Electronic component classification prediction method and device

Abstract

The embodiment of the invention provides an electronic component classification prediction method and device. The method comprises the following steps: acquiring a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; the simplified three-dimensional circuit board model comprises an electronic component entity model; inputting the simplified three-dimensional circuit board model into a trained deep learning model combined with a graph neural network to obtain a classification label of an electronic component entity model in the three-dimensional circuit board model output by the deep learning model; and the deep learning model extracts geometric features and arrangement position information from boundary representation data features of the electronic component entity model of the simplified three-dimensional circuit board model so as to predict classification labels of the electronic component entity model in the simplified three-dimensional circuit board model according to the geometric features and the arrangement position information. According to the embodiment of the invention, the electronic components can be accurately classified.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of computer technology, and in particular to a method for classifying and predicting electronic components, an apparatus for classifying and predicting electronic components, an electronic device, and a computer-readable storage medium. Background Art

[0002] When performing electronic cooling simulation on a PCB (Printed Circuit Board), each electronic component on the PCB has its own unique thermal characteristics, including but not limited to heat generation, thermal resistance, and heat capacity. These parameters directly affect the heat distribution and heat dissipation of the entire PCB. Only by accurately identifying the electronic components and inputting their corresponding physical property parameters can the accuracy and reliability of the electronic cooling simulation be ensured. Therefore, accurately identifying the electronic components on the PCB is crucial. Summary of the Invention

[0003] In view of the above problems, a method and device for electronic component classification prediction is proposed to overcome or at least partially solve the above problems. The specific technical solution is as follows: In a first aspect of the present invention, a method for electronic component classification prediction is provided, the method comprising: Obtaining a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes solid models of electronic components; and the simplified three-dimensional circuit board model has fewer geometric features than the unsimplified three-dimensional circuit board model; The simplified three-dimensional circuit board model is input into a trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

[0004] Optionally, the deep learning model includes an encoder and a neural network model, and the simplified three-dimensional circuit board model is input into the trained deep learning model to obtain the classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, including: The encoder is used to extract geometric features of the electronic component entity model in the simplified three-dimensional circuit board model; wherein the geometric features include geometric information and topological structure; The simplified three-dimensional circuit board model and the geometric features are input into the neural network model to obtain a classification label of the electronic component entity model in the simplified three-dimensional circuit board model output by the neural network model.

[0005] Optionally, before inputting the simplified three-dimensional circuit board model into a trained deep learning model to obtain a classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, the method further includes: Acquire a sample electronic component entity model of the sample electronic component; wherein the sample electronic component entity model has corresponding actual geometric features, and the actual geometric features include actual geometric information and actual topological structure; Using the sample electronic component physical model to perform self-supervised training on the encoder to be trained; When the encoder after self-supervised training reaches a preset training condition, the trained encoder is obtained.

[0006] Optionally, before inputting the simplified three-dimensional circuit board model into a trained deep learning model to obtain a classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, the method further includes: Obtaining a sample three-dimensional circuit board model of a sample printed circuit board; wherein the sample three-dimensional circuit board model includes a sample electronic component entity model, and the sample electronic component entity model has a corresponding actual classification label; Calculating the center point distance between the center points of the sample electronic component physical models; Calculating the intimacy matrix between the sample electronic component physical models according to the center point distance; Determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component physical models and a preset threshold; Converting the sample electronic component physical model into a coding unit; For each coding unit of the sample printed circuit board, construct subgraph data by combining the geometric feature vector of the coding unit with the geometric feature vectors of other coding units that have a connection relationship in reverse order of the center point distance of the coding unit, and combine all the subgraph data of the sample printed circuit board into an input sequence; The input sequence is used to pre-train the neural network model to be trained.

[0007] Optionally, determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component physical models and a preset threshold includes: When the intimacy matrix between the sample electronic component physical models is greater than the preset threshold, it is determined that a connection relationship exists between the sample electronic component physical models.

[0008] Optionally, pre-training the neural network model to be trained using the input sequence includes: Randomly performing masking on the coding units in the input sequence to obtain a masked input sequence; Inputting the masked input sequence into the neural network model to be trained for training, so that the neural network model predicts the predicted classification label of the masked coding unit based on the unmasked coding unit in the input sequence; Calculate the loss value according to the predicted classification label and the actual classification label corresponding to the masked encoding unit; When the loss value reaches a preset convergence condition, the pre-trained neural network model is obtained.

[0009] Optionally, the neural network model includes a multi-layer perceptual network, which is used to classify according to the feature vector corresponding to the first coding unit in the input sequence output by the neural network model, so as to output the predicted classification label corresponding to the first coding unit to perform classification fine-tuning training on the pre-trained neural network model to obtain a trained deep learning model.

[0010] In a first aspect of the present invention, there is also provided an electronic component classification prediction device, the device comprising: A three-dimensional circuit board model acquisition module is used to obtain a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes solid models of electronic components; and the geometric features of the simplified three-dimensional circuit board model are fewer than the geometric features of the unsimplified three-dimensional circuit board model; A classification label acquisition module is used to input the simplified three-dimensional circuit board model into a trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

[0011] In another aspect of the present invention, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium. When the computer-readable storage medium is run on a computer, the computer executes any of the above-mentioned electronic component classification prediction methods.

