HDI printed circuit board X-ray detection method and device

Abstract

The invention relates to the technical field of detection, in particular to an HDI printed circuit board X-ray detection method and device, and the method comprises the steps: obtaining a space reference system and initial position data of an HDI circuit board, and obtaining initial detection data; performing dynamic X-ray scanning and data acquisition on the HDI circuit board according to the initial detection data to obtain a motion control sequence; driving the X-ray source to execute dynamic multi-track motion and adaptive inclination angle adjustment according to the motion control sequence to obtain a synchronous control sequence; driving a detector to acquire an X-ray signal according to the synchronous control sequence, and generating a multi-dimensional projection data set; performing optimization processing on the multi-dimensional projection data set based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; and performing multi-layer structure feature extraction on the comprehensive detection image to obtain a deviation analysis report. The problem that penetration is difficult due to the fact that the thickness of the HDI circuit board in the length direction and the width direction is too large is effectively solved, and overlapping and signal attenuation of multi-layer structure features in a projection image are avoided.

Description

Technical Field

[0001] This invention relates to the field of testing technology, and more specifically, to a method and apparatus for X-ray inspection of HDI printed circuit boards. Background Technology

[0002] High-density interconnect (HDI) printed circuit boards play a crucial role in modern electronic device manufacturing due to their high integration, multilayer structure, and miniaturized design. Existing X-ray inspection methods typically employ static scanning at a fixed angle or traditional computed tomography (CT) scans, acquiring projected images and combining them with image processing algorithms to analyze the internal structure of the circuit board. However, because HDI circuit boards have a planar structure, their length and width dimensions are much larger than their height. Ordinary CT scans struggle to effectively penetrate the thickness in the length and width directions, making it difficult to generate clear projected images. This technical limitation makes it difficult for existing methods to accurately resolve the geometric features of complex multilayer structures when inspecting HDI circuit boards, resulting in insufficient inspection accuracy, especially when inspecting high-density microvias and fine linewidths.

[0003] To address the challenges of HDI circuit board inspection, existing technologies attempt to improve image resolution and feature extraction accuracy by increasing the fixed scanning angle of the X-ray source or adjusting the detector position to acquire multiple sets of projection data. For example, some methods attempt to compensate for the shortcomings of traditional CT scans by manually setting multi-angle scans or employing simple geometric correction algorithms. However, these methods still rely on preset fixed scanning paths and lack the ability to dynamically adjust the X-ray source's trajectory and angle, making it difficult to adapt to the complexity and diversity of the multi-layered structure of HDI circuit boards. Due to the special geometric characteristics of planar circuit boards, static or semi-static scanning methods can easily lead to overlap or signal attenuation of structural features at different levels in the projected image, thus affecting the reliability of subsequent feature extraction and deviation analysis. This technical limitation makes existing methods difficult to meet the requirements of high-precision inspection scenarios, especially in production environments where rapid inspection of large batches of HDI circuit boards is required.

[0004] Therefore, a new detection technology is urgently needed to solve the above-mentioned technical problems. Summary of the Invention

[0005] The main objective of this invention is to provide an X-ray inspection method and apparatus for HDI printed circuit boards, aiming to overcome the technical problem that existing technologies cannot accurately analyze the geometric features of complex multilayer HDI circuit boards.

[0006] To address the aforementioned problems, this invention proposes an X-ray inspection method for HDI printed circuit boards, the method comprising: Obtain the spatial reference frame and initial position data of the HDI circuit board to obtain initial detection data; Based on the initial detection data, dynamic X-ray scanning and data acquisition are performed on the HDI circuit board to obtain a motion control sequence; Based on the motion control sequence, the X-ray source is driven to perform dynamic multi-track motion and adaptive tilt angle adjustment to obtain a synchronous control sequence; The detector is driven to acquire X-ray signals according to the synchronous control sequence, and a multidimensional projection dataset is generated. The multidimensional projection dataset is optimized using a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; Multi-layer structural features are extracted from the comprehensive detection image to obtain a deviation analysis report.

[0007] Furthermore, the step of acquiring the spatial reference frame and initial position data of the HDI circuit board to obtain initial detection data includes: Laser triangulation scanning was performed on the HDI circuit board to obtain a set of three-dimensional geometric parameters; The material composition of the circuit board is obtained, and the material composition is partitioned and modeled according to the three-dimensional geometric parameter set to obtain the density distribution matrix. Based on the density distribution matrix, the initial motion trajectory of the HDI circuit board is planned to obtain the initial trajectory parameter set. The X-ray source and detector are aligned along their central axes based on the initial trajectory parameter set to obtain an alignment parameter set. The initial detection data is obtained by feedback calibration of the X-ray attenuation data based on the alignment parameter set.

[0008] Further, the step of performing dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain the motion control sequence includes: Based on the initial detection data, a spatial geometric transformation algorithm is used to analyze the spatial reference system and initial position data of the HDI circuit board to obtain the spatial distribution characteristic data of the HDI circuit board. Based on the spatial distribution characteristic data, a non-uniform ellipse-spiral composite trajectory planning algorithm is used to generate the preliminary trajectory parameters of the X-ray source. Based on the preliminary trajectory parameter set, the tilt angle of the X-ray source emission direction is calculated to obtain a dynamic tilt angle sequence; Based on the preliminary trajectory parameter set and dynamic tilt angle sequence, the detector motion is mirrored and synchronized to obtain the detector motion sequence. The X-ray attenuation data is collected and processed according to the detector motion sequence to obtain a dynamic attenuation dataset; The motion control sequence is obtained by dynamically optimizing the preliminary trajectory parameter set and tilt angle based on the dynamic decay dataset.

[0009] Further, the step of driving the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence to obtain the synchronization control sequence includes: The motion control sequence is decomposed into independent motion components along the X, Y, and Z axes. The motion sequence along each axis is analyzed in the frequency domain by Fourier transform to generate a multidimensional trajectory component set. Based on the set of multidimensional trajectory components, the motion parameters of the X-ray source are dynamically planned to obtain a set of trajectory optimization parameters. The energy distribution of the X-ray source is divided into regions based on the trajectory optimization parameter set to obtain an energy partitioning sequence. The emission parameters of the X-ray source are subjected to dual-mode modulation processing based on the energy partitioning sequence to obtain the modulation control sequence; The relative positions of the X-ray source and the detector are calibrated according to the modulation control sequence to obtain the synchronization control sequence.

[0010] Further, the step of driving the detector to acquire X-ray signals according to the synchronization control sequence and generating a multidimensional projection dataset includes: The detector's acquisition parameters are initialized and configured according to the synchronization control sequence to obtain an initial acquisition parameter set; The detector is driven to acquire X-ray signals based on the initial acquisition parameter set, resulting in a raw signal dataset containing intensity, phase contrast, and dark field signals. The detector's receiving position is calibrated by combining the tilt angle parameter in the synchronization control sequence to generate a calibrated signal dataset; Based on the intensity, phase contrast, and dark field signals in the calibrated signal dataset, a multidimensional projection reconstruction algorithm is used to integrate and spatially map the signals, generating a multidimensional projection dataset containing multidimensional spatial information.

