Multi-layer circuit board quality inspection system based on machine learning
Through a multi-layer circuit board quality inspection system based on machine learning, image features are automatically learned and scratches are identified, and the problem of large errors and low efficiency in the existing technology is solved, achieving a more efficient and accurate quality inspection process.
Patent Information
- Application Number
- CN202411499709.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-02-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing multi-layer circuit board quality inspection methods mainly rely on manual inspection, with large errors and low detection efficiency.
Adopt a multi-layer circuit board quality inspection system based on machine learning, which includes data acquisition module, data preprocessing module, feature extraction module, machine learning model module, model evaluation and optimization module, system integration module and real-time detection module, through depth The learning model automatically learns image features, recognizes and classifies scratches.
It improves the accuracy and efficiency of multi-layer circuit board quality inspection, reduces human errors, and improves the automation level and product quality of the production line.
Smart Images
Figure CN119443934A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of circuit board quality inspection, and specifically relates to a multi-layer circuit board quality inspection system based on machine learning. Background Art
[0002] Multilayer Printed Circuit Board (MLPCB) is a support for electronic components, which is composed of multiple layers of printed circuits, including vias and insulating layers. It is mainly used for printed circuit boards and is widely used in electronic devices. The quality of multilayer circuit boards directly affects the performance and life of electronic products. Therefore, the quality inspection of multilayer circuit boards is crucial. The main contents of the quality inspection of multilayer circuit boards include: Appearance inspection: The quality inspector will inspect the appearance of the multilayer circuit board to ensure that its surface is flat, free of scratches, bubbles, warping and other defects. Dimension measurement: The quality inspector will measure the dimensions of the multilayer circuit board, including length, width, thickness, etc., to ensure that it meets the design requirements. Electrical performance test: The electrical performance test of the multilayer circuit board includes resistance, capacitance, insulation resistance, withstand voltage and other test items to ensure that it has good electrical performance. Solder joint inspection: The quality inspector will inspect the solder joints of the multilayer circuit board to ensure that it is firmly welded, without defects such as false solder joints and leaking solder joints. Functional test: The functional test of the multilayer circuit board includes signal integrity, high-speed signal transmission, noise and other test items to ensure that it has good functional performance. Environmental adaptability test: Quality inspectors will conduct environmental adaptability tests on multi-layer circuit boards, including temperature, humidity, vibration and other test items to ensure that they have good environmental adaptability.
[0003] However, common quality inspection methods mostly use manual inspection, which not only has large errors but also makes the inspection efficiency relatively low. Summary of the invention
[0004] The purpose of the present invention is to provide a multi-layer circuit board quality inspection system based on machine learning in order to solve the above-mentioned problems.
[0005] The technical solution adopted by the present invention is as follows: a multi-layer circuit board quality inspection system based on machine learning, the quality inspection system comprising:
[0006] Data acquisition module, data preprocessing module, feature extraction module, machine learning model module, model evaluation and optimization module, system integration module and real-time detection module;
[0007] The machine learning model module is internally provided with a network design module, a model training module and a model verification module.
[0008] In a preferred embodiment, the data acquisition module uses a high-resolution camera or image scanning device to capture images of the circuit board. These images need to contain various types of scratches, as well as normal circuit boards without scratches, so that the system can learn to distinguish between the two. The collected images should have high contrast and clear details for subsequent processing. At the same time, the data acquisition process should ensure stable environmental conditions, such as constant lighting and angle, to reduce interference with image analysis. In order to improve the generalization ability of the model, images covering different scenes and different lighting conditions should be collected.
[0009] In a preferred embodiment, the feature extraction module is responsible for extracting information useful for scratch detection from the preprocessed image. In traditional machine learning methods, this may include manually designed features such as edge detection and texture analysis. In deep learning methods, such as convolutional neural networks (CNNs), feature extraction is automatic, and the network learns the mapping from raw pixels to scratch features. Deep learning models are able to learn higher-level and more abstract feature representations, which are crucial for distinguishing scratched and non-scratched areas.
[0010] In a preferred embodiment, the network design module includes: defining the model: selecting a suitable neural network architecture, such as a fully connected network Dense Neural Network, a convolutional neural network CNN or a recurrent neural network RNN, etc. Adding layers: adding input layers, hidden layers and output layers according to the complexity of the problem. Each hidden layer can use an activation function, such as ReLU, Sigmoid or Tanh. Compiling the model: configuring the model's loss function, such as mean square error MSE, cross entropy Cross-Entropy, etc., optimizers such as Adam, SGD, etc. and evaluation indicators such as accuracy Accuracy.