[0012] In another aspect of the present invention, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to execute any of the above-mentioned electronic component classification prediction methods.

[0013] Compared with the related art, the embodiments of the present invention have at least the following advantages: In an embodiment of the present invention, a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation is obtained, the simplified three-dimensional circuit board model may include an electronic component physical model, and the geometric features of the simplified three-dimensional circuit board model are fewer than the geometric features of the unsimplified three-dimensional circuit board model. The simplified three-dimensional circuit board model is input into a trained deep learning model combined with a graph neural network, and the deep learning model can output classification labels corresponding to the electronic component physical models in the simplified three-dimensional circuit board model, wherein the deep learning model can extract geometric features and arrangement position information from the boundary representation data features of the electronic component physical models of the simplified three-dimensional circuit board model. In this way, the deep learning model can predict the classification labels of the electronic component physical models in the simplified three-dimensional circuit board model based on the geometric features and arrangement position information. When performing electronic heat dissipation simulation on a printed circuit board in an embodiment of the present invention, the deep learning model can accurately predict the geometric features and arrangement position information of the electronic component physical model of the simplified three-dimensional circuit board model, and then determine the classification labels of the electronic components in the printed circuit board based on the geometric features and arrangement position information, thereby improving the recognition accuracy of the electronic components. Furthermore, the physical properties required for the electronic component simulation can be accurately determined based on the identified classification labels, thereby ensuring the accuracy and reliability of the electronic heat dissipation simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments or the prior art.

[0015] Figure 1 A flowchart of the steps of a method for classifying and predicting electronic components provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of an electronic component classification prediction device provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings in the embodiments of the present invention.

[0017] Currently, the identification method for electronic components on printed circuit boards (PCBs) mainly relies on manual inspection, which is inefficient and prone to errors, especially when there are a large number of electronic components on the PCB.

[0018] With the development of artificial intelligence technology, it has become possible to use deep learning models to automatically identify electronic components on printed circuit boards. However, current deep learning models mostly classify and identify electronic components by extracting their geometric features. During electronic heat dissipation simulation, electronic components are generally simplified to reduce the differences between their geometric features. There are many categories of electronic components, and some have similar geometric shapes. Therefore, relying solely on the geometric features of electronic components for identification is difficult to achieve the accuracy requirements of electronic heat dissipation simulation.

[0019] In a specific implementation, the arrangement of electronic components on a printed circuit board has certain regularities (such as the relative positions between electronic components, the layout regularity of electronic components with different classification labels on a PCB, etc.). Identifying electronic components with the help of the arrangement position information (position information) of electronic components on a printed circuit board helps to improve the accuracy of identifying electronic components. Therefore, the AI (Artificial Intelligence) model of an embodiment of the present invention, i.e., a deep learning model, extracts the geometric features of electronic components while combining the arrangement position information of electronic components on a printed circuit board, and integrates multiple aspects of information to identify electronic components on a printed circuit board. The embodiment of the present invention realizes the extraction of geometric features of electronic components and the extraction of arrangement position information on a printed circuit board. Combining geometric features and arrangement position information, electronic components can be predicted and identified, and then electronic heat dissipation simulation can be accurately and quickly completed based on the identification results. It should be added that the geometric features are 3D geometric features (three-dimensional geometric features). The geometric features may at least include the surface area of the geometric body, the length and curvature of the edges, and the UV mesh features of the faces. The topological structure mainly includes the connection relationship between faces connected by common edges, forming the topological relationship of the graph structure.

[0020] Reference Figure 1 , is a flowchart of the steps of a method for classifying and predicting electronic components provided in an embodiment of the present invention, such as Figure 1 As shown, the method may specifically include the following steps: Step 101: Obtain a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes solid models of electronic components; and the geometric features of the simplified three-dimensional circuit board model are fewer than the geometric features of the unsimplified three-dimensional circuit board model.

[0021] In specific implementation, electronic heat dissipation simulation is a method that uses computer simulation technology to predict and optimize the heat dissipation effect of printed circuit boards. In this way, the heat dissipation performance of printed circuit boards can be estimated in advance during the design stage of printed circuit boards, and possible problems with printed circuit boards can be identified, such as excessively high temperatures in local areas of the printed circuit boards, excessively high temperatures in areas where key electronic components are located, and other problems. In this way, the design of printed circuit boards can be optimized in advance to ensure the reliability and stability of products using printed circuit boards.

[0022] In a specific implementation, the three-dimensional circuit board model (3D model) of the printed circuit board may include electronic component physical models (physical model / 3D model) of electronic components, such as physical models of electronic components such as on-board devices, radiators, axial fans, flat fans, blowers, and centrifugal fans.

[0023] An embodiment of the present invention obtains a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation based on an actual simulation scenario, wherein the simplified three-dimensional circuit board model includes solid models of electronic components, and the geometric features of the simplified three-dimensional circuit board model are fewer than those of the unsimplified three-dimensional circuit board model. This can improve simulation efficiency, especially when there are a large number of electronic components on the printed circuit board.

