Board card detection method, device and equipment and storage medium

Abstract

The invention discloses a board card detection method, device and equipment and a storage medium, and relates to the technical field of server testing, and the method comprises the steps: detecting a board card image through employing a pre-training model constructed based on a preset large-scale selective kernel network, and obtaining a first detection result; performing tomography on each to-be-detected board card to obtain a three-dimensional image, and determining a second detection result; sending the first detection result and the second detection result to an upper computer, obtaining a physical defect detection result returned by the upper computer, and determining an initial qualified board card according to the physical defect detection result; and acquiring electric signal data of each initial qualified board card, and determining a target qualified board card in the initial qualified board cards according to the electric signal data. The pre-training model is constructed based on the large-scale selective kernel network to perform board card image detection, the three-dimensional image is obtained through tomography, the appearance detection result is obtained, the qualified board card is further determined according to the electric signal data, and the accuracy of the board card test result is improved through comprehensive test of the appearance and the signal channel.

Description

Technical Field

[0001] The present application relates to the field of server testing technology, and in particular to a board detection method, apparatus, device, and storage medium. Background Art

[0002] In current server design and production, some relatively independent functions are increasingly being isolated as separate boards. This facilitates optimized server layout and differentiated configurations. However, each board within a server often includes numerous functional interfaces, which enable signal interaction with other boards. This interaction typically occurs through plug-in connections or cable connections. Before shipping, each board undergoes a power-on test. FCT (Functional Circuit Test) is required to run the board's basic functions to verify the hardware and related links, ensuring board quality.

[0003] Currently, the testing of boards before use is still in the manual testing stage. Manual testing has disadvantages such as long testing time and lack of accuracy. At the same time, when testing a small number of boards, only a single project can be tested at a time, which is inefficient and very inconvenient to operate. Summary of the Invention

[0004] The present application provides a board card detection method, device, equipment and storage medium, which solves the current problems of insufficient precision and low efficiency in board card testing. Through comprehensive testing from appearance to signal channel, it helps to improve the accuracy of board card test results.

[0005] This application provides a board detection method, including:

[0006] Obtaining a card image of each card to be inspected, and inspecting the card image using a pre-trained model to obtain a first inspection result of each card to be inspected; the pre-trained model is a model constructed based on a preset large selective kernel network;

[0007] Performing a tomographic scan on each board to be inspected to obtain a three-dimensional image of each board to be inspected, and determining a second inspection result of each board to be inspected based on the three-dimensional image;

[0008] Sending the first test result and the second test result to a preset host computer, obtaining the physical defect test results of each board to be tested returned by the preset host computer, and determining an initial qualified board among the boards to be tested based on the physical defect test results;

[0009] The electrical signal data of each initially qualified board card during operation is collected, and the target qualified board card among the initially qualified boards is determined according to the electrical signal data.

[0010] The present application also provides a board detection device, comprising:

[0011] An image detection module is used to obtain a card image of each card to be detected and detect the card image using a pre-trained model to obtain a first detection result of each card to be detected; the pre-trained model is a model built based on a preset large selective kernel network;

[0012] a board scanning module, configured to perform a tomographic scan on each board to be detected to obtain a three-dimensional image of each board to be detected, and determine a second detection result of each board to be detected based on the three-dimensional image;

[0013] The defect detection module is used to send the first detection result and the second detection result to a preset host computer, obtain the physical defect detection result of each board to be detected returned by the preset host computer, and determine the initial qualified board among the boards to be detected based on the physical defect detection result;

[0014] The signal detection module is used to collect the electrical signal data of each initially qualified board during operation, and drive the target qualified board among the initially qualified boards according to the electrical signal data.

[0015] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the above-mentioned board detection methods when executing the computer program.

[0016] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned board detection methods are implemented.