[0011] Furthermore, the step of optimizing the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image includes: A multi-layer network model is constructed, and the tilt angle parameters in the synchronization control sequence are input into the multi-layer network model for deep learning weighting to obtain a weighted signal set. Identify high-density interconnected regions within the weighted signal set, and enhance the features of these high-density interconnected regions based on gradient operators to generate an enhanced feature set. Based on the feature intensity in the enhanced feature set, the mean and variance of gray level in each region are calculated, and the gray level distribution is optimized by dynamic range compression algorithm to obtain the optimized gray level set. The optimized grayscale set is input into a bilateral filter for initial smoothing and artifact features are extracted. The optimized grayscale set is then iteratively updated based on the artifact features to generate a de-artifacted signal set. Based on the signal distribution in the artifact removal signal set, the signal components are integrated using a multi-dimensional image reconstruction algorithm to generate the comprehensive detection image.

[0012] Further, the step of extracting multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report includes: A linear transformation is performed on the pixel grayscale values ​​of the comprehensive detection image to calculate the mean and standard deviation of the grayscale distribution of each pixel, and the grayscale values ​​are mapped to a preset range to generate standardized image data. The region-growing segmentation algorithm sets seed points according to the preset interval and calculates the gradient value of each pixel point, performs region growing operation, and generates a layered structure mask containing inner copper wires, micropores and interlayer medium. Based on the edge detection algorithm and tilt angle parameter, the layered structure mask is subjected to boundary tracking, and the spatial coordinates and curvature of the boundary points are calculated to generate a multi-layer geometric feature set; Calculate the mean linewidth, aperture distribution statistics, and interlayer spacing deviation of each layer in the multi-layer geometric feature set, and generate a quantitative feature parameter table based on the deviation values; The quantized feature parameter table is compared with the standard value database to obtain the deviation distribution dataset. The deviation distribution dataset is mapped to the spatial reference frame of the HDI circuit board to form the deviation analysis report.

[0013] The present invention also proposes an X-ray inspection device for HDI printed circuit boards, comprising: The acquisition module is used to acquire the spatial reference system and initial position data of the HDI circuit board to obtain initial detection data; The scanning module is used to perform dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence; The adjustment module is used to drive the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence, so as to obtain a synchronous control sequence; The acquisition module is used to drive the detector to acquire X-ray signals according to the synchronization control sequence and generate a multidimensional projection dataset; The optimization module is used to optimize the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; The extraction module is used to extract multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report.

[0014] The present invention also proposes a computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0015] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0016] Compared with the prior art, this application has the following beneficial effects: This application proposes an X-ray inspection method and apparatus for HDI printed circuit boards. By dynamically adjusting the motion trajectory and tilt angle of the X-ray source, it effectively overcomes the penetration difficulty caused by the excessive thickness of HDI circuit boards in both length and width directions, avoiding the overlap of multi-layer structural features and signal attenuation in the projected image. Simultaneously, the application of a deep learning dynamic weighting algorithm significantly improves image processing accuracy and optimizes the fusion effect of multi-dimensional data, enabling the accurate analysis and extraction of features such as micropores and fine linewidths in complex multi-layer structures. Overall, this application improves the accuracy, reliability, and efficiency of HDI circuit board inspection, making it particularly suitable for high-volume, rapid inspection of high-density, complex-structure circuit boards, significantly reducing false positive and false negative rates. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] The structures, proportions, sizes, etc., shown in the accompanying drawings are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the implementation conditions of this application. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size should still fall within the scope of the technical content disclosed in this application, provided that they do not affect the effects and purposes that this application can produce.

[0019] Figure 1 This is a schematic diagram of the steps of an X-ray inspection method for HDI printed circuit boards according to an embodiment of the present invention; Figure 2 This is a schematic block diagram of the structure of an HDI printed circuit board X-ray inspection device according to an embodiment of the present invention; Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention; 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

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when an element is referred to as “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connection or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0022] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0023] Reference Figure 1 This invention provides an X-ray inspection method for HDI printed circuit boards, comprising the following steps: S1: Obtain the spatial reference system and initial position data of the HDI circuit board to obtain initial detection data; In step S1, a detection device is designed. The X-ray source is mounted above the HDI circuit board, employing a multifocal microtube with an adjustable focal spot size ranging from 0.01 to 0.1 mm and an energy range covering 50-200 keV. This design adapts to the complexity of the multilayer structure of the HDI circuit board, such as the detection requirements for micro-holes and fine linewidths. The detector is mounted below the circuit board, using a high-resolution array detector with a pixel size of 0.025 mm and a dynamic range of 16 bits, capable of capturing high-precision X-ray penetration signals. This top-down layout optimizes the X-ray path, significantly reducing the penetration thickness of the HDI circuit board in the length and width directions, and enhancing the signal capture capability for minute features. The geometric parameters of the HDI circuit board, including length, width, height, and surface morphology, are accurately measured using a laser triangulation system. Laser triangulation technology utilizes the reflection of the laser beam on the circuit board surface to calculate the precise dimensions and surface features of the circuit board using trigonometric principles, such as detecting minute bumps or tilt angles on the circuit board surface. These geometric parameters form the basis for constructing a spatial reference system, which describes the relative position and orientation of the circuit board within the detection system. A high-precision servo motor drives the X-ray source and detector, aligning their central axes with the center of the circuit board to avoid signal acquisition errors caused by positional deviations. Initial trajectory parameters are calibrated using real-time feedback of X-ray attenuation data, generating a configuration file containing geometric parameters, the initial motion trajectory, and material density distribution.

