PCBA defect image intelligent analysis and diagnosis system and method thereof

Abstract

The invention relates to the technical field of PCBA defect analysis and diagnosis, in particular to a PCBA defect image intelligent analysis and diagnosis system and method, and the system comprises an image collection module, a feature extraction module, a feature transformation module connected with the image collection module, and a detection module connected with the feature transformation module. Based on the continuous coherence feature vector, executing group theory transformation to generate a transformed feature vector; the power system mapping module is connected with the feature transformation module; based on the transformed feature vector, executing nonlinear power system mapping to generate a power system state vector; the matrix manifold projection module is connected with the power system mapping module; based on the power system state vector, matrix manifold projection is executed, and a Stiefel matrix is generated; and the quantum state classification module is connected with the matrix manifold projection module and is used for executing quantum state classification and outputting PCBA defect categories based on the Stiefel matrix, so that the accuracy of PCBA defect detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of PCBA defect analysis and diagnosis, in particular to a PCBA defect image-based intelligent analysis and diagnosis system and method thereof. Background Art

[0002] In the electronics manufacturing industry, quality control of printed circuit board assembly (PCBA) has always been an important and challenging area. With the increasing complexity and miniaturization of electronic products, the detection and diagnosis of PCBA defects have become increasingly difficult. Traditional manual inspection methods are not only time-consuming and labor-intensive, but also prone to omissions, making it difficult to meet the high efficiency and high quality requirements of modern industrial production.

[0003] In recent years, machine vision technology has been widely used in the field of PCBA defect detection. The closest existing technologies mainly include methods based on traditional image processing and methods based on deep learning. Traditional image processing methods usually use edge detection, morphological operations and other technologies to extract PCBA image features, and then use machine learning algorithms such as support vector machines (SVM) to classify defects. This method performs well when dealing with simple defects, but its performance is often unsatisfactory when facing complex or minor defects.

[0004] Deep learning-based methods, such as convolutional neural networks (CNNs), have made significant progress in recent years. These methods can automatically learn image features and outperform traditional methods in some complex scenarios. However, deep learning methods also have some inherent problems. First, they require a large amount of labeled data for training, while in actual production, it is often difficult and expensive to obtain a large number of high-quality defect samples. Second, deep learning models are often "black box" and lack interpretability, which is a significant drawback in industrial applications that require high reliability and traceability.

[0005] In addition, both traditional methods and deep learning methods mainly focus on the local features or statistical properties of images, while ignoring the global topological structure that PCBA defects may have. This leads to their poor performance in dealing with certain complex defect patterns, especially when dealing with multi-scale or cross-regional defects.

[0006] Another common problem is that most existing technologies focus on defect detection and classification, but rarely on defect cause analysis and repair suggestions. This makes the quality control process incomplete and difficult to achieve closed-loop management.

[0007] In the face of these technical challenges, a new PCBA defect image intelligent analysis and diagnosis system is urgently needed that can comprehensively improve the accuracy, reliability and efficiency of defect detection, while providing in-depth defect diagnosis and repair suggestions. Summary of the invention

[0008] The present invention aims to solve the above technical problems and proposes an innovative PCBA defect image intelligent analysis and diagnosis system. The system realizes the intelligent processing of the whole process from defect detection to diagnosis suggestion by integrating advanced technologies such as multi-angle image acquisition, topological feature extraction, group theory transformation, nonlinear dynamic system mapping, matrix manifold projection and quantum state classification.

[0009] The present invention proposes a PCBA defect image intelligent analysis and diagnosis system, comprising:

[0010] Image acquisition module for:

[0011] Get multi-angle images of the PCBA surface;

[0012] Align and stitch the acquired multi-angle images to generate a complete PCBA surface image;

[0013] A feature extraction module is connected to the image acquisition module and is used to:

[0014] Receiving the complete PCBA surface image sent by the image acquisition module;

[0015] Based on the complete PCBA surface image, topological feature extraction is performed to generate a continuous homology feature vector;

[0016] A feature transformation module is connected to the feature extraction module and is used to:

[0017] Receiving the continuous homology feature vector sent by the feature extraction module;

[0018] Based on the continuous homology eigenvector, performing group theory transformation to generate a transformed eigenvector;

[0019] A power system mapping module, connected to the feature transformation module, is used to:

[0020] Receiving the transformed feature vector sent by the feature transformation module;

[0021] Based on the transformed eigenvector, performing nonlinear power system mapping to generate a power system state vector;

[0022] A matrix manifold projection module, connected to the power system mapping module, is used to:

[0023] Receiving a power system state vector sent by the power system mapping module;

[0024] Based on the state vector of the power system, matrix manifold projection is performed to generate a Stiefel matrix; a quantum state classification module is connected to the matrix manifold projection module and is used to:

[0025] Receiving the Stiefel matrix sent by the matrix manifold projection module;

[0026] Based on the Stiefel matrix, quantum state classification is performed to output the PCBA defect category. Preferably, the feature extraction module includes:

[0027] Continuous Coherent Computing Unit, used for:

[0028] Receive complete PCBA surface image;

[0029] Calculating the persistent homology group of the complete PCBA surface image;

[0030] A Betti number extraction unit is connected to the continuous coherence calculation unit and is used for:

[0031] extracting Betti numbers from the persistent homology group;

[0032] Generates persistent homology eigenvectors containing multidimensional Betti numbers.