[0024] In an optional embodiment of the present invention, the simplified processing of the three-dimensional circuit board model can specifically simplify the complex geometric shapes of the electronic component physical model in the three-dimensional circuit board model to obtain an electronic component physical model with a relatively simple geometric shape, thereby reducing the computer's processing complexity of the three-dimensional circuit board model and improving processing efficiency.

[0025] Step 102: Input the simplified three-dimensional circuit board model into the trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

[0026] In an embodiment of the present invention, a trained deep learning model (AI model) combined with a graph neural network can be pre-deployed in a computer. The deep learning model is a Graph-Transformer. Graph-Transformer is a deep learning method that combines the architectural advantages of a graph neural network (GNN) and a transformer (a neural network architecture based on an attention mechanism). Compared with traditional models such as CNN (Convolutional Neural Networks), it has significant advantages in processing non-Euclidean data (such as graph structures, temporal relationships, etc.). In other words, based on the deep learning model, the required data can be accurately extracted from the simplified three-dimensional circuit board model.

[0027] It should be noted that the simplified three-dimensional circuit board model has lost a large number of geometric features. Therefore, it is much more difficult to predict the category of electronic components based on the simplified three-dimensional circuit board model. For this reason, in an embodiment of the present invention, the physical model of the electronic components after the simplification of the three-dimensional circuit board is predicted by combining the geometric features and arrangement position information with the deep learning model of the graph neural network. In this way, even if the simplified three-dimensional circuit board model has lost a large number of geometric features, the category of electronic components can still be accurately predicted by combining the arrangement position information with the deep learning model of the graph neural network.

[0028] Specifically, the simplified three-dimensional circuit board model can be input into the deep learning model. The deep learning model can extract geometric features and arrangement position information from the B-Rrep data features (boundary representation data features) of each electronic component physical model in the simplified three-dimensional circuit board model. Then, based on the geometric features and arrangement position information, the classification labels of each electronic component physical model in the simplified three-dimensional circuit board model can be predicted, such as on-board devices, radiators, axial fans, flat fans and other classification labels.

[0029] In a specific implementation, each electronic component in a printed circuit board has its own specific thermal characteristics, including but not limited to parameters such as heat generation, thermal resistance, and heat capacity. These parameters directly affect the heat distribution and heat dissipation effect of the entire printed circuit board. Therefore, in an embodiment of the present invention, when determining the classification label corresponding to each electronic component physical model, the physical parameters required for the simulation of each electronic component physical model can be accurately determined based on the classification label. Then, electronic heat dissipation simulation of the printed circuit board is performed based on the physical parameters of the electronic component physical model, thereby ensuring the accuracy and reliability of the electronic heat dissipation simulation.

[0030] In the above-mentioned electronic component classification prediction method, a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation is obtained, the simplified three-dimensional circuit board model may include an electronic component physical model, and the geometric features of the simplified three-dimensional circuit board model are less than the geometric features of the unsimplified three-dimensional circuit board model. The simplified three-dimensional circuit board model is input into a trained deep learning model combined with a graph neural network, and the deep learning model can output a classification label corresponding to the electronic component physical model in the simplified three-dimensional circuit board model, wherein the deep learning model can extract geometric features and arrangement position information from the boundary representation data features of the electronic component physical model of the simplified three-dimensional circuit board model. In this way, the deep learning model can predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and arrangement position information. When performing electronic heat dissipation simulation on a printed circuit board in an embodiment of the present invention, the deep learning model can accurately predict the geometric features and arrangement position information of the electronic component physical model of the simplified three-dimensional circuit board model, and then determine the classification labels of the electronic components in the printed circuit board based on the geometric features and arrangement position information, thereby improving the recognition accuracy of the electronic components. Furthermore, the physical properties required for the electronic component simulation can be accurately determined based on the identified classification labels, thereby ensuring the accuracy and reliability of the electronic heat dissipation simulation.

[0031] In one embodiment of the present invention, the deep learning model may include an encoder and a neural network model. Step 102, inputting the simplified three-dimensional circuit board model into the trained deep learning model, and obtaining a classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model, includes: The encoder is used to extract geometric features of the electronic component entity model in the simplified three-dimensional circuit board model; wherein the geometric features include geometric information and topological structure; The simplified three-dimensional circuit board model and the geometric features are input into the neural network model to obtain a classification label of the electronic component entity model in the simplified three-dimensional circuit board model output by the neural network model.

[0032] Among them, the geometric features of the electronic component physical model may include geometric information and topological structure. Specifically, geometric information refers to the structural information such as the shape, size, position relationship, etc. of the electronic component physical model. The topological structure refers to the mutual relationship and connection method of the electronic component physical model in the printed circuit board. The topological structure describes the mutual relationship between the geometric elements such as points, lines, and surfaces of the electronic component physical model, as well as the relative storage address in the memory space. These relationships determine how the electronic component physical model is constructed, modified, and rendered in CAD (Computer Aided Design).

