represent the Y-axis deviation of the i-th adjacent layer pair, α represent the rotation-translation conversion factor used to convert the radian value into an equivalent value with length dimensions, and θ represent the rotation-translation conversion factor used to convert the radian value into an equivalent value with length dimensions. i This represents the rotational deviation of the i-th adjacent layer pair; through the above calculation, the deviations of each adjacent layer pair in different degrees of freedom are merged into a total, comparable deviation; through " "Calculate the overall alignment deviation index, where D represents the overall alignment deviation index, sqrt represents the square root mathematical operation symbol, n represents the number of adjacent layer pairs on the PCB board, i represents the sequence number of the adjacent layer pair, and ω..." i M represents the fusion weight of the i-th adjacent layer pair. i This represents the comprehensive deviation of the single-layer pair of the i-th adjacent layer pair. This calculation method amplifies the contribution of major deviations and critical layer deviations to the final deviation result, thus providing a more comprehensive reflection of the misalignment of the entire laminate alignment.

[0038] This application also provides a system for identifying alignment deviations in multilayer PCB boards, including: The data acquisition module is used to acquire X-ray inspection images of the PCB board after multilayer board lamination, and to acquire the layered structure design parameters and target design parameters of the PCB board. The projection synthesis module is used to perform multi-layer target projection synthesis based on the layered structure design parameters and the target design parameters to obtain a theoretical projection image of the multi-layer target. The image analysis module is used to obtain the frequency domain residual spectrum based on the X-ray detection image and the multi-layer target theoretical projection image, and to perform frequency domain feature analysis on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix. The image separation module is used to perform fusion and deconvolution operations on the X-ray detection image and the multi-layer target theoretical projection image based on the pixel-level confidence weight matrix and the layered structure design parameters to obtain the deconvolution result image, and to obtain the single-layer target separation image of each target layer according to the target design parameters and the deconvolution result image. The target localization module is used to fit the single-layer target separation image of each target layer with an elliptical model to obtain the corresponding optimal elliptical model, and to perform deformation analysis based on the optimal elliptical model to obtain the actual center coordinates of each target layer. The deviation fusion module is used to obtain the fusion weight corresponding to each adjacent layer pair according to the layered structure design parameters, and to perform inter-layer deviation fusion based on the fusion weight, target design parameters and actual center coordinates to obtain a comprehensive alignment deviation index. The deviation determination module is used to determine whether the comprehensive alignment deviation index is greater than the deviation threshold. If the comprehensive alignment deviation index is greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be substandard. If the comprehensive alignment deviation index is not greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be up to standard.

[0039] In one embodiment, the image separation module includes: The diffusion simulation unit is used to fuse the X-ray detection image and the multi-layer target theoretical projection image based on the pixel-level confidence weight matrix to generate a fused guidance image, and to simulate the point diffusion characteristics of the multipath propagation and attenuation of the X-rays based on the layered structure design parameters to generate a point diffusion matrix. The iterative optimization unit is used to obtain initial deconvolution iteration parameters based on the fused guide image, the X-ray detection image and the point spread matrix, and to perform deconvolution iterative optimization on the fused guide image based on the initial deconvolution iteration parameters and the pixel-level confidence weight matrix to obtain multiple iterative intermediate images; An image generation unit is used to obtain an adjacent iteration difference sequence based on multiple intermediate images of the iteration, and to determine the iteration termination based on the difference sequence to obtain a deconvolution result image; The image separation unit is used to extract the target region based on the target design parameters and the deconvolution result image to obtain single-layer target separation images of each target layer.

[0040] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for identifying alignment deviations in multilayer PCB boards described above.

[0041] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for identifying alignment deviations in multilayer PCB boards described above.