[0017] The present application first obtains the board image of each board to be inspected, and uses a pre-trained model constructed based on a preset large-scale selective kernel network to detect the board image to obtain a first detection result of each board to be inspected, and then performs a tomography scan on each board to be inspected to obtain a three-dimensional image of each board to be inspected and determine a second detection result of each board to be inspected; the first detection result and the second detection result are sent to a preset host computer, and the returned physical defect detection result is obtained. The initial qualified board among the boards to be inspected is determined based on the physical defect detection result, and then the electrical signal data of each initial qualified board during operation is collected, and the target qualified board among the initial qualified boards is determined based on the electrical signal data. Through this application, a pre-trained model can be built based on a preset large-scale selective kernel network to perform board image detection, and a tomographic scan of the board to be inspected can be performed to obtain a three-dimensional image to obtain the physical defect detection results of the board appearance, perform preliminary screening of the boards, and then collect the electrical signal data of the initial qualified boards, detect the signal channels of the boards, and further determine the target qualified boards among the initial qualified boards. In this way, dual automatic testing of board physical defects and signal channel defects is achieved, solving the current problems of inaccuracy and low efficiency in board testing. Comprehensive testing from appearance to signal channels helps to improve the accuracy of board test results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A flow chart of a board detection method provided in an embodiment of the present application;

[0020] Figure 2 This is a diagram of the architecture of a board detection system provided in an embodiment of the present application;

[0021] Figure 3 A flow chart of board detection provided in an embodiment of the present application;

[0022] Figure 4 A flow chart of a specific board detection method provided in an embodiment of the present application;

[0023] Figure 5 A schematic structural diagram of a board detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0026] At present, the testing of boards and cards before use is still in the manual testing stage. Manual testing has disadvantages such as long testing time and lack of accuracy. At the same time, when testing a small number of boards and cards, only a single project can be tested at a time, which is inefficient and very inconvenient to operate. The present application can use a pre-trained model to perform board and card image detection, and perform tomography scanning on the board to be tested to obtain a three-dimensional image to obtain the physical defect detection results of the board and card appearance, and then collect the electrical signal data of the initial qualified board and card, detect the signal channel of the board and card, and further determine the target qualified board and card among the initial qualified board and card. Through comprehensive testing from appearance to signal channel, it helps to improve the accuracy of the board and card test results.

[0027] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0028] like Figure 1 As shown, the embodiment of the present application provides a board detection method, which is described in detail in conjunction with the execution process of the board detection method, including:

[0029] Step S11: Acquire a board image of each board to be detected, and use a pre-trained model to detect the board image to obtain a first detection result of each board to be detected; the pre-trained model is a model constructed based on a preset large selective kernel network.

[0030] In this embodiment, a card image of each card to be inspected is first obtained, and the card image is inspected using a pre-trained model to obtain a first inspection result for each card to be inspected. The pre-trained model is constructed based on a preset Large Selective Kernel Network (LSKNet). It is understood that pre-training is a strategy for training deep learning models, the core of which is to initially train the model using a large-scale dataset, enabling the model to learn universal feature representations. In this embodiment, a model training task can be pre-designed based on a card image library, and a large-scale neural network algorithm structure can be trained to achieve learning. The resulting large-scale neural network algorithm structure and parameters are the pre-trained model. Subsequent tasks can be used to extract features or fine-tune this model to achieve specific task objectives. By pre-training on the image library, the large-scale neural network model can extract rich feature information from the card image to facilitate qualified card inspection. The LSKNet is a convolutional neural network architecture, and the pre-trained model in this embodiment is specifically the YOLOv10 model (You Only Look Once: Unified, Real-Time Object Detection, a target detection model).

[0031] like Figure 2 As shown, in this embodiment, board testing can be achieved using a board interface test fixture, which includes a host computer, a wireless communication module, a main control chip, a time-frequency domain collaborative test system, a signal acquisition module, a multi-level defect intelligent identification module, a data acquisition module, and the board under test. After the main control chip issues a command, the data collected by the data acquisition module can be imported into the multi-level defect intelligent identification module. The multi-level defect intelligent identification module then detects physical defects in the board under test, and the results are input to the host computer via the wireless communication module, which then outputs the first detection result. In this process, the YOLOv10 model can be used to achieve real-time detection without NMS (Non-Maximum Suppression), combined with the LSKNet attention mechanism to enhance the detection of minor board defects.

[0032] Step S12: Perform a tomographic scan on each board to be inspected to obtain a three-dimensional image of each board to be inspected, and determine a second inspection result of each board to be inspected based on the three-dimensional image.