[0024] S2: Perform dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence; In step S2, the X-ray source moves upward along a non-uniform elliptical-spiral composite trajectory. This trajectory is composed of the superposition of an elliptical trajectory and a spiral lifting motion, mathematically expressed as x = a·cos(θ) + Δx, y = b·sin(θ) + Δy, z = k·θ. The major axis a and minor axis b of the elliptical trajectory are dynamically adjusted according to the geometry and material density distribution of the circuit board. For example, a and b may be reduced in high-density areas to improve scanning resolution, and appropriately increased in low-density areas to improve scanning efficiency. Δx and Δy are dynamic offsets determined by the real-time detected material density distribution, and k is the spiral step coefficient, related to the circuit board thickness, typically adjusted within the range of 0.05-0.2 mm / turn. For example, if the circuit board thickness is 1.2 mm, k may be set to 0.1 mm / turn to ensure the spiral motion covers the entire thickness range while avoiding excessively dense scanning that would reduce efficiency. Compared to existing single circular motion, this composite trajectory design can more flexibly adapt to the complex geometry and multilayer characteristics of HDI circuit boards, thereby capturing more precise micropore and defect information. The emission direction of the X-ray source is dynamically adjusted in real time by a servo motor, with a tilt angle range of -20° to +20°. The specific adjustment is based on a deep learning-based adaptive algorithm, with the formula φ = atan(ρ·Δh / d) + k·Δρ, where ρ is the local material density, Δh is the interlayer thickness, d is the X-ray path length, and Δρ is the density gradient. For example, when inspecting an HDI circuit board, if the local density of a certain area is ρ = 8.96 g / cm³, ... 3 Interlayer thickness Δh = 0.1 mm, ray path length d = 10 mm, density gradient Δρ = 0.5 g / cm³ 3With a radius of 0.02 mm and k=0.02, φ≈atan(8.96×0.1 / 10)+0.02×0.5≈5.14° can be calculated. In this way, the tilt angle of the X-ray source can be dynamically optimized according to material properties and structural changes, ensuring that the rays penetrate the target area at the optimal angle, thereby enhancing the imaging contrast of micropores and interlayer connections. The detector moves synchronously below with a mirror trajectory and corresponding tilt angle, maintaining a precise match between the incident and receiving angles of the rays, with an error controlled within 0.08°. This synchronous movement is achieved through a high-precision servo system, ensuring the consistency of movement between the X-ray source and the detector under complex trajectories. For example, when the X-ray source moves along an elliptical trajectory (a=50mm, b=30mm) and is superimposed with a spiral rise and fall (k=0.1 mm / turn), the detector is controlled to move in a mirror manner along an elliptical-spiral trajectory with the same parameters, while simultaneously adjusting the receiving angle to match the tilt angle φ of the X-ray source. This coordinated movement effectively reduces signal distortion caused by angular deviations and improves the quality of multidimensional projection data. The trajectory parameters (a, b, k) and tilt angle φ are dynamically optimized using real-time attenuation data. For example, during scanning, if an X-ray attenuation signal in a certain area indicates the presence of a high-density copper layer, the major axis a and minor axis b of the elliptical trajectory are automatically reduced, while the helical step coefficient k is increased to improve the scanning density in that area. Through this dynamic X-ray scanning and data acquisition process, a motion control sequence is generated. This sequence contains the complete motion trajectory parameters (a, b, k, Δx, Δy) and tilt angle adjustment parameter (φ) for the X-ray source and detector.

[0025] S3: Drive the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence to obtain the synchronization control sequence; In step S3, the X-ray source is driven by a motion control sequence to achieve dynamic multi-track motion in three-dimensional space via a robotic arm or servo system. Simultaneously with the trajectory motion, the X-ray source undergoes adaptive tilt angle adjustment to dynamically adjust its emission angle. During adjustment, a three-loop servo algorithm (track, tilt angle, energy synchronization) ensures that the X-ray source and detector are synchronized. Furthermore, this embodiment introduces a dual-mode energy modulation mechanism, combining high-frequency pulse width modulation (PWM) and low-frequency continuous modulation to dynamically adjust the X-ray source's energy output, ranging from 50-200 keV, with an energy change rate of up to 12 keV / s. This energy modulation mechanism adaptively adjusts based on the material density and thickness of the HDI circuit board. For example, for high-density copper layer areas (density approximately 8.96 g / cm³), the system switches to a high-energy rapid switching mode (150-200 keV) to enhance the penetration ability of the rays and reduce signal attenuation caused by high-density materials. In low-density substrate regions (such as FR-4 material, with a density of approximately 1.8 g / cm³), a low-energy fine modulation mode (50-100 keV) is employed to reduce scattering interference and improve the signal-to-noise ratio of the image. This differentiated energy modulation strategy optimizes the penetration effect of X-rays on different materials, reduces the radiation dose during the detection process, and improves the safety and efficiency of the system. The dynamically adjusted parameters, including trajectory coordinates, tilt angle (e.g., adjusted to 15° to optimize microaperture penetration), and energy value, are recorded and linked with the detector via a three-loop servo algorithm to ensure that the detector can capture high-quality X-ray signals in real time. The generated synchronous control sequence includes motion trajectory and tilt angle information and also integrates dual-mode energy parameters.

[0026] S4: Drive the detector to acquire X-ray signals according to the synchronous control sequence and generate a multidimensional projection dataset; In step S4, the detector is driven by a synchronous control sequence to acquire X-ray signals and generate a multidimensional projection dataset containing various signal types (such as intensity signals, phase contrast signals, and dark field signals). In actual operation, the detector acquires X-ray signals at a high frequency of 150 frames per second, capturing instantaneous signal changes generated by the circuit board during dynamic scanning, forming a continuous multidimensional projection dataset. During signal acquisition, a correction mechanism is used to address projection distortions that may be caused by unevenness on the HDI circuit board surface or multi-angle scanning. Due to the complex multilayer structure of the HDI circuit board, there are slight height differences or morphological changes on its surface, which introduce geometric distortions in the X-ray projection. The dynamic geometric correction mechanism utilizes the top-down characteristics of the X-ray source, combined with the pre-stored circuit board surface morphology data and dynamic tilt angle parameters in the initial detection data, to perform real-time correction of the projection data through an adaptive inverse projection transformation algorithm. For example, assuming a height deviation of 0.1 mm in a certain area of ​​the circuit board, the distortion vector of the projection image is calculated based on the tilt angle φ of the X-ray source (e.g., 30°) and trajectory parameters, and the error is controlled within 0.015 mm through inverse projection transformation. The detector can adjust its gain (ranging from 8 to 70 dB) and exposure time (ranging from 0.5 to 5 ms) based on changes in the energy and angle of the X-ray signal, optimizing signal quality under different detection conditions. For example, when detecting high-density micropore structures in an HDI circuit board, low-energy X-ray signals may be weak due to insufficient penetration. In this case, the detector will automatically increase the gain (e.g., to 50 dB) and extend the exposure time (e.g., 2 ms) to enhance the signal strength. Furthermore, through dynamic denoising technology using nonlinear filtering, combined with the material density distribution characteristics of the circuit board, the acquired signal is denoised in real time to generate a high-resolution projection sequence. Notably, in this embodiment, the top-down configuration of the X-ray source enhances the detector's signal capture capability for micropores and fine lines, reduces interference from scattering artifacts, and allows the top-down X-ray source to irradiate the target area at a more vertical angle, reducing signal blurring caused by lateral scattering and thus improving the contrast of the micropore edges in the projected image.