[0011] In a preferred embodiment, the model training module first standardizes the feature matrix and then divides the data into a training set, a validation set, and a test set. Next, a simple fully connected neural network is designed, which includes two hidden layers with 64 and 32 nodes respectively, and uses the ReLU activation function. The output layer has one node and uses the Sigmoid activation function, which is suitable for binary classification problems. The model is compiled using the Adam optimizer and the binary cross entropy loss function, and then trained on the training set for 10 cycles, with a batch size of 32 for each cycle. Validation data is also provided to monitor the performance of the model during training.
[0012] In a preferred embodiment, the specific process of the model verification module includes:
[0013] Dataset segmentation: Divide the entire dataset into several mutually exclusive subsets, each of which is called a "fold". K-fold cross-validation is usually used, where k is the number of folds. For example, in 5-fold cross-validation, the dataset is divided into 5 folds.
[0014] Model training and validation: Perform k iterations, select a different fold as the validation set in each iteration, and the remaining k-1 folds as the training set. In each iteration, use the training set to train the model, and use the validation set to evaluate the performance of the model.
[0015] Performance evaluation: For each iteration, record the performance indicators of the model on the validation set, such as accuracy, recall, F1 score, etc. The choice of performance indicators depends on the specific application scenario and problem type.
[0016] Model evaluation: The performance metrics of k iterations are averaged to get the average performance of the model on the entire dataset. This average provides a more stable estimate of model performance because it takes into account the performance of the model on different subsets.
[0017] Parameter adjustment: According to the results of cross-validation, adjust the parameters of the model or choose a different model architecture. This process may need to be repeated many times until a model configuration with optimal performance and stability is found.
[0018] Final model training: Once the best model configuration is determined, the model is trained using the entire dataset (except for a held-out test set).
[0019] Test set evaluation: The final trained model is evaluated using the reserved test set to obtain an estimate of the model’s performance in real-world applications.
[0020] In a preferred embodiment, the machine learning model module uses features learned from the data set to identify and classify scratches. For circuit board scratch detection, the machine learning model module selects the Fast R-CNN machine learning model, which learns the features of scratches through the training data set and can accurately identify scratches during testing. The choice of model depends on the specific application requirements, computing resources, and detection accuracy requirements.
[0021] In a preferred embodiment, the model evaluation and optimization module is used to ensure that the performance of the model meets expectations. This involves using a test set to evaluate the model's performance indicators such as accuracy, recall, F1 score, etc. If the performance of the model does not meet the requirements, it needs to be tuned. This may include adjusting model parameters, using different feature combinations, or changing the model architecture. In addition, in order to prevent the model from overfitting, cross-validation and other techniques are usually used to ensure the generalization ability of the model on unknown data.
[0022] In a preferred embodiment, the system integration module integrates the trained model into a complete software system that can receive circuit board images uploaded by users, process them, and return scratch detection results. This usually involves building a user-friendly interface that allows users to easily upload images and view detection results. In addition, the system also needs to have the ability to handle multiple user requests, manage data and model updates, and ensure system stability and availability.
[0023] In a preferred embodiment, the real-time detection module is the final application stage of the system, which allows users to upload circuit board images and obtain scratch detection results in real time. This module needs to run efficiently in order to quickly respond to user requests. The system automatically processes the uploaded images, uses the trained model for analysis, identifies scratches, and outputs the location and type of scratches. In order to improve the user experience, the system can also provide visualization results to help users intuitively understand the specific situation of the scratches. In addition, the real-time detection module should also have the ability to self-update and adapt to new data to maintain the long-term effectiveness and accuracy of the system.
[0024] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0025] 1. In the present invention, the network design module selects a suitable neural network architecture, such as CNN, RNN or DNN, and the system can customize the model for specific quality inspection tasks to improve the accuracy and efficiency of detection.