[0033] In an embodiment of the present invention, a deep learning model may include an encoder (GNN (Graph Neural Network, graph neural network) - AutoEncoder) of a pre-trained generative model and a neural network model (Transformer). The encoder can be used to predict the geometric features (including geometric information and topological structure) of an electronic component physical model. The neural network model can be used to predict the classification label of the electronic component physical model based on the geometric features and a simplified three-dimensional circuit board model. The neural network model can predict the intimacy matrix between the electronic component physical models during prediction, and can determine whether there is a connection relationship between the electronic component physical models based on the density matrix.

[0034] In one embodiment of the present invention, before step 102, inputting the simplified three-dimensional circuit board model into a trained deep learning model to obtain classification labels of the electronic component physical models in the three-dimensional circuit board model output by the deep learning model, the method may further include: Acquire a sample electronic component entity model of the sample electronic component; wherein the sample electronic component entity model has corresponding actual geometric features, and the actual geometric features include actual geometric information and actual topological structure; Using the sample electronic component physical model to perform self-supervised training on the encoder to be trained; When the encoder after self-supervised training reaches a preset training condition, the trained encoder is obtained.

[0035] In an embodiment of the present invention, an encoder can be used to extract geometric features from a physical model of electronic components on a printed circuit board. The encoder capable of extracting geometric features can be obtained through training. Specifically, a suitable pre-trained generative model (i.e., an encoder to be trained, such as one based on a diffusion model architecture, specifically a GNN-AutoEncoder model structure) can be selected. The encoder of the pre-trained generative model can extract geometric features from the physical model of the electronic components.

[0036] In one example of the present invention, self-supervised pre-training is applied to obtain B-Rrep data of individual electronic component 3D models (sample electronic component physical models) from a large number of open-source CAD files. The B-Rrep data defines the boundaries between internal and external points by representing the physical model as a set of connected surface elements. The B-Rrep data includes actual geometric features. The generated encoder to be trained is then self-supervised pre-trained based on the sample electronic component physical models. During the training process, the encoder learns the geometric information and topological structure of the electronic component physical models, thereby improving the effectiveness of the encoder's pre-extraction of geometric features, enabling the encoder to accurately extract key features of the electronic component physical models and obtain feature vectors, which are used as the geometric features of the corresponding electronic components. The encoder can be considered a trained encoder when it meets preset training conditions, such as when the number of training iterations reaches a preset number and the error between the geometric features output by the encoder and the actual geometric features is within a preset error range.

[0037] The present invention obtains the CAD geometric features of different electronic component physical models in the CAD file as the input of the encoder, directly trains and learns the geometric features of each electronic component physical model, fully utilizes the geometric features of the electronic components, and effectively improves the accuracy of geometric feature recognition of electronic components.

[0038] In one embodiment of the present invention, before step 102, inputting the simplified three-dimensional circuit board model into a trained deep learning model to obtain classification labels of the electronic component physical models in the three-dimensional circuit board model output by the deep learning model, the method may further include: Obtaining a sample three-dimensional circuit board model of a sample printed circuit board; wherein the sample three-dimensional circuit board model includes a sample electronic component entity model, and the sample electronic component entity model has a corresponding actual classification label; Calculating the center point distance between the center points of the sample electronic component physical models; Calculating the intimacy matrix between the sample electronic component physical models according to the center point distance; Determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component physical models and a preset threshold; Converting the sample electronic component physical model into a coding unit; For each coding unit of the sample printed circuit board, construct subgraph data from geometric feature vectors of the coding unit and other coding units that have a connection relationship in reverse order of the center point distance of the coding unit, and combine all the subgraph data of the sample printed circuit board into an input sequence; The input sequence is used to pre-train the neural network model to be trained.

[0039] In an embodiment of the present invention, an input sequence is constructed according to the distribution of electronic component physical models on a printed circuit board. Specifically, based on the distribution position information of the electronic component physical models on the printed circuit board, the center point distances between the electronic component physical models are calculated. The intimacy matrix between the electronic component physical models is obtained by calculating the center point distances. Based on the intimacy matrix and a preset threshold, it is determined whether there is a connection relationship between the electronic component physical models. If there is a connection relationship, it indicates that there are edges between the electronic component physical models. Sub-graph data is constructed. Then, based on the sub-graph data, each electronic component physical model is converted into a coding unit (token) and pre-training of the Transformer (neural network model) is performed. Specifically, each electronic component physical model is used as a starting token, and the electronic component physical model is compared with other electronic components according to the intimacy matrix and the preset threshold. Whether there is a connection relationship between the physical models. If there is a connection relationship, other electronic component physical models are used as tokens to construct sub-graph data in reverse order of center point distance, and then the sub-graph data of all electronic component physical models constitute the input sequence of Transformer, wherein the input sequence is composed of several geometric feature vectors, each node in the sequence represents the geometric feature of an electronic component, and the geometric feature is represented by a geometric feature vector. The input sequence reflects the arrangement position information of the electronic component physical model. Then, the neural network model to be trained with the input sequence can be used for training, so as to obtain the ability to predict the center point distance between the electronic component physical models, predict the intimacy matrix based on the center point distance, and then predict whether there is a connection relationship between the electronic component physical models based on the intimacy matrix and the preset threshold.