[0033] Preferably, the feature transformation module comprises:

[0034] Permutation group mapping unit, used for:

[0035] Map the persistent homology eigenvectors to the n-th order symmetry group;

[0036] A group action unit, connected to the permutation group mapping unit, is used to:

[0037] Define group action;

[0038] Based on the group action, a transformed feature vector is generated.

[0039] Preferably, the power system mapping module comprises:

[0040] Differential equation building blocks for:

[0041] Formulate differential equations for nonlinear dynamical systems;

[0042] A Runge-Kutta solving unit, connected to the differential equation building unit, for: solving the differential equation using the Runge-Kutta method;

[0043] Generate the powertrain state vector.

[0044] Preferably, the matrix manifold projection module comprises:

[0045] Stiefel manifold definition unit, used for:

[0046] Define the Stiefel manifold;

[0047] A projection function unit, connected to the Stiefel manifold definition unit, is used to:

[0048] Define the projection function from the state vector of the dynamical system to the Stiefel manifold;

[0049] Perform a projection operation to generate a Stiefel matrix.

[0050] Preferably, the quantum state classification module comprises:

[0051] Quantum state building blocks for:

[0052] Construct quantum states based on Stiefel matrices;

[0053] Measurement operator definition unit, used to:

[0054] Define a set of measurement operators;

[0055] A probability calculation unit, connected to the quantum state construction unit and the measurement operator definition unit, is used to:

[0056] Calculate the probability of different defect categories;

[0057] A maximum probability selection unit is connected to the probability calculation unit and is used to:

[0058] The category with the highest probability is selected as the PCBA defect category.

[0059] Preferably, the image acquisition module comprises:

[0060] Multi-angle image acquisition unit, used for:

[0061] Collect PCBA surface images from different angles;

[0062] An image alignment unit is connected to the multi-angle image acquisition unit and is used to:

[0063] Align the collected multi-angle images;

[0064] An image stitching unit, connected to the image alignment unit, is used to:

[0065] The aligned multi-angle images are stitched together into a complete PCBA surface image.

[0066] As a preference, it also includes:

[0067] A defect diagnosis module, connected to the quantum state classification module, is used to:

[0068] Receiving the PCBA defect category output by the quantum state classification module;

[0069] Generate defect diagnosis results based on preset defect diagnosis rules.

[0070] As a preference, it also includes:

[0071] A repair suggestion module is connected to the quantum state classification module and the defect diagnosis module, and is used to:

[0072] Receiving the PCBA defect category output by the quantum state classification module and the defect diagnosis result output by the defect diagnosis module;

[0073] Generate targeted repair suggestions based on the preset repair suggestion library.

[0074] The PCBA defect image intelligent analysis and diagnosis method comprises the following steps:

[0075] S1, obtaining multi-angle images of the PCBA surface, and aligning and splicing the obtained multi-angle images to generate a complete PCBA surface image;

[0076] S2, performing topological feature extraction on the complete PCBA surface image to generate a continuous homology feature vector;

[0077] S3, performing group theory transformation on the continuous homology eigenvector to generate a transformed eigenvector;

[0078] S4, performing nonlinear power system mapping based on the transformed eigenvector to generate a power system state vector;

[0079] S5, performing matrix manifold projection on the power system state vector to generate a Stiefel matrix;

[0080] S6, performing quantum state classification based on the Stiefel matrix and outputting a PCBA defect category;

[0081] S7: Generate defect diagnosis results and repair suggestions based on the PCBA defect category.

[0082] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0083] The beneficial effects of the present invention are multifaceted, reflecting the innovation and practicality of the system design. First, at the macro level, the present invention establishes a complete PCBA defect analysis framework, covering all aspects from image acquisition to defect diagnosis, and realizing closed-loop management of quality control. This comprehensive approach not only improves the accuracy of defect detection, but also provides strong support for the continuous improvement of the production process.

[0084] From a microscopic perspective, the present invention has achieved technological breakthroughs in multiple key modules. The multi-angle image acquisition module provides comprehensive and high-quality PCBA surface information for subsequent analysis through innovative alignment and splicing algorithms. The topological feature extraction module uses the theory of persistent homology to introduce global topological features in PCBA defect analysis for the first time, effectively capturing the structural information of complex defect patterns. The group theory transformation and nonlinear dynamic system mapping modules cleverly map defect features to high-dimensional space, enhancing the distinguishability of features.

[0085] It is particularly worth mentioning that the present invention introduces the idea of ​​quantum computing through matrix manifold projection and quantum state classification module, which is a brand new attempt in the field of PCBA defect analysis. This method not only improves the accuracy of classification, but also provides a new idea for processing high-dimensional complex data.

[0086] The collaborative design between modules is another significant feature of the present invention. For example, the results of topological feature extraction directly affect the subsequent group theory transformation, and the output of nonlinear dynamic system mapping provides a basis for matrix manifold projection. This close module collaboration solves the problem of feature extraction and classification separation in traditional methods, and realizes a seamless transition from low-level features to high-level semantics.

[0087] In terms of effect, the present invention has shown significant advantages. The defect detection accuracy rate is as high as 98.7% and the classification accuracy rate is 97.3%, far exceeding traditional methods and general deep learning methods. Especially when dealing with complex and subtle defects, the advantages of the present invention are more obvious, which is of great significance for improving the overall quality of electronic products.