[0027] S5: The multidimensional projection dataset is optimized based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; In step S5, a deep learning dynamic weighting algorithm is used to optimize and fuse multidimensional data, generating a high-quality comprehensive inspection image reflecting the internal structure of the HDI circuit board. In implementation, the deep learning dynamic weighting algorithm integrates intensity, phase, and dark field signals through a multidimensional signal adaptive fusion algorithm, and dynamically assigns weights based on the dual-modal energy parameters from step S3 and the tilt angle parameters from step S2. Specifically, the intensity signal reflects the degree of absorption of X-rays as they pass through the HDI circuit board, highlighting high-density areas (such as copper layers); the phase signal is more sensitive to material boundaries and minute structural changes, suitable for detecting micropores and edge features; the dark field signal can enhance scattering information, helping to identify minute defects within the material. The deep learning algorithm uses a convolutional neural network (CNN) to extract and fuse these signals. The weight of each signal is dynamically adjusted according to the material density distribution and geometric characteristics of the HDI circuit board. For example, in high-density interconnect regions, the weight of the intensity signal may be increased to enhance the imaging effect of the copper layer. During the fusion process, the region adaptive enhancement technology is combined with the analysis of material density distribution to automatically adjust the gray value range, making the details of the copper layer area more prominent, while avoiding the loss of information in the substrate area due to excessively low gray values, thereby generating a comprehensive detection image.

[0028] Step S6: Perform multi-layer structural feature extraction on the comprehensive detection image to obtain a deviation analysis report.

[0029] In step S6, multi-layer structural features are extracted from the comprehensive inspection image generated in step S5. Combining the top-down view of the X-ray source, the comprehensive inspection image is structurally segmented into multiple sub-regions, each corresponding to a different layer or structure of the circuit board. For example, edge detection and region growing algorithms are used to identify the boundaries of inner copper lines, and morphological operations are used to further separate the contours of microvias. During this process, the adaptive tilt angle in step S3 provides projection data from different perspectives, thereby enhancing the algorithm's ability to analyze complex three-dimensional structures. After structural segmentation, geometric features are extracted, including parameters such as linewidth, aperture, and interlayer spacing. These measurements are compared with standard values ​​to generate a deviation analysis report. The deviation analysis report quantifies the differences between the geometric parameters of each layer of the HDI circuit board and the expected values, and outputs them in a visual form. For example, the report includes a two-dimensional heatmap showing the distribution of linewidth deviations in each region of the circuit board; red areas indicate larger deviations, and green areas indicate deviations within acceptable ranges. It also lists detailed parameters for each layer, such as the average linewidth of the first layer copper lines and the average aperture of the second layer microvias, and marks areas exceeding tolerances.

[0030] In one embodiment, the step of acquiring the spatial reference frame and initial position data of the HDI circuit board to obtain initial detection data includes: Laser triangulation scanning was performed on the HDI circuit board to obtain a set of three-dimensional geometric parameters; The material composition of the circuit board is obtained, and the material composition is partitioned and modeled according to the three-dimensional geometric parameter set to obtain the density distribution matrix. Based on the density distribution matrix, the initial motion trajectory of the HDI circuit board is planned to obtain the initial trajectory parameter set. The X-ray source and detector are aligned along their central axes based on the initial trajectory parameter set to obtain an alignment parameter set. The initial detection data is obtained by feedback calibration of the X-ray attenuation data based on the alignment parameter set.

[0031] In the above embodiments, a laser beam is projected onto the surface of the HDI circuit board, and the surface geometry is calculated by receiving changes in the angle of the reflected light. In practice, a laser scanner can be used to illuminate the circuit board surface at a specific angle. The reflected light is captured by a sensor, and the spatial coordinates of each point on the surface relative to the scanner are calculated using trigonometric relationships, thus forming a three-dimensional point cloud dataset containing height, width, and depth. This point cloud data is the three-dimensional geometric parameter set, describing the surface morphology of the HDI circuit board, including the layout of components on the board, solder joint height, and possible manufacturing defects such as warpage or dents. The material composition of the HDI circuit board is obtained, and the material composition is partitioned and modeled according to the three-dimensional geometric parameter set. Obtaining the material composition can be achieved in various ways, such as referring to the circuit board design documents (Gerber files or BOM lists) or directly detecting the material composition through spectral analysis techniques (such as X-ray fluorescence spectroscopy). After clarifying the material composition, the three-dimensional point cloud data can be spatially mapped with the material information, dividing the circuit board into different regions, such as conductor regions (copper foil), insulating regions (substrate), and cover layer regions (solder resist). Each region is assigned a corresponding density value based on its material properties; for example, the density of copper is approximately 8.96 g / cm³. 3 The density of the FR4 substrate is approximately 1.85 g / cm³. 3Through mathematical modeling, these density values ​​are combined with a set of three-dimensional geometric parameters to form a density distribution matrix. Initial trajectory planning is performed based on this density distribution matrix, utilizing path planning algorithms (such as the A* algorithm or gradient-based optimization algorithms) to design the motion trajectories of the X-ray source and detector. In practice, the target scanning area is determined, and the scanning path is optimized based on the density distribution matrix to minimize scanning time and ensure coverage of all critical areas. The initial trajectory parameter set includes parameters such as the spatial coordinates, moving speed, and angle adjustment of the X-ray source and detector. The X-ray source and detector are aligned along their central axes based on this initial trajectory parameter set. Central axis alignment means aligning the emission center of the X-ray source with the receiving center of the detector. In implementation, mechanical adjustment devices, such as stepper motors or servo systems, can be used to control the position and angle of the X-ray source and detector. Based on the initial trajectory parameter set, the ideal positions of the X-ray source and detector for each scan are calculated and adjusted in real time using a feedback control system (such as a PID controller). During calibration, standard templates or reference objects with known geometry are used to verify alignment accuracy. Based on the alignment parameter set, the X-ray source is driven to emit X-rays, and the detector acquires the penetration signal, generating a raw dataset containing attenuation intensity. By using a real-time feedback algorithm and comparing attenuation data with the density distribution matrix, the focal size and energy output of the X-ray source are calibrated to generate initial detection data containing geometric parameters, calibration trajectory, and attenuation data.

[0032] In one embodiment, the step of performing dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence includes: Based on the initial detection data, a spatial geometric transformation algorithm is used to analyze the spatial reference system and initial position data of the HDI circuit board to obtain the spatial distribution characteristic data of the HDI circuit board. Based on the spatial distribution characteristic data, a non-uniform ellipse-spiral composite trajectory planning algorithm is used to generate the preliminary trajectory parameters of the X-ray source. Based on the preliminary trajectory parameter set, the tilt angle of the X-ray source emission direction is calculated to obtain a dynamic tilt angle sequence; Based on the preliminary trajectory parameter set and dynamic tilt angle sequence, the detector motion is mirrored and synchronized to obtain the detector motion sequence. The X-ray attenuation data is collected and processed according to the detector motion sequence to obtain a dynamic attenuation dataset; The motion control sequence is obtained by dynamically optimizing the preliminary trajectory parameter set and tilt angle based on the dynamic decay dataset.