[0040] In one embodiment of the present invention, determining the connection relationship between the sample electronic components based on the intimacy matrix between the sample electronic component physical models and a preset threshold may include: When the intimacy matrix between the sample electronic component physical models is greater than the preset threshold, it is determined that a connection relationship exists between the sample electronic component physical models.

[0041] In an embodiment of the present invention, if the intimacy matrix between the sample electronic component physical models is greater than a preset threshold, it can be determined that there is a connection relationship between the sample electronic component physical models. Conversely, if the intimacy matrix between the sample electronic component physical models is less than or equal to the preset threshold, it can be determined that there is no connection relationship between the sample electronic component physical models.

[0042] In a specific implementation, the center point distance and intimacy are inversely related. The smaller the center point distance between electronic components, the greater the intimacy in the intimacy matrix. Conversely, the greater the center point distance between electronic components, the smaller the intimacy in the intimacy matrix. For example, the intimacy value in the intimacy matrix is compared with a preset threshold. If it is less than the preset threshold, it is set to 0, and if it is greater than or equal to the preset threshold, it is set to 1. The intimacy matrix with only 0 and 1 is obtained as the adjacency matrix. 1 indicates that there is a connection relationship between the sample electronic component physical models, and 0 indicates that there is no connection relationship between the sample electronic component physical models. In this way, the connection relationship can be described based on the adjacency matrix.

[0043] In some embodiments of the present invention, the calculation process of the intimacy matrix may include:

[0044] in, is the coordinate of the center point of node i, is the coordinate of the center point of node j, is the center-point distance between node i and node j.

[0045] The intimacy value (numeric value) in the intimacy matrix can be calculated as follows:

[0046] in, Represents the intimacy value in the intimacy matrix. Generally, the greater the distance between the center points, the lower the probability of a connection between the nodes (electronic components). The calculated intimacy values can then be combined to form an intimacy matrix.

[0047] In specific implementations, the connection relationships of electronic components in a three-dimensional circuit board model are usually determined based on electrical connections (such as wires and solder joints). However, in actual applications, non-electrical connection relationships may exist between electronic components (such as thermal coupling and physical proximity). These connection relationships are not clearly shown in the three-dimensional circuit board model, especially in a simplified three-dimensional circuit board model, which may also lose some electrical connections.

[0048] In an embodiment of the present invention, different types of connection relationships are required for different simulation tasks. For example, in a thermal simulation task, additional thermal coupling relationship information of electronic components is required, and this information may be lost in the simplified three-dimensional circuit board model. Therefore, an embodiment of the present invention proposes that whether there is a connection relationship between electronic components can be determined based on the center point distance. In particular, when the simplified three-dimensional circuit board model fails to show non-electrical connection relationships (such as thermal coupling, physical proximity), the center point distance can be used as a supplementary means to ensure that important connection relationships in the three-dimensional circuit board model are not ignored. For example, if two components are not directly connected electrically, but because the center point distance is close, there is a high probability that they will show a strong thermal coupling relationship in the electronic heat dissipation simulation. In addition, the embodiment of the present invention can also meet the simulation requirements of multiple simulation tasks by adjusting the preset threshold according to different simulation task requirements (such as different simulation task types, simulation accuracy requirements, simulation speed, electronic component types in the three-dimensional circuit board model, working environment, etc.).

[0049] For example, in high-precision thermal simulation, the preset threshold is lowered to capture more thermal coupling relationships; in fast electrical simulation, the preset threshold is increased to reduce the number of connection relationships and increase the simulation speed; for high-power components or high-temperature environments, the preset threshold is further lowered to enhance the accuracy of thermal management. In this way, the embodiments of the present invention can adapt to diverse simulation needs by flexibly adjusting the threshold, ensuring the accuracy and efficiency of the simulation results.

[0050] In one embodiment of the present invention, pre-training the neural network model to be trained using the input sequence may include: Randomly performing masking on the coding units in the input sequence to obtain a masked input sequence; Inputting the masked input sequence into the neural network model to be trained for training, so that the neural network model predicts the predicted classification label of the masked coding unit based on the unmasked coding unit in the input sequence; Calculate the loss value according to the predicted classification label and the actual classification label corresponding to the masked encoding unit; When the loss value reaches a preset convergence condition, the pre-trained neural network model is obtained.

[0051] In an embodiment of the present invention, during the training process of the neural network model, the Masking prediction technology is used for model training. Specifically, by randomly masking (masking) part of the coding unit (token) of the input sequence, the neural network model needs to predict the predicted classification label corresponding to the masked data based on the unmasked part (unmasked), that is, the predicted classification label of the coding unit after masking, and then calculate the loss value based on the predicted classification label corresponding to the coding unit after masking and the actual classification label. If the loss value reaches the preset convergence condition, for example, when the loss value is within the preset loss value range, a trained neural network model can be obtained. In addition, the adjacency matrix (connection relationship) is also combined during the training of the neural network model, so that the neural network model learns a more representative and universal feature representation.