[0088] In addition, the present invention achieves a low false detection rate (0.8%) and missed detection rate (1.2%) while maintaining high accuracy. This balance is extremely important for actual production, which can maximize product quality and avoid unnecessary rework and material waste.

[0089] Finally, the present invention not only stops at the defect detection level, but also provides strong support for the continuous improvement of the production process through the defect diagnosis module and the repair suggestion module. This closed-loop design from detection to diagnosis and then to improvement reflects the far-reaching impact of the present invention in practical applications.

[0090] In summary, the PCBA defect image intelligent analysis and diagnosis system of the present invention comprehensively improves the performance of PCBA defect detection and diagnosis through innovative technology combination and ingenious module collaboration, and provides an efficient, reliable and forward-looking solution for quality control in the electronics manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 It is a logic block diagram of the entire system of the present invention.

[0092] Figure 2 It is a logic block diagram of the image acquisition module of the present invention.

[0093] Figure 3 It is a logic block diagram of the feature extraction module of the present invention.

[0094] Figure 4 It is a logic block diagram of the feature transformation module of the present invention.

[0095] Figure 5 It is a logic block diagram of the power system mapping module of the present invention.

[0096] Figure 6 It is a logic block diagram of the matrix manifold projection module of the present invention.

[0097] Figure 7 It is a logic block diagram of the quantum state classification module of the present invention.

[0098] Figure 8 It is a logic block diagram of the defect diagnosis module of the present invention.

[0099] Fig. 9 It is a logic block diagram of the repair suggestion module of the present invention.

[0100] Fig.10 Comparison diagram of the original image and the topological feature map.

[0101] Fig.11 is the state evolution trajectory of the power system.

[0102] Fig.12 This is the Stiefel matrix distribution and classification result diagram. DETAILED DESCRIPTION

[0103] See also Figure 1-12 The present invention provides a PCBA defect image intelligent analysis and diagnosis system and method thereof. The system realizes efficient and accurate identification and classification of PCBA defects by innovatively combining cutting-edge mathematical theories such as topology, group theory, dynamical system theory, differential geometry and quantum mechanics.

[0104] First, the PCBA defect image intelligent analysis and diagnosis system of the present invention includes an image acquisition module 1, a feature extraction module 2, a feature transformation module 3, a power system mapping module 4, a matrix manifold projection module 5 and a quantum state classification module 6. These modules are connected in sequence to form a complete data processing pipeline.

[0105] The image acquisition module 1 is used to obtain multi-angle images of the PCBA surface, and align and stitch these images to generate a complete PCBA surface image. Preferably, the image acquisition module 1 uses a high-resolution camera array to ensure that the subtle details of the PCBA surface are captured. In one embodiment of the present invention, the camera array includes 5 cameras, which respectively shoot the PCBA surface from 0°, 45°, 90°, 135° and 180° angles to obtain comprehensive surface information.

[0106] The feature extraction module 2 is connected to the image acquisition module 1, and is used to receive the complete PCBA surface image and perform topological feature extraction. The present invention innovatively applies the continuous homology theory to extract the topological features of the PCBA image. Specifically, the feature extraction module 2 includes a continuous homology calculation unit 21 and a Betti number extraction unit 22.

[0107] The continuous coherence calculation unit 21 first receives the complete PCBA surface image, and then calculates the continuous coherence group of the image. In the present invention, the calculation of the continuous coherence group adopts the following innovative algorithm:

[0108]

[0109] Where I is the PCBA surface image, H k (I) is the k-dimensional continuous homology group, and d is the highest dimension considered. In practical applications, d=2 is usually taken, because the PCBA surface mainly involves the topological features of 0-dimensional (connected components), 1-dimensional (rings) and 2-dimensional (space, holes). The Betti number extraction unit 22 extracts the Betti number from the calculated continuous homology group and generates a continuous homology feature vector containing multi-dimensional Betti numbers. The calculation formula of the Betti number is as follows:

[0110] b k =rank(Z k )-rank(B k ),

[0111] Among them, b k is the k-dimensional Betti number, Z k is a k-dimensional cyclic group, B k is the k-dimensional boundary group.

[0112] In the system of the present invention, continuous coherence is used to extract the topological features of the PCBA surface image. The PCBA surface may have defects such as poor welding, missing components, and component misalignment, which will change the geometric structure of the PCBA surface. Through continuous coherence, the system can capture these structural changes and convert them into topological feature vectors.

[0113] For example, for areas with poor welding, new connected components or ring structures may be formed, while missing components may cause the voids in certain areas to grow. By calculating the Betti numbers at different scales, the system can identify these topological changes and input them as features into the subsequent classification module. This not only improves the robustness of the system, but also effectively distinguishes different types of defects, especially those tiny defects that are difficult to detect with the naked eye.

[0114] In practical applications, the input of continuous homology is the complete PCBA surface image after alignment and stitching. The system first converts the image into a simplicial complex, which is a discretized representation consisting of geometric elements such as points, edges, and faces. Then, the system constructs a filter sequence that gradually increases the resolution of the image to capture topological features at different scales. Finally, the system generates continuous homology barcodes to represent the life cycle of each topological feature (that is, the time when it appears and disappears). These barcodes are further converted into Betti number feature vectors as the basis for subsequent analysis.

[0115] The feature transformation module 3 is connected to the feature extraction module 2 and is used to receive the continuous homology feature vector and perform group theory transformation. The module includes a permutation group mapping unit 31 and a group action unit 32.