[0033] In this embodiment, a multi-level defect intelligent identification module is used to detect physical defects on the board to be tested, and the results are input into the host computer through the wireless communication module. During the process of the host computer outputting the first detection result, each board to be tested can also be subjected to a tomographic scan to obtain a three-dimensional image of each board to be tested, so as to determine the second detection result of each board to be tested based on the three-dimensional image. Specifically, each board to be tested can first be subjected to a tomographic scan to obtain an initial image of each board to be tested, and then the initial image can be processed using a preset denoising algorithm, and a corresponding three-dimensional visualization model can be generated based on the processed initial image to obtain a three-dimensional image based on the three-dimensional visualization model. The above-mentioned preset denoising algorithm is any one of spatial domain filtering, time domain filtering, transform domain filtering, and non-local mean filtering.

[0034] That is to say, if Figure 3 As shown, this embodiment can deploy a 3D X-ray tomography system to detect microvoid defects at hybrid bond interfaces with nanometer-level resolution (<5nm). Furthermore, the 3D X-ray tomography system also applies a denoising algorithm to reduce image noise and enhance detail contrast. It generates a 3D visualization model through volume rendering or multi-planar reconstruction (MPR), supporting cross-sectional observation from any angle. Specifically, the denoising algorithm can employ any of spatial, temporal, transform, and non-local mean filtering.

[0035] Step S13: Send the first test result and the second test result to a preset host computer, obtain the physical defect test results of each board to be tested returned by the preset host computer, and determine the initial qualified board among the boards to be tested based on the physical defect test results.

[0036] In this embodiment, the first and second test results can be sent to a pre-set host computer, and the physical defect test results of each board to be tested returned by the pre-set host computer are obtained. Based on the physical defect test results, the initial qualified boards among the boards to be tested are determined. In this way, 3D X-ray tomography and YOLOv10 model detection are combined to detect the integrity of the external boards and perform preliminary screening of the boards, reducing the number of channel tests. The results are then input into the host computer, which outputs the initial test results and removes the boards that failed the initial test.

[0037] Step S14: collecting electrical signal data of each initially qualified board during operation, and determining a target qualified board among the initially qualified boards based on the electrical signal data.

[0038] In this embodiment, after removing the board bodies that fail the initial inspection, the electrical signal data of each initially qualified board during operation can be collected, and the target qualified board among the initially qualified boards can be determined based on the electrical signal data. Specifically, an oscilloscope can be used to capture the signal waveform of each initially qualified board during operation to generate a corresponding signal eye diagram, and a vector network analyzer can be used to scan the signal frequency range of each initially qualified board during operation. Based on the signal frequency range, the corresponding insertion loss and return loss can be determined, and the corresponding frequency domain parameters can be obtained based on the insertion loss and return loss. Then, a time domain reflectometer can be used to detect the impedance mutation point of the preset connection position in each initially qualified board to obtain the electrical signal data of each initially qualified board during operation based on the signal eye diagram, frequency domain parameters, and impedance mutation point. In other words, if Figure 2 As shown, an acquisition command can be issued through the main control chip. At this time, the signal acquisition module collects the electrical signal data of the test board body and tests the electrical signal integrity of the test board body through the time-frequency domain collaborative test system. The test results are input to the host computer through the wireless communication module, and the host computer outputs a second test result to determine the target qualified board card. In this way, through the board card interface test fixture, dual testing of the board card is achieved, achieving comprehensive testing from the appearance to the signal channel. In addition, during the above signal transmission process, the signal acquisition module inputs a test digital signal to the main control chip, the data acquisition module inputs a test digital signal to the main control chip, and the test board body inputs an analog voltage signal to the signal acquisition module.

[0039] That is to say, if Figure 3 As shown, this embodiment can construct a time-frequency domain collaborative test system, combining a vector network analyzer (VNA) and a high-speed oscilloscope to implement S-parameter matrix analysis and simultaneous eye height / jitter detection. Secondary testing of the remaining boards can be performed, specifically testing the board channel signals through simultaneous eye height / jitter measurement. The secondary test results are then input to a host computer via the main control chip, which then outputs the secondary test results. Boards that failed the secondary test are then removed to determine the target qualified boards.