[0052] In one embodiment of the present invention, the neural network model includes a multi-layer perceptual network, which is used to classify according to the feature vector corresponding to the first coding unit in the input sequence output by the neural network model to output a predicted classification label corresponding to the first coding unit.

[0053] Among them, Transformer (neural network model) can also include MLP (Multilayer Perceptron) network. MLP network is a feedforward neural network composed of multiple layers of neurons and can learn complex nonlinear mapping relationships.

[0054] In an embodiment of the present invention, after the input sequence passes through the pre-trained transformer model, the feature vector of the first token can be obtained, and the MLP is applied to perform classification fine-tuning training on the feature vector to obtain the final classification and recognition result of the electronic component physical model, that is, the classification label.

[0055] Among them, B-Rrep data is boundary representation data, a geometric shape representation method widely used in solid modeling and computer-aided design (CAD). B-Rrep data defines the boundaries between internal and external points through connected surface elements, and records in detail the geometric information of all geometric elements that constitute the shape and their interconnected topological structure. Therefore, B-Rrep data is suitable for representing objects with complex geometric shapes and topological structures, such as electronic components in PCB boards. Based on the geometric information and topological structure recorded in B-Rrep data, the performance and behavior of electronic components in actual PCB boards can be accurately simulated. The embodiment of the present invention extracts B-Rrep data of each electronic component solid model in the PCB board, constructs subgraph data based on the spatial distribution of the components (arrangement position information) of the electronic component solid model, and predicts the classification labels of the electronic components based on the AI model (deep learning model) of the GNN and Transformer architecture, thereby achieving the classification of electronic components.

[0056] To help those skilled in the art better understand the embodiments of the present invention, a specific example is used below to illustrate the process of predicting electronic components in a printed circuit board based on a deep learning model. The specific steps of the embodiment of the present invention may include: Step 1: Use the pre-trained generative model encoder to pre-extract features: Select an appropriate pre-trained generative model (e.g., one based on a diffusion model, such as the GNN-AutoEncoder model structure) whose encoder can extract geometric features from PCB 3D models (physical models of electronic components / 3D models of electronic components). In practical applications, a large amount of open-source B-Rrep data of independent 3D electronic component models is collected and fed into the pre-trained generative model for self-supervised pre-training. During self-supervised pre-training, the B-Rrep reconstruction task (Reconstruct) is used to enhance the encoder's ability to extract geometric information and topological structures of electronic components, thereby improving the effectiveness of the encoder's pre-extraction of geometric features. Specifically, Input B-Rrep: B-Rrep data of the 3D model of the electronic component; GNN tokenizer: converts the B-Rrep data of each component on the PCB board into graph structure data according to the connection relationship, regards each component as a tokenizer of the transformer, and then arranges them in reverse order according to the distance between the components to form a sequence as the input of the transformer; Surface_Conv and Curve_Conv: perform UV coordinate grid sampling on the surfaces and edges of the B-Rrep data to obtain point coordinates, and perform convolution operations on the point coordinates; Node Update: Update the node information in the B-Rrep data according to the message passing mechanism of the graph neural network; Edge Update: Update the edge information in the B-Rrep data according to the message passing mechanism of the graph neural network; GNN: Graph neural network in deep learning; Reconstruct: Use the learned node or graph representation to reconstruct or reconstruct the B-Rrep data.

[0057] In a preferred embodiment of the present invention, the pre-trained generative model is a Graph-Transformer deep learning model. Graph-Transformer is a deep learning method that combines the advantages of graph neural networks and Transformer architectures. Specifically, it aggregates neighbor information through a message passing mechanism (such as GNN) to preserve the topological structure of electronic components. In addition, the self-attention of Transformer can capture the dependency relationship between any two electronic components in the graph. Therefore, in complex relationship prediction tasks, especially in scenarios where the data itself has an explicit associative structure, such as a simplified three-dimensional circuit board model with a large number of electronic components, the geometric information and topological structure can be obtained more accurately than models such as CNN.

[0058] Step 2: Sub-image data construction of PCB electronic components: Taking each electronic component on the PCB board as the center, the center point distance to other electronic components is calculated respectively. The intimacy matrix between the electronic components is obtained through the center point distance calculation. According to the intimacy matrix and the preset threshold, it is judged whether there is a connection relationship between the electronic components. The eigenvectors of the electronic components (nodes) are represented by geometric eigenvectors. If there is a connection relationship, there are edges between the nodes, which are constructed into a graph structure, thereby reflecting the geometric characteristics and position information of the electronic components.