[0116] The permutation group mapping unit 31 maps the continuous homology feature vector to the n-order symmetry group. Here, the selection of n is related to the complexity of the PCBA. For example, for a PCBA with 100 key feature points, n=100 may be selected. The mapping function can be expressed as:

[0117] ψ:v1→S n ,

[0118] Among them, v1 is the continuous homology eigenvector, S n is an n-order symmetric group. The group action unit 32 defines the group action and generates the transformed eigenvector. The group action is defined as follows:

[0119] α:S n ×v1→v1′,

[0120] Among them, v′1 is the transformed eigenvector.

[0121] Through this innovative group theory transformation, the present invention can capture the complex interrelationships between PCBA defect features and improve the accuracy of subsequent classification.

[0122] Group theory transformation is used to enhance the expressive power of feature representation. Defects on the PCBA surface usually have complex geometric structures, and traditional feature extraction methods may not be able to fully capture the intrinsic relationships of these structures. Through group theory transformation, the system can encode the symmetry and interaction between feature points into the feature vector, thereby better describing the geometric characteristics of the defects.

[0123] For example, for poorly welded areas, the relative position and arrangement of welding points may be the key to distinguishing different types of welding defects. Through group theory transformation, the system can capture these arrangement relationships and input them as features into the subsequent dynamical system mapping module. This not only improves the classification accuracy of the system, but also effectively handles defects with similar appearance but different internal structures.

[0124] The input of the group theory transformation is the continuous homology eigenvector v1, which contains the topological characteristics of the PCBA surface. The system first maps the eigenvector to the symmetry group S n Then, the system defines the group action function α, through the symmetry group S n The eigenvector is transformed to generate a new eigenvector v′1. This new eigenvector not only retains the original topological information, but also introduces the symmetry and interaction between the feature points, thus enhancing the expressive power of the feature.

[0125] This innovative design of the present invention not only improves the accuracy of PCBA defect detection, but also greatly enhances the robustness and generalization ability of the system. By combining multiple advanced mathematical theories, the present invention can handle various complex PCBA defect situations and provide a powerful tool for quality control in the electronics manufacturing industry. The power system mapping module 4 of the present invention is connected to the feature transformation module 3, and is used to receive the transformed feature vector and perform nonlinear power system mapping. The innovation of this module is to map the PCBA defect features into a nonlinear power system, thereby capturing the complex dynamic relationship between the features.

[0126] The power system mapping module 4 includes a differential equation building unit 41 and a Runge-Kutta solving unit 42. The differential equation building unit 41 is responsible for building the differential equation of the nonlinear power system. In a preferred embodiment of the present invention, the differential equation takes the following form:

[0127]

[0128] Where v3 is the state vector of the dynamic system, v2 is the transformed eigenvector, f is the nonlinear function, and g is the input function. Preferably, f can be selected as a Lorenz system to introduce chaotic dynamic characteristics:

[0129]

[0130] Wherein, σ, ρ, β are system parameters, and are usually set to σ = 10, ρ = 28, β = 8 / 3. The Runge-Kutta solution unit 42 uses the fourth-order Runge-Kutta method to solve the above differential equation. The specific solution process is as follows:

[0131]

[0132] in,

[0133]

[0134] Here, h is the time step, and h=0.01 is usually taken to ensure calculation accuracy.

[0135] Nonlinear dynamic system mapping is used to capture the complex dynamic relationship between PCBA defect features. Defects on the PCBA surface are usually the result of multiple factors, and traditional static feature extraction methods may not be able to fully describe these dynamic changes. By introducing a nonlinear dynamic system, the system can model the dynamic evolution process between features, thereby better understanding the formation mechanism of defects.

[0136] For example, a poorly welded area may gradually deteriorate over time, causing changes in the strength of the connection between welds. Through nonlinear dynamic system mapping, the system can capture these dynamic changes and input them as features into the subsequent classification module. This not only improves the classification accuracy of the system, but also predicts the development trend of defects, helping engineers take timely repair measures.

[0137] The input of the nonlinear dynamical system mapping is the eigenvector v2 after the group theory transformation. The system first introduces the eigenvector v2 as the input function g(v2) into the nonlinear dynamical system. Then, the system uses the Lorenz system as the nonlinear function f(v3) to construct the evolution equation of the dynamical system. In order to solve the equation, the system uses the fourth-order Runge-Kutta method to gradually update the dynamical system state vector v3. Finally, the dynamical system state vector v3 generated by the system contains the dynamic relationship between the features, which provides the basis for the subsequent matrix manifold projection.

[0138] The matrix manifold projection module 5 of the present invention is connected to the power system mapping module 4 and is used to receive the power system state vector and perform matrix manifold projection. The module includes a Stiefel manifold definition unit 51 and a projection function unit 52.

[0139] The Stiefel manifold definition unit 51 defines a Stiefel manifold, which is a special matrix manifold used in the present invention to represent the high-dimensional structure of PCBA defect features. The definition of the Stiefel manifold is as follows:

[0140] St(p,n)={X∈R n×p :X T X=I p},

[0141] Wherein, p and n are parameters selected according to the dimension of PCBA defect features. In one embodiment of the present invention, p=10 and n=100 may be selected to capture sufficient feature information while maintaining computational efficiency.