[0040] After obtaining the frequency domain parameters based on the insertion loss and return loss, the frequency domain parameters can be converted into a corresponding time domain impulse response through an inverse fast Fourier transform. The signal waveform and the time domain impulse response are then compared. When the comparison result between the signal waveform and the time domain impulse response meets a preset error condition, electrical signal data can be obtained based on the signal eye diagram, the frequency domain parameters, and the impedance mutation point. Accordingly, when determining the target qualified boards among the initial qualified boards based on the electrical signal data, the signal amplitude range, signal stabilization time window, and signal jitter components corresponding to each signal eye diagram are first determined, and the signal jitter components are decomposed into random jitter and deterministic jitter. Then, based on the signal waveform, the rising edge time or falling edge time of each initial qualified board during operation is determined. The signal edge rate of the signal waveform is determined based on the rising edge time or falling edge time. The target qualified boards among the initial qualified boards are determined based on the signal amplitude range, signal stabilization time window, random jitter, deterministic jitter, signal edge rate, frequency domain parameters, and impedance mutation point. In this way, the frequency domain data measured by the VNA is converted into a time domain impulse response through an inverse fast Fourier transform and compared with the time domain waveform measured by the oscilloscope, which helps to verify the accuracy of the signal transmission model and ensure that the collected electrical signal truly reflects the actual working status of the board. Through the above technical solution steps, comprehensive acquisition of the board electrical signal is achieved, which can provide raw data for subsequent time-frequency domain collaborative analysis and obtain accurate real-time detection results. In addition, through the board interface test fixture and board interface test method of this embodiment, through the coordinated cooperation of the YOLOv10 model and the time-frequency domain collaborative test system, the board is preliminarily screened through a deep learning model, and through dual signal channel testing in the time domain and frequency domain, the test efficiency is improved while ensuring more accurate testing.

[0041] That is, this embodiment can convert the frequency domain parameters into the corresponding time domain impulse response through inverse fast Fourier transform to perform time-frequency domain collaborative testing. First, the time-frequency domain collaborative testing system is connected to the signal of the board to be tested (such as through cables or probes), and the main control chip issues an acquisition command to trigger a high-speed oscilloscope or other equipment to start capturing the electrical signal of the board when it is running, and then trigger the oscilloscope to capture ≥10 ^6The oscilloscope uses a signal waveform with a UI (unit interval) to generate an eye diagram and quantify the eye height (signal amplitude range), eye width (signal stable time window), and jitter components (time deviation of signal transitions). The dual-Dirac algorithm is then used to separate the total jitter (TJ) into random jitter (RJ (random jitter) <0.5 ps RMS (Root Mean Square)) and deterministic jitter (DJ (deterministic jitter) <9.5 ps) to verify the timing tolerance of the high-speed interface. The oscilloscope then captures the time it takes for the signal to transition from a low level to a high level (rising edge) or a high level to a low level (falling edge). The rise / fall time test measures the signal edge rate to determine whether it complies with the corresponding protocol specifications of the board and verify whether it meets the rate requirements. The signal frequency range is simultaneously scanned, recording insertion loss and return loss. TDR (time domain reflectometry) is used to detect impedance mutations at locations such as PCB vias and connectors. The S-parameters measured by the VNA are then converted to time-domain pulse responses using IFFT (inverse fast Fourier transform). This is then compared with the waveform measured by the oscilloscope to verify channel model errors and locate physical defects in signal transmission (such as via mismatch and broken wires). By combining S-parameter impedance mutation points with time-domain reflection events, hidden defects such as PCB trace breakage and via impedance mismatch can be identified. Data is collected independently from three dimensions: time-domain waveform, frequency-domain characteristics, and physical impedance. These together form the basis for a comprehensive analysis of the board's electrical signal integrity, improving the accuracy of board testing. The total jitter formula for the aforementioned dual-Dirac algorithm is:

[0042] ;

[0043] in, Dirac Function, representing the discrete distribution characteristics of deterministic jitter (DJ); μDJ is the peak-to-peak value of deterministic jitter (usually corresponding to the separation of DJ); σRJ is the standard deviation of the Gaussian distribution of random jitter (RJ); is the convolution operator, which represents the superposition effect of RJ and DJ.