[0059] Step 3: Transformer pre-training: Each electronic component is used as a starting token. Based on the existence of a connection (connectivity is determined based on the adjacency matrix), other electronic components are used as tokens to construct the Transformer input sequence in reverse order of center-point distance. The input sequence consists of multiple subgraphs, each representing a feature vector sequence consisting of a target component entity / electronic component entity model (the first token) and neighboring component entity models / (surrounding component entity models) (other tokens). The input sequence is then concatenated with positional information as the Transformer input representation. This input representation considers both the structural and topological characteristics of each electronic component entity model and their spatial distribution on the PCB, helping the Transformer model better understand the relationships between different electronic components during processing. Masking prediction technology is employed during Transformer model training. Masking prediction technology randomly masks part of the input representation. The Transformer model needs to predict the masked data (masked tokens) based on the unmasked part (unmasked tokens). In addition, the Transformer can also predict the adjacency matrix of electronic components, allowing the Transformer model to learn more representative and universal feature representations.

[0060] Step 4: Transformer fine-tuning classification training: Fine-tune the Transformer model for classification training to identify electronic component classification labels. The first token in the Transformer's input sequence represents an electronic component. Based on the classification label of each electronic component, an MLP network is added to the Transformer output. The MLP network can fine-tune the classification training on the first token in each subgraph data in the input sequence to obtain the final electronic component recognition result (classification label).

[0061] In summary, the embodiments of the present invention are based on graph neural networks and the Transformer self-attention mechanism to capture the geometric features and spatial distribution information (position arrangement information) of each component entity, thereby solving the difficulty of identifying electronic component assemblies. Furthermore, the physical properties required for simulation can be assigned based on the classification labels of the identified electronic components, thereby efficiently and accurately realizing electronic heat dissipation simulation of printed circuit boards.

[0062] It should be noted that for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required for the embodiments of the present invention.

[0063] Reference Figure 2 , is a structural block diagram of an electronic component classification prediction device provided in an embodiment of the present invention, such as Figure 2 As shown, the device 20 may specifically include the following modules: The three-dimensional circuit board model acquisition module 201 is used to obtain a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes a physical model of electronic components; The classification label acquisition module 202 is used to input the simplified three-dimensional circuit board model into the trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

[0064] In one embodiment of the present invention, the classification label acquisition module 202 is configured to: The encoder is used to extract geometric features of the electronic component entity model in the simplified three-dimensional circuit board model; wherein the geometric features include geometric information and topological structure; The simplified three-dimensional circuit board model and the geometric features are input into the neural network model to obtain a classification label of the electronic component entity model in the simplified three-dimensional circuit board model output by the neural network model.

[0065] In one embodiment of the present invention, the apparatus further comprises: a model training module, configured to: Acquire a sample electronic component entity model of the sample electronic component; wherein the sample electronic component entity model has corresponding actual geometric features, and the actual geometric features include actual geometric information and actual topological structure; Using the sample electronic component physical model to perform self-supervised training on the encoder to be trained; When the encoder after self-supervised training reaches a preset training condition, the trained encoder is obtained.

[0066] In one embodiment of the present invention, the apparatus further comprises: a model training module, configured to: Obtaining a sample three-dimensional circuit board model of a sample printed circuit board; wherein the sample three-dimensional circuit board model includes a sample electronic component entity model, and the sample electronic component entity model has a corresponding actual classification label; Calculating the center point distance between the center points of the sample electronic component physical models; Calculating the intimacy matrix between the sample electronic component physical models according to the center point distance; Determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component physical models and a preset threshold; Converting the sample electronic component physical model into a coding unit; For each coding unit of the sample printed circuit board, construct subgraph data from geometric feature vectors corresponding to the coding unit and other coding units that have a connection relationship in reverse order of the center point distance of the coding unit, and combine all the subgraph data of the sample printed circuit board into an input sequence; The input sequence is used to pre-train the neural network model to be trained.

[0067] In one embodiment of the present invention, the model training module is used to: When the intimacy matrix between the sample electronic component physical models is greater than the preset threshold, it is determined that a connection relationship exists between the sample electronic component physical models.

[0068] In one embodiment of the present invention, the model training module is used to: Randomly performing masking on the coding units in the input sequence to obtain a masked input sequence; Inputting the masked input sequence into the neural network model to be trained for training, so that the neural network model predicts the predicted classification label of the masked coding unit based on the unmasked coding unit in the input sequence; Calculate the loss value according to the predicted classification label and the actual classification label corresponding to the masked encoding unit; When the loss value reaches a preset convergence condition, the pre-trained neural network model is obtained.

[0069] In one embodiment of the present invention, the neural network model includes a multi-layer perceptual network, which is used to classify according to the feature vector corresponding to the first coding unit in the input sequence output by the neural network model, so as to output the predicted classification label corresponding to the first coding unit to perform classification fine-tuning training on the pre-trained neural network model to obtain a trained deep learning model.