[0142] The projection function unit 52 defines the projection function from the state vector of the dynamic system to the Stiefel manifold and performs the projection operation. The projection function π can be expressed as:

[0143] π:v3→St(p,n),

[0144] The specific projection process can be achieved through QR decomposition:

[0145] V4=qr(v3)[:,:p],

[0146] Among them, qr represents the QR decomposition operation, and [:,:p] means taking the first p columns.

[0147] In the PCBA defect image intelligent analysis and diagnosis system, matrix manifold projection is used for dimensionality reduction and feature compression. The defect features on the PCBA surface are usually high-dimensional, and directly classifying them may lead to overfitting problems. Through matrix manifold projection, the system can compress high-dimensional features into low-dimensional space while maintaining the orthogonality between features. This not only improves the classification efficiency of the system, but also effectively avoids overfitting and improves the generalization ability of classification.

[0148] For example, for complex welding defects, the interrelationships between feature points may be high-dimensional. Through matrix manifold projection, the system can compress these high-dimensional features into a low-dimensional space, retaining the key features that can best distinguish different defect types. This enables the system to complete the classification task in a shorter time while ensuring the accuracy of the classification.

[0149] The input of the matrix manifold projection is the dynamical system state vector v3. The system first performs QR decomposition on v3 to generate an orthogonal matrix Q and an upper triangular matrix R. Then, the system takes the first p columns of Q as the projection result V4. This projection result not only retains the orthogonality between features, but also reduces the dimension of the features, providing an efficient feature representation for subsequent quantum state classification.

[0150] The quantum state classification module 6 of the present invention is connected to the matrix manifold projection module 5, and is used to receive the Stiefel matrix and perform quantum state classification, and finally output the PCBA defect category. The innovation of this module is to transform the traditional classification problem into a quantum measurement problem, thereby using the advantages of quantum theory to improve the classification accuracy.

[0151] The quantum state classification module 6 includes a quantum state construction unit 61, a measurement operator definition unit 62, a probability calculation unit 63 and a maximum probability selection unit 64. The quantum state construction unit 61 constructs a quantum state based on the Stiefel matrix. In a preferred embodiment of the present invention, the quantum state is defined as follows:

[0152] |ψ>=V4|0>,

[0153] Wherein, V4 is the Stiefel matrix, and |0> is the initial quantum state. The measurement operator definition unit 62 defines a set of measurement operators for representing different PCBA defect categories. This set of measurement operators {M i Satisfy the completeness condition:

[0154]

[0155] In practical applications, this set of measurement operators can be designed according to the types of defects that may occur in PCBA. For example, for common defects such as welding defects, component missing, component misalignment, etc., corresponding measurement operators can be defined respectively.

[0156] The probability calculation unit 63 calculates the probability of different defect categories. The calculation formula is as follows:

[0157]

[0158] The maximum probability selection unit 64 selects the category with the highest probability as the PCBA defect category:

[0159] c = argmax i p i ,

[0160] The quantum state classification method of the present invention can not only effectively process high-dimensional features, but also capture the quantum entanglement effect between features, thereby performing well in complex PCBA defect classification tasks.

[0161] In the PCBA defect image intelligent analysis and diagnosis system, quantum state classification is used to accurately classify PCBA defects. Traditional classification methods usually rely on Euclidean distance or similarity measurement, which may not fully capture the complex relationship between high-dimensional features. By introducing quantum state classification, the system can use the probability amplitude of the quantum state to represent the features, thereby better describing the quantum entanglement effect between the features. This not only improves the accuracy of classification, but also effectively processes high-dimensional features and avoids overfitting problems.

[0162] For example, for common defects such as poor welding, missing components, and component misalignment, the system can define different measurement operators {M i} to represent these defect categories. Through quantum state classification, the system can capture the quantum entanglement effect between features to more accurately distinguish different types of defects. This enables the system to excel in complex PCBA defect classification tasks, especially in the case of high-dimensional features.

[0163] The input of quantum state classification is the Stiefel matrix V4. The system first uses V4 as the initial quantum state |ψ> and uses a set of measurement operators {M i}Measure the quantum state. Each measurement operator M i Corresponding to a specific defect category, the system calculates the probability p corresponding to each category i , generate probability distribution. Finally, the system selects the category with the highest probability as the classification result. This process not only improves the accuracy of classification, but also effectively handles high-dimensional features and avoids overfitting problems.

[0164] Through the above innovative module design, the PCBA defect image intelligent analysis and diagnosis system of the present invention can efficiently and accurately identify and classify various PCBA defects, providing strong support for quality control in the electronics manufacturing industry. Each module of this system has been carefully designed, which not only reflects the application of cutting-edge mathematical theories, but also takes into account actual engineering needs, achieving a perfect combination of theory and practice. The image acquisition module 1 of the present invention is the starting point of the entire system, and its innovative design ensures the accuracy and reliability of subsequent analysis. The module includes a multi-angle image acquisition unit 11, an image alignment unit 12 and an image stitching unit 13.

[0165] The multi-angle image acquisition unit 11 uses a high-precision camera array to collect PCBA surface images from different angles. In a preferred embodiment of the present invention, five high-resolution cameras are used, located at 0°, 45°, 90°, 135° and 180° angles respectively. This configuration can fully capture the geometric features and potential defects of the PCBA surface. The resolution of each camera can reach 20 million pixels, ensuring the visibility of tiny defects.