[0033] In the above embodiments, the HDI circuit board is subjected to multi-dimensional spatial analysis processing based on the initial detection data. Using point cloud data as input, the two-dimensional point cloud data is converted into a three-dimensional spatial distribution through stereomicroscope projection transformation, and spatial frequency components are extracted. High-frequency components reflect the characteristics of fine structures such as micropores, while low-frequency components characterize the changing trend of material density. Based on the generated spatial distribution feature data, a non-uniform ellipse-spiral composite trajectory planning algorithm is used, with material density distribution and micropore distribution as input, to design the dynamic motion trajectory of the X-ray source. Specifically, by calculating the major axis a and minor axis b (range 1-3 mm) of the ellipse trajectory and combining it with the spiral stepping, the trajectory equations x=a·cos(θ)+Δx, y=b·sin(θ)+Δy, and z=k·θ are generated. Here, Δx and Δy are dynamically adjusted according to the material density gradient, and the spiral stepping k is adaptively determined according to the circuit board thickness. In implementation, the gradient descent method is used to optimize the ellipse parameters a and b to ensure that the trajectory can cover the critical areas of the circuit board. Simultaneously, an iterative optimization algorithm is used to adjust the spiral stepping k to ensure the smoothness and continuity of the trajectory. After generating the initial trajectory parameter set, an adaptive tilt angle is calculated for the X-ray source's emission direction. A convolutional neural network (CNN) model is used to predict the material density gradient Δρ, and combined with the interlayer thickness Δh and the ray path length d, the dynamic tilt angle φ is calculated using the formula φ=atan(ρ·Δh / d)+k·Δρ, where k is an adaptive adjustment coefficient (range 0.1-0.5). In implementation, the CNN extracts density gradient features from the initial trajectory parameter set, calculates the initial tilt angle through a nonlinear regression model, and uses feedback data from the servo motor for real-time correction. This allows the X-ray source's emission direction to be optimized based on the local characteristics of the circuit board (such as density variations or structural complexity), thereby improving scanning accuracy and imaging quality. Based on the initial trajectory parameter set and the dynamic tilt angle sequence, using the X-ray source's trajectory equation and tilt angle φ as input, the detector's mirror trajectory equations x'=-x, y'=-y, z'=z+Δz are calculated, where Δz is the fixed distance between the detector and the X-ray source. The detector's tilt angle φ' is highly matched to the X-ray source's tilt angle φ. The detector's mirror trajectory is calculated based on the X-ray source's trajectory parameters. Real-time servo feedback corrects the detector's position and tilt angle, and a PID control algorithm optimizes the smoothness of the motion. After the detector motion sequence is generated, the detector array acquires X-ray signals based on the position and tilt angle parameters in the motion sequence. An analog-to-digital converter (ADC) quantizes the analog attenuation signal into a 16-bit digital signal, and a Kalman filter is used to denoise the signal, generating a dynamic attenuation dataset containing timestamps and spatial coordinates.Based on the dynamic attenuation dataset, a real-time feedback optimization algorithm based on reinforcement learning is adopted. The signal strength and noise level in the attenuation dataset are used as inputs, and the Q-learning algorithm is used to optimize the elliptical trajectory parameters a, b, spiral step k and tilt angle φ. The optimization objective is to minimize the variance of the ray attenuation signal, i.e., min(σ). 2 =min(∑(I_i-I_avg) 2 By calculating the signal variance through attenuation data, iteratively updating the trajectory parameters and tilt angle through Q-learning, and applying the optimized parameters in real time through the servo system, the motion trajectory of the X-ray source and detector can be adaptively adjusted according to the actual acquired data, thereby improving the stability and accuracy of scanning and generating a motion control sequence containing dynamic trajectory parameters and adaptive tilt angle.

[0034] In one embodiment, the step of driving the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence to obtain a synchronization control sequence includes: The motion control sequence is decomposed into independent motion components along the X, Y, and Z axes. The motion sequence along each axis is analyzed in the frequency domain by Fourier transform to generate a multidimensional trajectory component set. Based on the set of multidimensional trajectory components, the motion parameters of the X-ray source are dynamically planned to obtain a set of trajectory optimization parameters. The energy distribution of the X-ray source is divided into regions based on the trajectory optimization parameter set to obtain an energy partitioning sequence. The emission parameters of the X-ray source are subjected to dual-mode modulation processing based on the energy partitioning sequence to obtain the modulation control sequence; The relative positions of the X-ray source and the detector are calibrated according to the modulation control sequence to obtain the synchronization control sequence.

[0035] In the above embodiments, the motion control sequence is decomposed into independent motion components along the X, Y, and Z axes, decoupling the complex multidimensional motion trajectory into motion trajectories in three orthogonal directions. Fourier transform is used to perform frequency domain analysis on the motion sequence along each axis, extracting the dominant and secondary frequency components. Frequency domain characteristics reflect the oscillation characteristics of the motion sequence at different frequencies; for example, the dominant frequency component may correspond to the main motion cycle, while the secondary frequency component reflects subtle trajectory changes. By integrating this frequency domain information, a multidimensional trajectory component set is generated. Based on this multidimensional trajectory component set, the motion parameters of the X-ray source are dynamically programmed. The dynamic programming algorithm, combined with the geometric features of the HDI circuit board, comprehensively considers parameters such as velocity, acceleration, and trajectory curvature to generate a set containing these optimized parameters. After obtaining the trajectory optimization parameter set, the scanning area is divided into high-density and low-density regions according to the material density and thickness distribution of the HDI circuit board. The material properties of different regions have different requirements for X-ray penetration capabilities; for example, high-density regions require higher X-ray energy to ensure sufficient penetration, while low-density regions can use lower energy to reduce radiation dose and improve imaging contrast. An energy allocation algorithm assigns corresponding X-ray energy parameters to each region, forming an energy partitioning sequence containing high-energy and low-energy zones. Based on this energy partitioning sequence, the X-ray source emission parameters are subjected to dual-mode modulation. For high-density regions, high-frequency pulse width modulation (PWM) is used to generate high-energy short-pulse sequences. These short-pulse sequences can provide high-intensity X-ray output in a short time, suitable for penetrating thicker materials. For low-density regions, low-frequency continuous modulation (LCM) is used to generate low-energy stable sequences. The dual-mode modulation algorithm fuses these two sequences to form a modulation control sequence containing modulation frequency and energy intensity. The relative positions of the X-ray source and detector are calibrated according to the modulation control sequence. The calibration process uses a three-loop servo algorithm to monitor and adjust the position, velocity, and acceleration in real time to match the X-ray source's trajectory with the detector's receiving position, generating a servo synchronization parameter set containing position deviation, velocity error, and tilt angle adjustment. Based on this parameter set, a time-series synchronization algorithm aligns the X-ray source's trajectory, tilt angle, and energy parameters with the detector's receiving parameters along the time axis, generating a synchronization control sequence containing all these parameters.