[0026] The network design module can automatically learn image features through deep learning architecture, reduce reliance on manual feature engineering, and make the system more adaptable to the diversity of scratches on circuit boards. By adding different types of layers and activation functions, the network design module can build a model that can handle complex problems and improve the robustness of the system. The model training module standardizes the feature matrix so that the model can better adapt to data distribution, improve training efficiency and model performance. Providing a validation data set can monitor the performance of the model during training, adjust the training strategy in time, and prevent overfitting. The model training module uses modern optimization algorithms such as Adam, combined with appropriate loss functions, to accelerate the convergence of the model and improve training efficiency.
[0027] 2. In the present invention, the model verification module can evaluate the generalization ability of the model on unknown data through the cross-validation method, ensuring that the model has good stability and predictive ability. Using performance indicators such as accuracy, recall rate, F1 score, etc., the classification ability of the model can be comprehensively evaluated to ensure that the model can meet the quality inspection requirements in practical applications. Adjusting the model parameters according to the results of cross-validation can help find the optimal model configuration and further improve the detection accuracy and reliability of the system. Overall, the setting of these modules enables the entire system to handle multi-layer circuit board quality inspection tasks more flexibly and efficiently, improve the automation level and product quality of the production line, reduce human errors, improve production efficiency, and provide data support for management to optimize production processes. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 is the overall system block diagram of the present invention;
[0029] Figure 2 This is a system block diagram of the machine learning model module in the present invention. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0031] Reference Figure 1-2 ,
[0032] A multi-layer circuit board quality inspection system based on machine learning, the quality inspection system comprising:
[0033] Data acquisition module, data preprocessing module, feature extraction module, machine learning model module, model evaluation and optimization module, system integration module and real-time detection module;
[0034] The machine learning model module is internally configured with a network design module, a model training module, and a model verification module.
[0035] The data acquisition module uses a high-resolution camera or image scanning device to capture images of the circuit board. These images need to contain various types of scratches, as well as normal circuit boards without scratches, so that the system can learn to distinguish between the two. The acquired images should have high contrast and clear details for subsequent processing. At the same time, the data acquisition process should ensure stable environmental conditions, such as constant lighting and angle, to reduce interference with image analysis. In order to improve the generalization ability of the model, images covering different scenes and different lighting conditions should be collected.
[0036] The feature extraction module is responsible for extracting useful information for scratch detection from the preprocessed image. In traditional machine learning methods, this may include manually designed features such as edge detection and texture analysis. In deep learning methods, such as convolutional neural networks (CNN), feature extraction is automatic, and the network learns the mapping from raw pixels to scratch features. Deep learning models are able to learn higher-level and more abstract feature representations, which are crucial for distinguishing scratched and non-scratched areas.
[0037] The network design module includes: Define model: Select a suitable neural network architecture, such as DenseNeural Network, Convolutional Neural Network (CNN), or Recurrent Neural Network (RNN). Add layer: Add input layer, hidden layer, and output layer according to the complexity of the problem. Each hidden layer can use an activation function, such as ReLU, Sigmoid, or Tanh. Compile model: Configure the model's loss function, such as mean square error (MSE), cross entropy (Cross-Entropy), optimizer, such as Adam, SGD, and evaluation indicators, such as accuracy.
[0038] The model training module first normalizes the feature matrix and then splits the data into training, validation, and test sets. Next, a simple fully connected neural network is designed, which contains two hidden layers with 64 and 32 nodes respectively, and uses the ReLU activation function. The output layer has one node and uses the Sigmoid activation function, which is suitable for binary classification problems. The model is compiled using the Adam optimizer and the binary cross entropy loss function, and then trained on the training set for 10 cycles, with a batch size of 32 per cycle. Validation data is also provided to monitor the performance of the model during training.
[0039] The specific process of the model verification module includes:
[0040] Dataset segmentation: Divide the entire dataset into several mutually exclusive subsets, each of which is called a "fold". K-fold cross-validation is usually used, where k is the number of folds. For example, in 5-fold cross-validation, the dataset is divided into 5 folds.
[0041] Model training and validation: Perform k iterations, select a different fold as the validation set in each iteration, and the remaining k-1 folds as the training set. In each iteration, use the training set to train the model, and use the validation set to evaluate the performance of the model.
[0042] Performance evaluation: For each iteration, record the performance indicators of the model on the validation set, such as accuracy, recall, F1 score, etc. The choice of performance indicators depends on the specific application scenario and problem type.
[0043] Model evaluation: The performance metrics of k iterations are averaged to get the average performance of the model on the entire dataset. This average provides a more stable estimate of model performance because it takes into account the performance of the model on different subsets.