[0166] The image alignment unit 12 receives the output of the multi-angle image acquisition unit 11 and performs alignment processing on the acquired multi-angle images. The present invention adopts an innovative feature point matching algorithm for image alignment. The algorithm first extracts the SIFT feature points in each image, and then uses the RANSAC algorithm to perform feature point matching and geometric transformation estimation. Preferably, the threshold of feature point matching is set to 0.7, and the number of RANSAC iterations is set to 1000 to ensure accuracy while taking into account computational efficiency.

[0167] The image stitching unit 13 stitches the aligned multi-angle images into a complete PCBA surface image. The present invention adopts an improved multi-band image fusion algorithm for image stitching. The algorithm first performs wavelet decomposition on each image, then fuses them at different scales, and finally obtains the final stitching result by inverse wavelet transform. In one embodiment of the present invention, the wavelet decomposition adopts a 5-layer Daubechies wavelet, and the fusion rule adopts an adaptive weight method based on regional energy.

[0168] The present invention further includes a defect diagnosis module 7 connected to the quantum state classification module 6. The defect diagnosis module 7 receives the PCBA defect category output by the quantum state classification module 6, and generates a defect diagnosis result based on a preset defect diagnosis rule. In a preferred embodiment of the present invention, the defect diagnosis rule adopts a fuzzy logic reasoning system, which can handle the uncertainty of the classification result.

[0169] Specifically, the defect diagnosis module 7 first converts the defect category probability output by the quantum state classification module 6 into a fuzzy set. For example, for three common defects, welding defects, component missing and component misalignment, the following fuzzy sets can be defined:

[0170]

[0171] Among them, x is the probability of the corresponding defect category.

[0172] Then, the defect diagnosis module 7 applies a series of fuzzy rules for reasoning. For example:

[0173] 1. If the soldering defect level is high and the component missing level is low, the PCBA quality is poor.

[0174] 2. If the component misalignment is moderate and the soldering defects are low, the PCBA quality is average.

[0175] 3. If all defect levels are low, the PCBA quality is good.

[0176] Finally, the defuzzification is performed through the centroid method to obtain the final defect diagnosis result.

[0177] The present invention also includes a repair suggestion module 8, which is connected to the quantum state classification module 6 and the defect diagnosis module 7. The repair suggestion module 8 receives the PCBA defect category output by the quantum state classification module 6 and the defect diagnosis result output by the defect diagnosis module 7, and generates targeted repair suggestions based on a preset repair suggestion library.

[0178] In one embodiment of the present invention, the repair suggestion library adopts a knowledge graph structure, which contains information in multiple dimensions such as defect type, defect degree, possible cause and repair method. The repair suggestion module 8 uses a graph reasoning algorithm, combined with defect category and diagnosis results, to find the most relevant repair suggestion in the knowledge graph.

[0179] For example, for welding defects, the knowledge graph may contain the following information:

[0180] (Welding defect)-[type]->(cold weld);

[0181] (Cold welding)-[Possible reasons]->(Welding temperature is too low);

[0182] (Cold welding)-[Repair method]->(Increase welding temperature);

[0183] (Cold welding)-[Repair method]->(Increase welding time);

[0184] The repair suggestion module 8 finds the most matching repair suggestion path through graph traversal and similarity calculation, and generates repair suggestions in natural language form.

[0185] Finally, the present invention also provides a PCBA defect image intelligent analysis and diagnosis method, which includes the following steps:

[0186] S1, obtaining multi-angle images of the PCBA surface, and aligning and splicing the obtained multi-angle images to generate a complete PCBA surface image;

[0187] S2, performing topological feature extraction on the complete PCBA surface image to generate a continuous homology feature vector;

[0188] S3, performing group theory transformation on the continuous homology eigenvector to generate a transformed eigenvector;

[0189] S4, performing nonlinear power system mapping based on the transformed eigenvector to generate a power system state vector;

[0190] S5, performing matrix manifold projection on the power system state vector to generate a Stiefel matrix;

[0191] S6, performing quantum state classification based on the Stiefel matrix and outputting a PCBA defect category;

[0192] S7, generating defect diagnosis results and repair suggestions based on the PCBA defect category.

[0193] This method innovatively combines multiple advanced mathematical theories to achieve full automation of the entire process from image acquisition to defect diagnosis and repair suggestion generation. Each step has been carefully designed to ensure both the advancement of the theory and the feasibility of practical application.

[0194] For example, in step S2, the continuous homology feature extraction adopts an algorithm based on the continuous homology barcode. The algorithm first constructs a filter sequence of the PCBA image:

[0195]

[0196] Among them, K is the simplicial complex representation of the PCBA image. Then, the persistent homology group in each dimension is calculated:

[0197]

[0198] in, K k The i-dimensional cyclic group in K j Finally, the topological features of PCBA defects are extracted by analyzing the length and distribution of persistent homology barcodes.

[0199] Through this innovative method, the present invention can effectively identify and classify various complex PCBA defects, providing strong technical support for quality control in the electronics manufacturing industry. In order to verify the superiority of the PCBA defect image intelligent analysis and diagnosis system and method of the present invention, a series of comparative experiments were conducted. The following is a detailed comparison of the embodiment of the present invention and two comparative examples:

[0200] Embodiment 1: PCBA defect image intelligent analysis and diagnosis system of the present invention

[0201] In this embodiment, a complete system configuration is used, including innovative modules such as multi-angle image acquisition, topological feature extraction, group theory transformation, nonlinear dynamic system mapping, matrix manifold projection and quantum state classification. The system is tested on a data set containing 1,000 PCBA samples, which cover a variety of common PCBA defect types.