[0042] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media provided in this application and in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0043] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0044] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for identifying alignment deviations in multilayer PCB boards, characterized in that, include: Obtain X-ray inspection images of the PCB board after multilayer board lamination, and obtain the layered structure design parameters and target design parameters of the PCB board; Based on the layered structure design parameters and the target design parameters, multi-layer target projection synthesis is performed to obtain a theoretical projection image of the multi-layer target. The frequency domain residual spectrum is obtained based on the X-ray detection image and the multi-layer target theoretical projection image, and frequency domain feature analysis is performed on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix. Based on the pixel-level confidence weight matrix and the layered structure design parameters, the X-ray detection image and the multi-layer target theoretical projection image are fused and deconvolved to obtain the deconvolution result image. Then, based on the target design parameters and the deconvolution result image, the single-layer target separation image of each target layer is obtained. Elliptical model fitting is performed on the single-layer target separation image of each target layer to obtain the corresponding optimal elliptical model, and deformation analysis is performed based on the optimal elliptical model to obtain the actual center coordinates of each target layer. The fusion weights corresponding to each adjacent layer pair are obtained according to the layered structure design parameters, and inter-layer deviation fusion is performed based on the fusion weights, target design parameters and actual center coordinates to obtain the comprehensive alignment deviation index. Determine whether the comprehensive alignment deviation index is greater than the deviation threshold; If the comprehensive alignment deviation index is greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be substandard. If the comprehensive alignment deviation index is not greater than the deviation threshold, the alignment accuracy of the PCB multilayer board is determined to be up to standard.

2. The method for identifying alignment deviations in multilayer PCB boards according to claim 1, characterized in that, The step of synthesizing multi-layer target projections based on the layered structure design parameters and the target design parameters to obtain a theoretical projection image of the multi-layer target includes: The layered structure design thickness and attenuation coefficient are obtained based on the layered structure design parameters. Then, based on the Lambert-Beer law, the layered penetration intensity is simulated layer by layer according to the layered structure design thickness and attenuation coefficient of each layer of the PCB board to obtain the theoretical irradiation intensity of X-rays reaching the target position of each layer. The target design thickness, target attenuation coefficient, target shape parameters and target design coordinates are obtained based on the target design parameters. The target absorption intensity is simulated based on the Lambert-Beer law, the target design thickness, the target attenuation coefficient and the theoretical irradiation intensity, and the X-ray emission intensity after penetrating the target is obtained. The theoretical attenuation intensity of each target layer is obtained based on the theoretical irradiation intensity and the emitted intensity. An ideal target image is constructed based on the theoretical attenuation intensity, target shape parameters and target design coordinates to obtain multiple single-layer theoretical projection images. Pixel-level grayscale superposition of multiple single-layer theoretical projection images is performed to obtain multi-layer target theoretical projection images.

3. The method for identifying alignment deviations in multilayer PCB boards according to claim 1, characterized in that, The step of obtaining the frequency domain residual spectrum based on the X-ray detection image and the multi-layer target theoretical projection image, and performing frequency domain feature analysis on the frequency domain residual spectrum to obtain a pixel-level confidence weight matrix includes: Pixel-level difference operations are performed on the X-ray detection image and the theoretical projection image of the multi-layer target to generate an initial residual image, and a two-dimensional fast Fourier transform is performed on the initial residual image to obtain the frequency domain residual spectrum. Radial energy distribution analysis is performed on the frequency domain residual spectrum to obtain a radial energy distribution vector, wherein the radial energy distribution vector includes multiple frequency bands and their corresponding frequency band energy values, and the multiple frequency bands are screened by energy peak detection based on the frequency band energy values ​​to obtain the main coupling frequency band; A frequency domain filtering matrix is ​​generated based on the main coupling frequency band, and the filtered enhanced frequency domain spectrum is obtained based on the frequency domain filtering matrix and the frequency domain residual spectrum. An inverse Fourier transform is performed on the filtered enhanced frequency domain spectrum to obtain an enhanced residual image, and a local variance analysis is performed on the enhanced residual image using a sliding window statistical method to obtain a local variance map. The local variance map is mapped with an inverse Gaussian function to obtain a pixel-level confidence weight matrix.