[0044] This embodiment first obtains a board image of each board to be inspected, and uses a pre-trained model constructed based on a preset large-scale selective kernel network to inspect the board image to obtain a first inspection result of each board to be inspected, and then performs a tomography scan on each board to be inspected to obtain a three-dimensional image of each board to be inspected and determine a second inspection result of each board to be inspected; the first inspection result and the second inspection result are sent to a preset host computer, and the returned physical defect detection result is obtained. The initial qualified board among the boards to be inspected is determined based on the physical defect detection result, and then the electrical signal data of each initial qualified board during operation is collected, and the target qualified board among the initial qualified boards is determined based on the electrical signal data. It can be understood that the above technical solution can be applied to the automatic testing of all boards and cards on computer products such as servers. Through the above technical solution, this embodiment can build a pre-trained model based on a preset large-scale selective kernel network to perform board and card image detection, and perform tomography scanning on the board to be inspected to obtain a three-dimensional image to obtain the physical defect detection results of the board and card appearance, perform preliminary screening of the boards, and then collect electrical signal data of the initial qualified boards and cards, and detect the signal channels of the boards and cards, and further determine the target qualified boards and cards among the initial qualified boards and cards. In this way, dual automatic testing of physical defects and signal channel defects of the boards and cards is achieved, solving the current problems of insufficient accuracy and low efficiency in board and card testing. In addition, the frequency domain data measured by the VNA is converted into a time domain impulse response through an inverse fast Fourier transform, and compared with the time domain waveform measured by the oscilloscope, which helps to verify the accuracy of the signal transmission model, ensure that the collected electrical signal truly reflects the actual working status of the board and card, and achieve comprehensive collection of the board and card electrical signals. Through comprehensive testing from appearance to signal channel, it helps to improve the accuracy of the board and card test results.

[0045] Based on the previous embodiment, it can be seen that the present application can use the pre-trained model to perform board image detection and generate a three-dimensional image of the board to obtain physical defect detection results, and then collect electrical signal data to further determine the qualified board. Next, the process of constructing the pre-trained model will be described in detail in this embodiment. Figure 4 As shown, the embodiment of the present application provides a specific board detection method, including:

[0046] Step S21: Obtain the board data of qualified boards and unqualified boards, and construct corresponding training data based on the board data. Grayscale, size normalization and Gaussian filtering are performed on the training data to obtain preprocessed data, and the preprocessed data is processed based on preset geometric transformation rules to obtain target data.

[0047] In this embodiment, the board data of qualified boards and unqualified boards can be obtained, and corresponding training data can be constructed based on the board data. The training data is grayscaled, size normalized and Gaussian filtered to obtain preprocessed data, and the preprocessed data is processed based on preset geometric transformation rules to obtain target data.

[0048] It is understandable that in this embodiment, YOLOv10 adopts an Anchor-Free design, optimizing detection performance through the following constraints: the point anchoring mechanism simplifies traditional box-based anchors into center point predictions, reducing computational complexity and improving small target detection capabilities; the decoupling head separates classification and regression tasks, alleviating the problem of inconsistent objectives between the two, but requires feature alignment learning through a loss function; dynamic label allocation dynamically allocates positive and negative samples based on the IoU (Intersection over Union) between the predicted box and the true box, avoiding missed detections or false detections caused by fixed thresholds. In addition, YOLOv10's loss function integrates multi-task optimization objectives, specifically the classification loss (VFL, Varifocal Loss), using VFL instead of traditional cross entropy, addressing the problem of class imbalance through dynamic weighting, and focusing on high-quality predicted samples:

[0049] ;

[0050] Among them, qi is the IoU weighted label, pi is the predicted probability, and γ is the adjustment factor.

[0051] Step S22: construct an initial model based on the depthwise separable convolutional model and the preset large selective kernel network, train the initial model using the target data to obtain a pre-trained model, and use the pre-trained model to detect the board image.

[0052] In this embodiment, an initial model can be constructed based on a depthwise separable convolutional model and a preset large selective kernel network. The initial model can be trained using target data to obtain a pre-trained model, and the pre-trained model can be used to detect board images. In addition, during the process of training the initial model using the target data, board defect target points can be determined in the board data, and the corresponding prediction box of the board data can be determined based on the board defect target points. Then, the ground truth box corresponding to the board data can be determined, and the overlap between the prediction box and the ground truth box can be determined. Based on the overlap, the positive and negative samples in the training data can be determined, so that the initial model can be trained based on the positive samples, negative samples, and target data to obtain a pre-trained model.