[0036] In one embodiment, the step of driving the detector to acquire X-ray signals according to the synchronization control sequence and generating a multidimensional projection dataset includes: The detector's acquisition parameters are initialized and configured according to the synchronization control sequence to obtain an initial acquisition parameter set; The detector is driven to acquire X-ray signals based on the initial acquisition parameter set, resulting in a raw signal dataset containing intensity, phase contrast, and dark field signals. The detector's receiving position is calibrated by combining the tilt angle parameter in the synchronization control sequence to generate a calibrated signal dataset; Based on the intensity, phase contrast, and dark field signals in the calibrated signal dataset, a multidimensional projection reconstruction algorithm is used to integrate and spatially map the signals, generating a multidimensional projection dataset containing multidimensional spatial information.

[0037] In the above embodiments, the synchronization control sequence is combined with the initial detection data of the HDI circuit board, and the initial acquisition parameter set of the detector is configured through a parameter mapping algorithm. The detector is set according to the frame rate and exposure time in the initial acquisition parameter set, and captures X-ray signals using a multi-channel parallel sampling algorithm. X-ray signals contain various information types: intensity signals reflect the attenuation of X-rays after passing through the object being detected; phase contrast signals capture the phase shift of the X-ray wavefront due to changes in material density; and dark-field signals reveal signal changes caused by scattering due to the microstructure of the material. The multi-channel parallel sampling algorithm, by processing multiple signal channels simultaneously, can capture these different types of signals, thus forming a raw signal dataset containing intensity, phase contrast, and dark-field signals. To address potential distortion issues in the raw signal dataset, the detector's receiving position is calibrated by combining the tilt angle parameters in the synchronization control sequence and the surface topography data of the HDI circuit board. Specifically, through an adaptive inverse projection transformation algorithm, the detector's receiving position can be dynamically adjusted according to the tilt angle parameters in the synchronization control sequence to correct distortions caused by surface unevenness or multi-angle scanning. The adaptive inverse projection transform algorithm analyzes the geometric characteristics of the signal, compares the actual received signal projection with the ideal projection, calculates the correction matrix, and generates a calibrated signal dataset. This dataset has higher spatial consistency and accuracy compared to the original signal dataset. After generating the calibrated signal dataset, the detector gain and exposure time are dynamically adjusted using dynamic gain adjustment and exposure time optimization algorithms to adjust the detector's operating parameters in real time. For example, when the signal intensity distribution shows weak signals in certain areas, the dynamic gain adjustment algorithm increases the detector gain to enhance the signal. The exposure time optimization algorithm adjusts the exposure time for each acquisition based on the signal's dynamic range to ensure the signal-to-noise ratio reaches its optimal state, generating an optimized acquisition parameter set containing optimized gain and exposure time. Based on the optimized acquisition parameters, the calibrated signal dataset undergoes nonlinear denoising processing. A nonlinear filtering algorithm is used, and wavelet transform technology is employed to decompose the signal into high-frequency and low-frequency components. The high-frequency components typically contain noise, while the low-frequency components retain the main characteristics of the signal. Combined with an adaptive thresholding algorithm, high-frequency noise can be effectively separated and removed while preserving key signal information. The advantage of nonlinear denoising is its ability to adapt to the non-uniformity of signals, avoiding the loss of signal details that may be caused by traditional linear filtering, and generating a denoised signal dataset. By integrating and spatially mapping signals through a multidimensional projection reconstruction algorithm, intensity, phase contrast, and dark field signals are integrated into a unified multidimensional dataset. The spatial mapping algorithm then associates these signals with the spatial structure of the object being detected, reconstructing the multidimensional spatial information of the object from projection data from different angles.Multidimensional projection datasets contain intensity information and incorporate additional dimensional information provided by phase contrast and dark field signals, enabling a more comprehensive characterization of the internal structure and material properties of the detected object.

[0038] In one embodiment, the step of optimizing the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image includes: A multi-layer network model is constructed, and the tilt angle parameters in the synchronization control sequence are input into the multi-layer network model for deep learning weighting to obtain a weighted signal set. Identify high-density interconnected regions within the weighted signal set, and enhance the features of these high-density interconnected regions based on gradient operators to generate an enhanced feature set. Based on the feature intensity in the enhanced feature set, the mean and variance of gray level in each region are calculated, and the gray level distribution is optimized by dynamic range compression algorithm to obtain the optimized gray level set. The optimized grayscale set is input into a bilateral filter for initial smoothing and artifact features are extracted. The optimized grayscale set is then iteratively updated based on the artifact features to generate a de-artifacted signal set. Based on the signal distribution in the artifact removal signal set, the signal components are integrated using a multi-dimensional image reconstruction algorithm to generate the comprehensive detection image.

[0039] In the above embodiments, signal component decomposition processing can be performed on the multidimensional projection dataset to separate the intensity signal, phase contrast signal, and dark field signal from the dataset using a signal decomposition algorithm. Specifically, the energy and tilt angle parameters from the synchronization control sequence are first read. The multidimensional projection dataset is then transformed to the frequency domain using a Fast Fourier Transform (FFT), and the spectral features of each signal component are extracted. Deep learning weighting is then performed to generate a weighted signal set. The intensity, phase contrast, and dark field signals from the signal set are input into a CNN model. Simultaneously, the energy and tilt angle parameters from the synchronization control sequence are combined, and the trained weight matrix dynamically weights each signal component. The weights are calculated based on the projection density and geometric features of the high-density interconnected regions, reflecting the importance of the signal in its spatial distribution. For example, high-density interconnected regions typically have higher projection density and more complex geometric structures, thus requiring higher weights to highlight their features. During implementation, the CNN model iteratively optimizes the weight matrix based on the feature distribution of the input signal and the synchronization control sequence parameters, extracting deep features of the signal through multi-layer convolution and pooling operations to generate the weighted signal set. Feature enhancement processing is performed on high-density interconnected regions within the weighted signal set. Edge features are extracted from the weighted signal set using gradient operators, which effectively capture areas with significant signal intensity variations. Local contrast of the high-density interconnected regions is calculated using top-view characteristic parameters, and adaptive histogram equalization (AHE) is employed to enhance the imaging features of edges and micropores. The AHE algorithm enhances image contrast by adjusting the gray-level distribution of local regions, making edge and micropore features more prominent. Logically, high-density interconnected regions within the weighted signal set are first identified, and then gradient operators and the AHE algorithm are applied sequentially to generate an enhanced feature set containing enhanced edge and micropore features. Based on the enhanced feature set, the gray-level distribution of the signal set is optimized. The mean and variance of gray levels in each region are calculated according to the feature intensity within the enhanced feature set, and a dynamic range compression algorithm is used to optimize the gray-level distribution. In implementation, based on the feature distribution of the enhanced feature set, the signal set is divided into high-density interconnected regions and other regions. Gray-level statistical parameters, such as mean and variance, are calculated for each region. These parameters reflect the gray-level distribution characteristics within the region. A dynamic range compression algorithm is applied to adjust the gray-level values, making the gray-level distribution more uniform, thereby improving the overall image clarity and contrast, generating an optimized gray-level set. Artifact suppression processing is then applied to the optimized gray-level set. A bilateral filtering algorithm is used for initial smoothing. A deep learning residual network (ResNet) is used to identify and separate artifact features. ResNet, through its deep structure and residual connections, can effectively distinguish between normal signals and artifact components. An iterative optimization algorithm is used to remove artifact components, update the signal set, and generate a de-artifacted signal set.Image reconstruction is performed based on the artifact-removed signal set. The back-projection filtering (FBP) algorithm maps the artifact-removed signal set to a two-dimensional image space, fusing intensity, phase contrast, and dark field signals. The FBP algorithm reconstructs the initial structure of the image through back-projection operations, and further optimizes the image resolution and contrast using a deep learning optimization model. The deep learning model, through training, learns the feature distribution of the image, effectively enhancing image details and reducing reconstruction errors, outputting a comprehensive detection image. This process, through the integration and optimization of multi-dimensional signals, generates a high-resolution, high-contrast detection image.