[0044] Parameter adjustment: According to the results of cross-validation, adjust the parameters of the model or choose a different model architecture. This process may need to be repeated many times until a model configuration with optimal performance and stability is found.
[0045] Final model training: Once the best model configuration is determined, the model is trained using the entire dataset (except for a held-out test set).
[0046] Test set evaluation: The final trained model is evaluated using the reserved test set to obtain an estimate of the model’s performance in real-world applications.
[0047] The machine learning model module uses the features learned from the dataset to identify and classify scratches. For circuit board scratch detection, the machine learning model module selects the Fast R-CNN machine learning model, which learns the features of scratches through the training dataset and can accurately identify scratches during testing. The choice of model depends on the specific application requirements, computing resources, and detection accuracy requirements.
[0048] The model evaluation and optimization module is used to ensure that the performance of the model meets expectations. This involves using a test set to evaluate the model's performance indicators such as accuracy, recall, and F1 score. If the model's performance does not meet the requirements, it needs to be tuned. This may include adjusting model parameters, using different feature combinations, or changing the model architecture. In addition, in order to prevent the model from overfitting, techniques such as cross-validation are usually used to ensure the model's generalization ability on unknown data.
[0049] The system integration module integrates the trained model into a complete software system that can receive circuit board images uploaded by users, process them, and return scratch detection results. This usually involves building a user-friendly interface that enables users to easily upload images and view detection results. In addition, the system also needs to have the ability to handle multiple user requests, manage data and model updates, and ensure the stability and availability of the system.
[0050] The real-time detection module is the final application stage of the system, which allows users to upload circuit board images and obtain scratch detection results in real time. This module needs to run efficiently in order to respond quickly to user requests. The system automatically processes the uploaded images, uses the trained model for analysis, identifies scratches, and outputs the location and type of scratches. In order to improve the user experience, the system can also provide visualization results to help users intuitively understand the specific situation of the scratches. In addition, the real-time detection module should also have the ability to self-update and adapt to new data to maintain the long-term effectiveness and accuracy of the system.
[0051] In the present invention, the network design module selects a suitable neural network architecture, such as CNN, RNN or DNN, and the system can customize the model for specific quality inspection tasks to improve the accuracy and efficiency of detection.
[0052] The network design module can automatically learn image features through deep learning architecture, reduce reliance on manual feature engineering, and make the system more adaptable to the diversity of scratches on circuit boards. By adding different types of layers and activation functions, the network design module can build a model that can handle complex problems and improve the robustness of the system. The model training module standardizes the feature matrix so that the model can better adapt to data distribution, improve training efficiency and model performance. Providing a validation data set can monitor the performance of the model during training, adjust the training strategy in time, and prevent overfitting. The model training module uses modern optimization algorithms such as Adam, combined with appropriate loss functions, to accelerate the convergence of the model and improve training efficiency.
[0053] In the present invention, the model verification module can evaluate the generalization ability of the model on unknown data by the method of cross-validation, and ensure that the model has good stability and predictive ability. Using performance indicators such as accuracy, recall rate, F1 score, etc., the classification ability of the model can be comprehensively evaluated to ensure that the model can meet the quality inspection requirements in practical applications. Adjusting the model parameters according to the results of cross-validation can help find the best model configuration and further improve the detection accuracy and reliability of the system. Overall, the setting of these modules enables the entire system to handle multi-layer circuit board quality inspection tasks more flexibly and efficiently, improve the automation level and product quality of the production line, reduce human errors, improve production efficiency, and provide data support for management to optimize the production process.
[0054] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0055] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A multi-layer circuit board quality inspection system based on machine learning, characterized in that: The quality inspection system comprises: Data acquisition module, data preprocessing module, feature extraction module, machine learning model module, model evaluation and optimization module, system integration module and real-time detection module; The machine learning model module is internally provided with a network design module, a model training module and a model verification module.
2. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The data acquisition module uses a high-resolution camera or image scanning device to capture images of circuit boards; these images need to contain various types of scratches, as well as normal circuit boards without scratches, so that the system can learn to distinguish between the two; the captured images should have high contrast and clear details for subsequent processing; at the same time, the data acquisition process should ensure stable environmental conditions, such as constant lighting and angle, to reduce interference with image analysis; in order to improve the generalization ability of the model, images covering different scenes and different lighting conditions should be collected.
3. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The feature extraction module is responsible for extracting information useful for scratch detection from the preprocessed image; in traditional machine learning methods, this may include manually designed features such as edge detection and texture analysis; while in deep learning methods, such as convolutional neural networks, feature extraction is automatic, and the network learns the mapping from raw pixels to scratch features; The deep learning model is able to learn higher-level and more abstract feature representations that are crucial for distinguishing scratch and non-scratch areas.
4. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The network design module includes: defining the model: selecting a suitable neural network architecture, such as a fully connected network Dense Neural Network, a convolutional neural network CNN or a recurrent neural network RNN; adding layers: adding input layers, hidden layers and output layers according to the complexity of the problem; using activation functions for each hidden layer: configuring the model's loss function such as mean square error MSE, cross entropy Cross-Entropy, optimizers such as Adam, SGD and evaluation indicators such as accuracy.
5. The multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The model training module first standardized the feature matrix and then divided the data into training set, validation set and test set; then, a simple fully connected neural network was designed, which contained two hidden layers with 64 and 32 nodes respectively, and used the ReLU activation function; the output layer had one node and used the Sigmoid activation function, which was suitable for binary classification problems; the model was compiled using the Adam optimizer and the binary cross entropy loss function, and then trained on the training set for 10 cycles, with a batch size of 32 for each cycle; validation data was also provided to monitor the performance of the model during the training process.
6. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The specific process of the model verification module includes: Dataset segmentation: Divide the entire dataset into several mutually exclusive subsets, each subset is called a "fold"; k-fold cross-validation is usually used, where k is the number of folds; for example, in 5-fold cross-validation, the dataset is divided into 5 folds; Model training and validation: Perform k iterations, select a different fold as the validation set in each iteration, and use the remaining k-1 folds as the training set; in each iteration, use the training set to train the model, and use the validation set to evaluate the performance of the model; Performance evaluation: For each iteration, record the performance indicators of the model on the validation set, such as accuracy, recall, and F1 score; the choice of performance indicators depends on the specific application scenario and problem type; Model evaluation: The performance metrics of k iterations are averaged to obtain the average performance of the model on the entire dataset; this average provides a more stable estimate of model performance because it takes into account the performance of the model on different subsets; Parameter adjustment: According to the results of cross-validation, adjust the parameters of the model or choose a different model architecture; this process may need to be repeated multiple times until a model configuration with optimal performance and stability is found; Final model training: Once the best model configuration is determined, the model is trained using the entire dataset (except for the reserved test set). Test set evaluation: The final trained model is evaluated using the reserved test set to obtain an estimate of the model’s performance in real-world applications.
7. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The machine learning model module uses the features learned from the data set to identify and classify scratches. For circuit board scratch detection, the machine learning model module selects the Fast R-CNN machine learning model, which learns the features of scratches through the training data set and can accurately identify scratches during testing. The choice of model depends on the specific application requirements, computing resources and detection accuracy requirements.
8. The multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The model evaluation and optimization module is used to ensure that the performance of the model meets expectations; this involves using a test set to evaluate the model's accuracy, recall, and F1 score performance indicators; if the model's performance does not meet the requirements, it needs to be tuned; this may include adjusting model parameters, using different feature combinations, or changing the model architecture; in addition, in order to prevent the model from overfitting, cross-validation techniques are usually used to ensure the model's generalization ability on unknown data.
9. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The system integration module integrates the trained model into a complete software system that receives circuit board images uploaded by users, processes them, and returns scratch detection results. This usually involves building a user-friendly interface that enables users to easily upload images and view detection results. In addition, the system also needs to have the ability to handle multiple user requests, manage data and model updates, and ensure the stability and availability of the system.
10. A multi-layer circuit board quality inspection system based on machine learning as claimed in claim 1, characterized in that: The real-time detection module is the final application stage of the system, which allows users to upload circuit board images and obtain scratch detection results in real time; This module needs to run efficiently in order to quickly respond to user requests; the system will automatically process uploaded images, use the trained model for analysis, identify scratches, and output the location and type of scratches; in order to improve the user experience, the system also provides visualization results to help users intuitively understand the specific circumstances of the scratches; in addition, the real-time detection module should also have the ability to self-update and adapt to new data to maintain the long-term effectiveness and accuracy of the system.