[0053] Specifically, during model training in this embodiment, multimodal data containing both normal and defective samples, such as PCB surface images and 3D CAD (Computer Aided Design) models, is first collected to generate a dataset, which is then annotated and divided into a test set and a validation set. Grayscaling, size normalization, and noise removal (Gaussian filtering) are then performed on the captured images to improve data consistency. Random geometric transformations, such as ±5° rotation, 10% translation, and ±20% color perturbation brightness adjustment, are also applied to enhance data diversity and improve the model's adaptability to complex industrial scenarios. The YOLOv10 model is then deployed for board defect recognition. Depthwise separable convolution is used to extract multi-scale features, and the LSKNet module is integrated to focus on small defect areas, such as ±0.1mm pin offset. Spatial-channel collaborative attention is used to improve feature discrimination. The YOLOv10 model is then trained on the test set and deployed on the inspection equipment for initial testing of the boards to be inspected. The above technical solution is used to build and train the YOLOv10 model, thereby improving the efficiency of identifying board defects.

[0054] Through the above technical solution, this embodiment uses the YOLOv10 model to achieve real-time detection without NMS, abandoning the preset anchor box method in traditional target detection, and directly predicting the center point and bounding box of the target (such as the defect locations of pins and interfaces on the board) through the point anchoring mechanism. There is no need to generate a large number of candidate anchor boxes, which fundamentally reduces the generation of overlapping boxes and reduces the dependence on NMS post-processing. At the same time, it reduces the amount of calculation and improves real-time performance. In addition, the head structure is decoupled, and the classification task (determining whether it is a defect) and the regression task (locating the bounding box of the defect) of the model are separated into two independent branches. Through their respective feature extraction and calculation, the conflict between the inconsistent objectives of the two types of tasks is alleviated. At the same time, the feature learning of the two types of tasks is aligned through subsequent loss functions (such as VarifocalLoss), ensuring that the predicted bounding box is more accurate and further reducing redundant boxes. During training, positive and negative samples are dynamically divided according to the intersection over union (IoU) between the predicted box and the actual defect box, avoiding missed detections (such as tiny defects being misclassified as negative samples due to low IoU) or false detections (such as background areas being misclassified as positive samples due to loose thresholds) caused by fixed thresholds. This allows the model to learn more accurate target box predictions during the training phase, and can output valid results without NMS during inference. Dynamic label allocation improves sample quality and ensures the accuracy and efficiency of defect recognition.

[0055] like Figure 5 As shown, an embodiment of the present application further provides a board detection device, comprising:

[0056] The image detection module 11 is used to obtain a card image of each card to be detected and detect the card image using a pre-trained model to obtain a first detection result of each card to be detected; the pre-trained model is a model constructed based on a preset large selective kernel network;

[0057] The board scanning module 12 is used to perform a tomographic scan on each board to be detected to obtain a three-dimensional image of each board to be detected, and to determine a second detection result of each board to be detected based on the three-dimensional image;

[0058] The defect detection module 13 is configured to send the first detection result and the second detection result to a preset host computer, obtain the physical defect detection result of each board to be detected returned by the preset host computer, and determine the initial qualified board among the boards to be detected based on the physical defect detection result;

[0059] The signal detection module 14 is used to collect the electrical signal data of each initially qualified board during operation, and drive the target qualified board among the initially qualified boards according to the electrical signal data.

[0060] In this embodiment, a pre-trained model can be constructed based on a preset large-scale selective kernel network to perform board image detection, and a tomographic scan of the board to be inspected can be performed to obtain a three-dimensional image to obtain the physical defect detection results of the board appearance, perform preliminary screening of the boards, and then collect electrical signal data of the initial qualified boards, detect the signal channels of the boards, and further determine the target qualified boards among the initial qualified boards. In this way, dual automatic testing of board physical defects and signal channel defects is achieved, solving the current problems of inaccuracy and low efficiency in board testing. Comprehensive testing from appearance to signal channels helps to improve the accuracy of board test results.

[0061] In some specific embodiments, the signal detection module specifically includes:

[0062] An eye diagram generating unit is used to capture the signal waveform of each initially qualified board card during operation using an oscilloscope to generate a corresponding signal eye diagram;

[0063] A signal analysis unit is used to use a vector network analyzer to scan the signal frequency range of each initially qualified board when it is running, and determine the corresponding insertion loss and return loss based on the signal frequency range, and obtain corresponding frequency domain parameters based on the insertion loss and return loss;

[0064] An impedance detection unit, configured to detect an impedance mutation point at a preset connection position in each initially qualified board using a time domain reflectometer;

[0065] The data generating unit is used to obtain the electrical signal data of each initially qualified board during operation according to the signal eye diagram, frequency domain parameters and impedance mutation point.