[0040] In one embodiment, the step of extracting multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report includes: A linear transformation is performed on the pixel grayscale values ​​of the comprehensive detection image to calculate the mean and standard deviation of the grayscale distribution of each pixel, and the grayscale values ​​are mapped to a preset range to generate standardized image data. The region-growing segmentation algorithm sets seed points according to the preset interval and calculates the gradient value of each pixel point, performs region growing operation, and generates a layered structure mask containing inner copper wires, micropores and interlayer medium. Based on the edge detection algorithm and tilt angle parameter, the layered structure mask is subjected to boundary tracking, and the spatial coordinates and curvature of the boundary points are calculated to generate a multi-layer geometric feature set; Calculate the mean linewidth, aperture distribution statistics, and interlayer spacing deviation of each layer in the multi-layer geometric feature set, and generate a quantitative feature parameter table based on the deviation values; The quantized feature parameter table is compared with the standard value database to obtain the deviation distribution dataset. The deviation distribution dataset is mapped to the spatial reference frame of the HDI circuit board to form the deviation analysis report.

[0041] In the above embodiments, the grayscale histogram of the image is extracted, the pixel frequency of each grayscale level is statistically analyzed, and then the mean and standard deviation of the grayscale distribution are calculated. By calculating the cumulative distribution function (CDF), the original grayscale values ​​can be mapped to a preset interval [0, 255] through linear interpolation. After grayscale normalization, the normalized image data is processed for region segmentation. Geometric parameters, such as the expected positions and sizes of inner copper wires or microvias, are read from the initial detection data. Seed points are set based on these parameters, and the pixel located at the geometric center of the target structure can be selected as the starting point. Region growth is performed by iteratively calculating the gradient difference between each pixel and its neighboring pixels. If the gradient difference between adjacent pixels is less than a preset threshold, they are assigned to the same region. This process continues until all pixels are assigned to the corresponding regions. In this way, a layered structure mask containing inner copper wires, microvias, and interlayer dielectric is generated, where each region corresponds to a different physical structure in the HDI circuit board. Feature boundary extraction is performed on the layered structure mask. The Canny edge detection algorithm is used, and the mask is preprocessed with Gaussian smoothing filtering to reduce noise interference in edge detection. The gradient magnitude and direction of the image are calculated to identify potential edge points, and continuous boundary contours are generated through double threshold detection and edge connection techniques. During boundary tracking, the spatial coordinates and local curvature of each boundary point are calculated by combining the tilt angle parameter, thereby extracting geometric features related to linewidth, aperture, and interlayer spacing. These features are integrated into a multi-layer geometric feature set, which contains geometric information of each layer of the HDI circuit board, such as the width of copper lines, the diameter of microvias, and the spacing of interlayer dielectrics. Statistical analysis methods are used to perform least-squares fitting on the boundary point set of each layer structure to calculate the mean of linewidth, the distribution statistics of aperture (such as mean and variance), and the deviation of interlayer spacing. For example, for inner layer copper lines, the average width is calculated by fitting the boundary point set; for microvias, the aperture distribution characteristics are statistically analyzed; and for interlayer dielectrics, the distance deviation between adjacent layers is measured. These statistical values ​​were compiled into a quantized feature parameter table, presenting the geometric characteristics of each layer. The quantized feature parameter table was compared with a standard value database, calculating the difference between the actual value and the standard value for each parameter. These deviation values ​​were then normalized to ensure the comparability of different parameters. The standard value database may contain the expected geometric parameters of HDI circuit board design, such as the standard width of copper traces, the diameter range of microvias, and the tolerance range of interlayer spacing. The normalized deviation values ​​were stored as a multidimensional array, forming a deviation distribution dataset. The deviation distribution dataset was then visualized using the OpenGL rendering pipeline, mapping the deviation values ​​to color gradients, for example, using red to represent larger deviations and green to represent smaller deviations, thus visually displaying the distribution of deviations.Simultaneously, a 3D model of the HDI circuit board is generated using a spatial reference frame, supplemented with statistical charts such as bar charts and heat maps to further highlight the areas and extent of deviations. These visualizations are then integrated into a deviation analysis report.

[0042] Reference Figure 2 An X-ray inspection device for HDI printed circuit boards, comprising: The acquisition module 100 is used to acquire the spatial reference system and initial position data of the HDI circuit board to obtain initial detection data. The scanning module 200 is used to perform dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence; The adjustment module 300 is used to drive the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence, so as to obtain a synchronous control sequence. Acquisition module 400 is used to drive the detector to acquire X-ray signals according to the synchronization control sequence and generate a multidimensional projection dataset; The optimization module 500 is used to optimize the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; The extraction module 600 is used to extract multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report.

[0043] Reference Figure 3 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores database-related data. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an HDI printed circuit board X-ray inspection method.

[0044] One embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements an X-ray inspection method for HDI printed circuit boards, including the following steps: acquiring the spatial reference frame and initial position data of the HDI circuit board to obtain initial inspection data; performing dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial inspection data to obtain a motion control sequence; driving an X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment based on the motion control sequence to obtain a synchronization control sequence; driving a detector to acquire X-ray signals based on the synchronization control sequence to generate a multi-dimensional projection dataset; optimizing the multi-dimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive inspection image; and extracting multi-layer structural features from the comprehensive inspection image to obtain a deviation analysis report.