4. The method for identifying alignment deviations in multilayer PCB boards according to claim 1, characterized in that, The steps of fusing and deconvolutionally performing the X-ray detection image and the multi-layer target theoretical projection image based on the pixel-level confidence weight matrix and the layered structure design parameters to obtain the deconvolution result image, and obtaining the single-layer target separation image of each target layer according to the target design parameters and the deconvolution result image, include: The X-ray detection image and the multi-layer target theoretical projection image are fused based on the pixel-level confidence weight matrix to generate a fused guidance image. The point diffusion characteristics of the multi-path propagation and attenuation of the X-rays are simulated based on the layered structure design parameters to generate a point diffusion matrix. The initial deconvolution iteration parameters are obtained based on the fused guide image, the X-ray detection image and the point spread matrix. The fused guide image is then subjected to deconvolution iteration optimization based on the initial deconvolution iteration parameters and the pixel-level confidence weight matrix to obtain multiple intermediate iteration images. The adjacent iteration difference sequence is obtained based on multiple intermediate images of the iteration, and the iteration termination is determined based on the difference sequence to obtain the deconvolution result image; Based on the target design parameters and the deconvolution result image, target regions are extracted to obtain single-layer target separation images of each target layer.

5. The method for identifying alignment deviations in multilayer PCB boards according to claim 1, characterized in that, The steps of fitting an elliptical model to the single-layer target separation image of each target layer to obtain the corresponding optimal elliptical model, and performing deformation analysis based on the optimal elliptical model to obtain the actual center coordinates of each target layer, include: The Zernike moment orthogonal polynomial algorithm is used to perform sub-pixel edge localization on the single-layer target separation image of each target layer to obtain the sub-pixel precision edge point set of each target layer, and multiple ellipse equations are constructed based on the sub-pixel precision edge point set. Based on the sub-pixel precision edge point set, the interior point statistical verification of each ellipse equation is performed to obtain the number of interior points corresponding to each ellipse equation. Based on the number of interior points, multiple ellipse equations are filtered to obtain the optimal ellipse model corresponding to each target layer. The initial center point coordinates and ellipse semi-axis parameters of each target layer are obtained according to the optimal ellipse model, and the target design shape and aspect ratio of each target layer are obtained according to the target design parameters. The shape deviation is obtained based on the target design shape, the design aspect ratio, and the ellipse semi-axis parameter. Obtain the deformation tolerance and determine whether the shape deviation is less than the deformation tolerance; If the shape deviation is less than the deformation tolerance, then the initial center point coordinates are used as the actual center coordinates of the target in that layer. If the shape deviation is not less than the deformation tolerance, then a smooth fitting based on the B-spline curve is performed based on the sub-pixel precision edge point set to obtain the geometric centroid coordinates of the polygon, and the geometric centroid coordinates of the polygon are used as the actual center coordinates of the target in this layer.

6. The method for identifying alignment deviations in multilayer PCB boards according to claim 1, characterized in that, The step of obtaining the fusion weight corresponding to each adjacent layer pair according to the layered structure design parameters, and performing inter-layer deviation fusion based on the fusion weight, target design parameters, and actual center coordinates to obtain a comprehensive alignment deviation index includes: The characteristic impedance sensitivity coefficient of each adjacent layer pair is obtained according to the layered structure design parameters, and the fusion weight corresponding to each adjacent layer pair is obtained according to the multiple characteristic impedance sensitivity coefficients. The design center coordinates of each target layer are obtained according to the target design parameters, and the translation deviation is obtained according to the actual center coordinates and the design center coordinates of each target layer, wherein the translation deviation includes the X-axis deviation and the Y-axis deviation. Based on a pair of actual center coordinates and a pair of design center coordinates corresponding to each pair of adjacent layers, construct actual connection vectors and theoretical connection vectors, and obtain rotation deviation based on the actual connection vectors and the theoretical connection vectors; The single-layer pair comprehensive deviation is obtained based on the X-axis deviation, Y-axis deviation and rotation deviation, and multiple single-layer pair comprehensive deviations are fused based on the fusion weight to obtain the comprehensive alignment deviation index.

7. A system for identifying alignment deviations in multilayer PCB boards, characterized in that, It includes multiple modules for implementing the steps of the method according to any one of claims 1 to 6.

8. The system for identifying alignment deviations in multilayer PCB boards according to claim 7, characterized in that, It includes multiple units, which are used to implement the steps of the method according to any one of claims 1 to 6.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images