[0066] In some specific embodiments, the signal detection module further includes:

[0067] a parameter conversion unit, configured to convert the frequency domain parameters into corresponding time domain impulse responses by inverse fast Fourier transform;

[0068] The signal comparison unit is used to compare the signal waveform and the time domain pulse response, so as to obtain the electrical signal data according to the signal eye diagram, frequency domain parameters and impedance mutation point when the comparison result of the signal waveform and the time domain pulse response meets the preset error condition.

[0069] In some specific embodiments, the signal detection module specifically includes:

[0070] A signal determination unit, used to determine the signal amplitude range, signal stabilization time window and signal jitter component corresponding to each signal eye diagram;

[0071] A signal decomposition unit, used for decomposing signal jitter components into random jitter and deterministic jitter;

[0072] a parameter determination unit, configured to determine a rising edge time or a falling edge time of each initially qualified board card during operation based on the signal waveform, so as to determine a signal edge rate of the signal waveform based on the rising edge time or the falling edge time;

[0073] The board detection unit is used to determine the target qualified board among the initial qualified boards based on the signal amplitude range, signal stable time window, random jitter, deterministic jitter, signal edge rate, frequency domain parameters and impedance mutation point.

[0074] In some specific embodiments, the board scanning module specifically includes:

[0075] The board scanning unit is used to perform a tomographic scan on each board to be detected to obtain an initial image of each board to be detected;

[0076] An image processing unit is used to process an initial image using a preset denoising algorithm, and generate a corresponding three-dimensional visualization model based on the processed initial image, so as to obtain a three-dimensional image based on the three-dimensional visualization model; the preset denoising algorithm is any one of spatial domain filtering, temporal domain filtering, transform domain filtering and non-local mean filtering.

[0077] In some specific embodiments, the board detection device further includes:

[0078] A data construction module is used to obtain board data of qualified boards and unqualified boards, and to construct corresponding training data based on the board data;

[0079] The data preprocessing module is used to grayscale, size normalize and Gaussian filter the training data to obtain preprocessed data, and process the preprocessed data based on preset geometric transformation rules to obtain target data;

[0080] A model building module for building an initial model based on a depthwise separable convolutional model and a preset large selective kernel network;

[0081] The model training module is used to train the initial model using the target data to obtain a pre-trained model.

[0082] In some specific embodiments, the model training module specifically includes:

[0083] A prediction frame determination unit is used to determine a board defect target point in the board data, and determine a prediction frame corresponding to the board data based on the board defect target point;

[0084] An overlap determination unit, used to determine the true frame corresponding to the board data and determine the overlap between the predicted frame and the true frame;

[0085] The model training unit is used to determine the positive samples and negative samples in the training data according to the overlap, so as to train the initial model based on the positive samples, the negative samples and the target data to obtain a pre-trained model.

[0086] For the description of the features in the embodiment corresponding to the board card detection device, reference can be made to the relevant description of the embodiment corresponding to the board card detection method, which will not be repeated here.

[0087] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0088] An embodiment of the present application further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned embodiments of the board detection method.

[0089] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above-mentioned embodiments of the board detection method when running.

[0090] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0091] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any of the above-mentioned embodiments of the board detection method are implemented.

[0092] An embodiment of the present application also provides another computer program product, including a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, implementing the steps in any of the above-mentioned board detection method embodiments.

[0093] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0094] The above is a detailed introduction to the board detection method, device, equipment and storage medium provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only intended to help understand the method and core ideas of the present application. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A board detection method, characterized in that: include: Obtaining a card image of each card to be detected, and detecting the card image using a pre-trained model to obtain a first detection result of each card to be detected; the pre-trained model is a model constructed based on a preset large selective kernel network; Performing a tomographic scan on each of the boards to be inspected to obtain a three-dimensional image of each of the boards to be inspected, and determining a second inspection result of each of the boards to be inspected based on the three-dimensional image; Sending the first test result and the second test result to a preset host computer, obtaining the physical defect detection result of each of the boards to be tested returned by the preset host computer, and determining an initial qualified board among the boards to be tested based on the physical defect detection result; The electrical signal data of each of the initially qualified boards during operation is collected, and a target qualified board among the initially qualified boards is determined according to the electrical signal data.