[0202] Comparative Example 1: Traditional Machine Vision Method

[0203] In this comparative example, a method based on traditional image processing and machine learning was used. Specifically, a combination of edge detection, morphological operations, and support vector machine (SVM) classifiers was used. This method represents the PCBA defect detection technology commonly used in the current industry.

[0204] Comparative Example 2: Deep Learning Method

[0205] This comparison uses a deep learning-based approach, specifically using a pre-trained ResNet50 model for feature extraction and then a fully connected layer for classification. This approach represents an advanced technique that has been widely used in the field of computer vision in recent years.

[0206] The following key indicators were selected to evaluate the performance of various methods:

[0207] 1. Defect detection accuracy: the proportion of defects correctly identified.

[0208] 2. Defect classification accuracy: the proportion of correctly classified defects based on detected defects.

[0209] 3. Missed detection rate: the proportion of defects that fail to detect actual defects.

[0210] 4. False detection rate: the proportion of normal parts mistakenly judged as defects.

[0211] 5. Processing time: The average time required to process a single PCBA sample.

[0212] The detection methods of these indicators are as follows:

[0213] Accuracy and missed detection rate: Compare the system output results with the real data set annotated by humans.

[0214] False positive rate: Defects reported by the system are reviewed by professionals.

[0215] Processing time: Use a high-precision timer to record the processing time of each sample.

[0216] The following are the detailed test results:

[0217] index Example 1 Comparative Example 1 Comparative Example 2 Defect detection accuracy 98.7 89.5 95.2 Defect classification accuracy 97.3 85.8 93.6 Missed detection rate 1.2 8.7 4.1 False positive rate 0.8 5.6 2.3 Average processing time 1.2 seconds 0.8 seconds 1.5 seconds

[0218] From the above results, it can be seen that the method of the present invention is significantly superior to traditional methods and deep learning methods in multiple key indicators. The specific analysis is as follows:

[0219] 1. Defect detection accuracy: The method of the present invention achieves a high accuracy of 98.7%, which is 9.2 percentage points higher than the traditional method and 3.5 percentage points higher than the deep learning method. This is mainly due to the innovative combination of topological feature extraction and quantum state classification in the present invention, which can capture the subtle features of PCBA defects.

[0220] 2. Defect classification accuracy: The present invention also performs well in defect classification, reaching an accuracy of 97.3%. This shows that the present invention can not only accurately detect defects, but also accurately identify the defect type, which is crucial for subsequent repair and quality control.

[0221] 3. Missed detection rate and false detection rate: The present invention performs well in both indicators, 1.2% and 0.8% respectively. A low missed detection rate means that almost all defects can be captured, while a low false detection rate indicates that the system will not be overly sensitive and falsely report normal parts as defects. This high-precision performance is of great significance for improving production efficiency and reducing unnecessary retooling.

[0222] 4. Processing time: Although the average processing time of the present invention (1.2 seconds) is slightly longer than that of the traditional method (0.8 seconds), this time difference is completely acceptable considering its significantly improved accuracy. Moreover, the processing speed of the present invention is still faster than that of the deep learning method (1.5 seconds).

[0223] These test results fully demonstrate the superiority of the present invention. By innovatively combining multiple advanced mathematical theories, the present invention not only significantly outperforms the prior art in terms of detection and classification accuracy, but also maintains a high processing efficiency. In particular, the outstanding performance in terms of missed detection rate and false detection rate demonstrates the great potential of the present invention in practical industrial applications.

[0224] It is worth noting that the advantages of the method of the present invention are more obvious when dealing with complex and subtle defects. For example, in some difficult-to-detect welding defect samples, the detection accuracy of the present invention reached 96.5%, while the traditional method and deep learning method were only 82.3% and 88.7% respectively. This fully demonstrates the excellent ability of the present invention in dealing with difficult PCBA defects.

[0225] In summary, the PCBA defect image intelligent analysis and diagnosis system and method of the present invention are significantly superior to the existing technology in all key indicators, especially in dealing with complex and subtle defects. This high-precision, low-error performance is of great significance for improving the quality control level of the electronics manufacturing industry, and is expected to bring significant economic benefits and quality improvement in actual industrial applications.

[0226] Please refer to Fig.10, showing the comparison between the original image and the topological feature map extracted by continuous coherence. In this figure, the topological feature map effectively captures the outline of the solder joint defect in the image, showing the superiority of the method of the present invention in image feature extraction. Through the topological feature map, the position and shape of the solder joint defect can be clearly identified, while traditional image processing methods often have difficulty in accurately locating defects in complex backgrounds. The continuous coherence method extracts the high-order structure of the image through its topological characteristics, and can capture nonlinearities and local details that cannot be identified by traditional feature extraction methods. The topological feature map has a certain robustness to noise during the extraction process, and can still effectively identify defects in the presence of noise interference.

[0227] Please refer to Fig.11 The evolution trajectory of the state vector of the dynamic system in the feature space is shown, reflecting the mapping process from the input features to the final feature space. The figure shows the significant separation of different defect categories in the state space. The trajectory in the state space shows the significant separation of different categories of defects (such as solder defects, non-solder defects) in the feature space, which helps to reduce confusion in the subsequent classification process. Through the evolution trajectory of the dynamic system, we can not only understand the distribution of each category, but also monitor and adjust the performance of the model during the training process. The significant separation of the state vector shows that the system can effectively distinguish different categories when processing feature mapping, thereby improving the accuracy and robustness of the classification.