[0045] 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 computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various 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), etc.

[0046] 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 X-ray inspection of HDI printed circuit boards, characterized in that, include: Obtain the spatial reference frame and initial position data of the HDI circuit board to obtain initial detection data; Based on the initial detection data, dynamic X-ray scanning and data acquisition are performed on the HDI circuit board to obtain a motion control sequence; Based on the motion control sequence, the X-ray source is driven to perform dynamic multi-track motion and adaptive tilt angle adjustment to obtain a synchronous control sequence; The detector is driven to acquire X-ray signals according to the synchronous control sequence, and a multidimensional projection dataset is generated. The multidimensional projection dataset is optimized using a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; Multi-layer structural features are extracted from the comprehensive detection image to obtain a deviation analysis report.

2. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of acquiring the spatial reference frame and initial position data of the HDI circuit board to obtain initial detection data includes: Laser triangulation scanning was performed on the HDI circuit board to obtain a set of three-dimensional geometric parameters; The material composition of the circuit board is obtained, and the material composition is partitioned and modeled according to the three-dimensional geometric parameter set to obtain the density distribution matrix. Based on the density distribution matrix, the initial motion trajectory of the HDI circuit board is planned to obtain the initial trajectory parameter set. The X-ray source and detector are aligned along their central axes based on the initial trajectory parameter set to obtain an alignment parameter set. The initial detection data is obtained by feedback calibration of the X-ray attenuation data based on the alignment parameter set.

3. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of performing dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence includes: Based on the initial detection data, a spatial geometric transformation algorithm is used to analyze the spatial reference system and initial position data of the HDI circuit board to obtain the spatial distribution characteristic data of the HDI circuit board. Based on the spatial distribution characteristic data, a non-uniform ellipse-spiral composite trajectory planning algorithm is used to generate the preliminary trajectory parameters of the X-ray source. Based on the preliminary trajectory parameter set, the tilt angle of the X-ray source emission direction is calculated to obtain a dynamic tilt angle sequence; Based on the preliminary trajectory parameter set and dynamic tilt angle sequence, the detector motion is mirrored and synchronized to obtain the detector motion sequence. The X-ray attenuation data is collected and processed according to the detector motion sequence to obtain a dynamic attenuation dataset; The motion control sequence is obtained by dynamically optimizing the preliminary trajectory parameter set and tilt angle based on the dynamic decay dataset.

4. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of driving the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence to obtain the synchronization control sequence includes: The motion control sequence is decomposed into independent motion components along the X, Y, and Z axes. The motion sequence along each axis is analyzed in the frequency domain by Fourier transform to generate a multidimensional trajectory component set. Based on the set of multidimensional trajectory components, the motion parameters of the X-ray source are dynamically planned to obtain a set of trajectory optimization parameters. The energy distribution of the X-ray source is divided into regions based on the trajectory optimization parameter set to obtain an energy partitioning sequence. The emission parameters of the X-ray source are subjected to dual-mode modulation processing based on the energy partitioning sequence to obtain the modulation control sequence; The relative positions of the X-ray source and the detector are calibrated according to the modulation control sequence to obtain the synchronization control sequence.

5. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of driving the detector to acquire X-ray signals according to the synchronous control sequence and generating a multidimensional projection dataset includes: The detector's acquisition parameters are initialized and configured according to the synchronization control sequence to obtain an initial acquisition parameter set; The detector is driven to acquire X-ray signals based on the initial acquisition parameter set, resulting in a raw signal dataset containing intensity, phase contrast, and dark field signals. The detector's receiving position is calibrated by combining the tilt angle parameter in the synchronization control sequence to generate a calibrated signal dataset; Based on the intensity, phase contrast, and dark field signals in the calibrated signal dataset, a multidimensional projection reconstruction algorithm is used to integrate and spatially map the signals, generating a multidimensional projection dataset containing multidimensional spatial information.

6. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of optimizing the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image includes: A multi-layer network model is constructed, and the tilt angle parameters in the synchronization control sequence are input into the multi-layer network model for deep learning weighting to obtain a weighted signal set. Identify high-density interconnected regions within the weighted signal set, and enhance the features of these high-density interconnected regions based on gradient operators to generate an enhanced feature set. Based on the feature intensity in the enhanced feature set, the mean and variance of gray level in each region are calculated, and the gray level distribution is optimized by dynamic range compression algorithm to obtain the optimized gray level set. The optimized grayscale set is input into a bilateral filter for initial smoothing and artifact features are extracted. The optimized grayscale set is then iteratively updated based on the artifact features to generate a de-artifacted signal set. Based on the signal distribution in the artifact removal signal set, the signal components are integrated using a multi-dimensional image reconstruction algorithm to generate the comprehensive detection image.

7. The X-ray inspection method for HDI printed circuit boards according to claim 1, characterized in that, The step of extracting multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report includes: A linear transformation is performed on the pixel grayscale values ​​of the comprehensive detection image to calculate the mean and standard deviation of the grayscale distribution of each pixel, and the grayscale values ​​are mapped to a preset range to generate standardized image data. The region-growing segmentation algorithm sets seed points according to the preset interval and calculates the gradient value of each pixel point, performs region growing operation, and generates a layered structure mask containing inner copper wires, micropores and interlayer medium. Based on the edge detection algorithm and tilt angle parameter, the layered structure mask is subjected to boundary tracking, and the spatial coordinates and curvature of the boundary points are calculated to generate a multi-layer geometric feature set; Calculate the mean linewidth, aperture distribution statistics, and interlayer spacing deviation of each layer in the multi-layer geometric feature set, and generate a quantitative feature parameter table based on the deviation values; The quantized feature parameter table is compared with the standard value database to obtain the deviation distribution dataset. The deviation distribution dataset is mapped to the spatial reference frame of the HDI circuit board to form the deviation analysis report.

8. An X-ray inspection device for HDI printed circuit boards, characterized in that, include: The acquisition module is used to acquire the spatial reference system and initial position data of the HDI circuit board to obtain initial detection data; The scanning module is used to perform dynamic X-ray scanning and data acquisition on the HDI circuit board based on the initial detection data to obtain a motion control sequence; The adjustment module is used to drive the X-ray source to perform dynamic multi-track motion and adaptive tilt angle adjustment according to the motion control sequence, so as to obtain a synchronous control sequence; The acquisition module is used to drive the detector to acquire X-ray signals according to the synchronization control sequence and generate a multidimensional projection dataset; The optimization module is used to optimize the multidimensional projection dataset based on a deep learning dynamic weighting algorithm to obtain a comprehensive detection image; The extraction module is used to extract multi-layer structural features from the comprehensive detection image to obtain a deviation analysis report.

9. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method as described in any one of claims 1 to 7.

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 method as described in any one of claims 1 to 7.

Images