2. The board detection method according to claim 1, characterized in that: The collecting of electrical signal data of each of the initially qualified boards during operation includes: Using an oscilloscope to capture the signal waveform of each of the initially qualified boards during operation, and generating a corresponding signal eye diagram; Scanning the signal frequency range of each of the initially qualified boards during operation using a vector network analyzer, determining corresponding insertion loss and return loss based on the signal frequency range, and obtaining corresponding frequency domain parameters based on the insertion loss and return loss; Using a time domain reflectometer to detect the impedance mutation point at a preset connection position in each of the initially qualified boards; The electrical signal data of each of the initially qualified boards during operation is obtained according to the signal eye diagram, the frequency domain parameters and the impedance mutation point.

3. The board detection method according to claim 2, characterized in that: After obtaining the corresponding frequency domain parameters according to the insertion loss and the return loss, the method further includes: Converting the frequency domain parameters into corresponding time domain impulse responses by inverse fast Fourier transform; The signal waveform and the time domain pulse response are compared, so that when a comparison result between the signal waveform and the time domain pulse response meets a preset error condition, the electrical signal data is obtained according to the signal eye diagram, the frequency domain parameters and the impedance mutation point.

4. The board detection method according to claim 3, characterized in that: The step of determining a target qualified board among the initially qualified boards according to the electrical signal data includes: Determining the signal amplitude range, signal stabilization time window and signal jitter component corresponding to each of the signal eye diagrams; Decomposing the signal jitter component into random jitter and deterministic jitter; Determine a rising edge time or a falling edge time of each of the initially qualified boards when running based on the signal waveform, so as to determine a signal edge rate of the signal waveform based on the rising edge time or the falling edge time; The target qualified board among the initial qualified boards is determined based on the signal amplitude range, the signal stable time window, the random jitter, the deterministic jitter, the signal edge rate, the frequency domain parameters, and the impedance mutation point.

5. The board detection method according to claim 1, characterized in that: The step of performing a tomographic scan on each of the boards to be inspected to obtain a three-dimensional image of each of the boards to be inspected includes: Performing a tomographic scan on each of the boards to be inspected to obtain an initial image of each of the boards to be inspected; The initial image is processed using a preset denoising algorithm, and a corresponding three-dimensional visualization model is generated based on the processed initial image, so as to obtain the three-dimensional image based on the three-dimensional visualization model; the preset denoising algorithm is any one of spatial domain filtering, temporal domain filtering, transform domain filtering and non-local mean filtering.

6. The board detection method according to any one of claims 1 to 5, characterized in that: Before detecting the board image using the pre-trained model, the method further includes: Obtaining board data of qualified boards and unqualified boards, and constructing corresponding training data based on the board data; Gray-scaling, size normalization, and Gaussian filtering are performed on the training data to obtain pre-processed data, and the pre-processed data is processed based on a preset geometric transformation rule to obtain target data; Building an initial model based on a depthwise separable convolutional model and the preset large selective kernel network; The initial model is trained using the target data to obtain the pre-trained model.

7. The board detection method according to claim 6, characterized in that: The process of using the target data to train the initial model to obtain the pre-trained model includes: Determining a board defect target point in the board data, and determining a corresponding prediction box of the board data based on the board defect target point; Determining a true frame corresponding to the board data, and determining an overlap between the predicted frame and the true frame; Positive samples and negative samples in the training data are determined according to the overlap, so as to train the initial model based on the positive samples, the negative samples and the target data to obtain the pre-trained model.

8. A board detection device, characterized in that: include: An image detection module is configured to obtain a card image of each card to be detected and detect the card image using a pre-trained model to obtain a first detection result of each card to be detected; the pre-trained model is a model constructed based on a preset large selective kernel network; a board scanning module, configured to perform a tomographic scan on each of the boards to be detected to obtain a three-dimensional image of each of the boards to be detected, and determine a second detection result of each of the boards to be detected based on the three-dimensional image; a defect detection module, configured to send the first detection result and the second detection result to a preset host computer, obtain the physical defect detection result of each of the boards to be detected returned by the preset host computer, and determine an initial qualified board among the boards to be detected based on the physical defect detection result; The signal detection module is used to collect the electrical signal data of each of the initially qualified boards during operation, and drive the target qualified boards among the initially qualified boards according to the electrical signal data.

9. An electronic device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the board detection method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the board detection method according to any one of claims 1 to 7 are implemented.

Images