[0228] Fig.12 The distribution of classification results based on Stiefel manifold projection is shown. The boundaries between categories clearly demonstrate the efficiency of classification. The category distribution after Stiefel matrix projection clearly shows the boundaries between different categories, further proving the efficiency of this method in classification tasks and the ability to accurately divide different categories in multi-dimensional space. Through the projection of the Stiefel matrix, this method can learn the most representative features in low-dimensional space, avoiding the complexity of traditional high-dimensional feature learning and simplifying the training and reasoning process of the model. Clear classification boundaries mean that the classification model can better adapt to complex data sets, reduce the occurrence of misclassification, and improve the overall classification accuracy.

[0229] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. PCBA defect image intelligent analysis and diagnosis system, characterized by ,include: Image acquisition module for: Get multi-angle images of the PCBA surface; Align and stitch the acquired multi-angle images to generate a complete PCBA surface image; A feature extraction module is connected to the image acquisition module and is used to: Receiving the complete PCBA surface image sent by the image acquisition module; Based on the complete PCBA surface image, topological feature extraction is performed to generate a continuous homology feature vector; A feature transformation module is connected to the feature extraction module and is used to: Receiving the continuous homology feature vector sent by the feature extraction module; Based on the continuous homology eigenvector, performing group theory transformation to generate a transformed eigenvector; A power system mapping module, connected to the feature transformation module, is used to: Receiving the transformed feature vector sent by the feature transformation module; Based on the transformed eigenvector, performing nonlinear power system mapping to generate a power system state vector; A matrix manifold projection module, connected to the power system mapping module, is used to: Receiving a power system state vector sent by the power system mapping module; Based on the state vector of the power system, performing matrix manifold projection to generate a Stiefel matrix; A quantum state classification module, connected to the matrix manifold projection module, is used to: Receiving the Stiefel matrix sent by the matrix manifold projection module; Based on the Stiefel matrix, quantum state classification is performed and PCBA defect categories are output.

2. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the feature extraction module includes: Continuous Coherent Computing Unit, used for: Receive complete PCBA surface image; Calculating the persistent homology group of the complete PCBA surface image; A Betti number extraction unit is connected to the continuous coherence calculation unit and is used for: extracting Betti numbers from the persistent homology group; Generates persistent homology eigenvectors containing multidimensional Betti numbers.

3. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the feature transformation module includes: Permutation group mapping unit, used for: Map the persistent homology eigenvectors to the n-th order symmetry group; A group action unit, connected to the permutation group mapping unit, is used to: Define group action; Based on the group action, a transformed feature vector is generated.

4. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the power system mapping module includes: Differential equation building blocks for: Formulate differential equations for nonlinear dynamical systems; A Runge-Kutta solver unit, connected to the differential equation building unit, is used to: Solving the differential equation using the Runge-Kutta method; Generate the powertrain state vector.

5. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the matrix manifold projection module includes: Stiefel manifold definition unit, used for: Define the Stiefel manifold; A projection function unit, connected to the Stiefel manifold definition unit, is used to: Define the projection function from the state vector of the dynamical system to the Stiefel manifold; Perform a projection operation to generate a Stiefel matrix.

6. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the quantum state classification module includes: Quantum state building blocks for: Construct quantum states based on Stiefel matrices; Measurement operator definition unit, used to: Define a set of measurement operators; A probability calculation unit, connected to the quantum state construction unit and the measurement operator definition unit, is used to: Calculate the probability of different defect categories; A maximum probability selection unit is connected to the probability calculation unit and is used to: The category with the highest probability is selected as the PCBA defect category.

7. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , the image acquisition module includes: Multi-angle image acquisition unit, used for: Collect PCBA surface images from different angles; An image alignment unit is connected to the multi-angle image acquisition unit and is used to: Align the collected multi-angle images; An image stitching unit, connected to the image alignment unit, is used to: The aligned multi-angle images are stitched together into a complete PCBA surface image.

8. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , also includes: A defect diagnosis module, connected to the quantum state classification module, is used to: Receiving the PCBA defect category output by the quantum state classification module; Generate defect diagnosis results based on preset defect diagnosis rules.

9. The PCBA defect image intelligent analysis and diagnosis system according to claim 1 is characterized in that , also includes: A repair suggestion module is connected to the quantum state classification module and the defect diagnosis module, and is used to: Receiving the PCBA defect category output by the quantum state classification module and the defect diagnosis result output by the defect diagnosis module; Generate targeted repair suggestions based on the preset repair suggestion library.

10. PCBA defect image intelligent analysis and diagnosis method, characterized by , including the following steps: S1, obtaining multi-angle images of the PCBA surface, and aligning and splicing the obtained multi-angle images to generate a complete PCBA surface image; S2, performing topological feature extraction on the complete PCBA surface image to generate a continuous homology feature vector; S3, performing group theory transformation on the continuous homology eigenvector to generate a transformed eigenvector; S4, performing nonlinear power system mapping based on the transformed eigenvector to generate a power system state vector; S5, performing matrix manifold projection on the power system state vector to generate a Stiefel matrix; S6, performing quantum state classification based on the Stiefel matrix and outputting a PCBA defect category; S7: Generate defect diagnosis results and repair suggestions based on the PCBA defect category.

Images