[0070] In an embodiment of the present invention, a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation is obtained, the simplified three-dimensional circuit board model may include an electronic component physical model, and the geometric features of the simplified three-dimensional circuit board model are less than the geometric features of the unsimplified three-dimensional circuit board model. The simplified three-dimensional circuit board model is input into a trained deep learning model combined with a graph neural network, and the deep learning model can output classification labels corresponding to the electronic component physical models in the simplified three-dimensional circuit board model, wherein the deep learning model can extract geometric features and arrangement position information from the boundary representation data features of the electronic component physical models of the simplified three-dimensional circuit board model. In this way, the deep learning model can predict the classification labels of the electronic component physical models in the simplified three-dimensional circuit board model based on the geometric features and arrangement position information. When performing electronic heat dissipation simulation on a printed circuit board in an embodiment of the present invention, the deep learning model can accurately predict the geometric features and arrangement position information of the electronic component physical model of the simplified three-dimensional circuit board model, and then determine the classification labels of the electronic components in the printed circuit board based on the geometric features and arrangement position information, thereby improving the recognition accuracy of the electronic components. Furthermore, the physical properties required for the electronic component simulation can be accurately determined based on the identified classification labels, thereby ensuring the accuracy and reliability of the electronic heat dissipation simulation.

[0071] As for the above-mentioned device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0072] The embodiment of the present invention further provides an electronic device, such as Figure 3As shown, it includes a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304. Memory 303, for storing computer programs; The processor 301 is configured to implement the electronic component classification prediction method described in any one of the above embodiments when executing the program stored in the memory 303, specifically including: Obtaining a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes a physical model of electronic components; The simplified three-dimensional circuit board model is input into a trained deep learning model to obtain a classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information.

[0073] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0074] The communication interface is used for communication between the above terminal and other devices.

[0075] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0076] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0077] In another embodiment provided by the present invention, a computer-readable storage medium is also provided, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the electronic component classification prediction method described in any of the above embodiments.

[0078] In another embodiment provided by the present invention, a computer program product including instructions is further provided, which, when executed on a computer, enables the computer to execute the electronic component classification prediction method described in any one of the above embodiments.

[0079] In the above embodiments, all or part of the embodiments can be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive (SSD)).

[0080] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so 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. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0081] Each embodiment in this specification is described in a related manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are described briefly because they are generally similar to the method embodiments. For related portions, reference can be made to the description of the method embodiments.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the scope of protection of the present invention.

Claims

1. A classification prediction method for electronic components, characterized in that: The method comprises: Obtaining a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes solid models of electronic components, and the geometric features of the simplified three-dimensional circuit board model are fewer than the geometric features of the unsimplified three-dimensional circuit board model; The simplified three-dimensional circuit board model is input into a trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

2. The method according to claim 1, characterized in that The deep learning model includes an encoder and a neural network model. The simplified three-dimensional circuit board model is input into the trained deep learning model to obtain the classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, including: The encoder is used to extract geometric features of the electronic component entity model in the simplified three-dimensional circuit board model; wherein the geometric features include geometric information and topological structure; The simplified three-dimensional circuit board model and the geometric features are input into the neural network model to obtain a classification label of the electronic component entity model in the simplified three-dimensional circuit board model output by the neural network model.

3. The method according to claim 2, characterized in that Before inputting the simplified three-dimensional circuit board model into the trained deep learning model to obtain the classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, the method further includes: Acquire a sample electronic component entity model of the sample electronic component; wherein the sample electronic component entity model has corresponding actual geometric features, and the actual geometric features include actual geometric information and actual topological structure; Using the sample electronic component physical model to perform self-supervised training on the encoder to be trained; When the encoder after self-supervised training reaches a preset training condition, the trained encoder is obtained.

4. The method according to claim 3, characterized in that Before inputting the simplified three-dimensional circuit board model into the trained deep learning model to obtain the classification label of the electronic component entity model in the three-dimensional circuit board model output by the deep learning model, the method further includes: Obtaining a sample three-dimensional circuit board model of a sample printed circuit board; wherein the sample three-dimensional circuit board model includes a sample electronic component entity model, and the sample electronic component entity model has a corresponding actual classification label; Calculating the center point distance between the center points of the sample electronic component physical models; Calculating the intimacy matrix between the sample electronic component physical models according to the center point distance; Determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component physical models and a preset threshold; Converting the sample electronic component physical model into a coding unit; For each coding unit of the sample printed circuit board, construct subgraph data from geometric feature vectors corresponding to the coding unit and other coding units that have a connection relationship in reverse order of the center point distance of the coding unit, and combine all the subgraph data of the sample printed circuit board into an input sequence; The input sequence is used to pre-train the neural network model to be trained.

5. The method according to claim 4, characterized in that Determining the connection relationship between the sample electronic components according to the intimacy matrix between the sample electronic component entity models and a preset threshold value includes: When the intimacy matrix between the sample electronic component physical models is greater than the preset threshold, it is determined that a connection relationship exists between the sample electronic component physical models.

6. The method according to claim 5, characterized in that Pre-training the neural network model to be trained using the input sequence includes: Randomly performing masking on the coding units in the input sequence to obtain a masked input sequence; Inputting the masked input sequence into the neural network model to be trained for training, so that the neural network model predicts the predicted classification label of the masked coding unit based on the unmasked coding unit in the input sequence; Calculate the loss value according to the predicted classification label and the actual classification label corresponding to the masked encoding unit; When the loss value reaches a preset convergence condition, the pre-trained neural network model is obtained.

7. The method according to claim 6, characterized in that The neural network model includes a multi-layer perception network, which is used to classify according to the feature vector corresponding to the first coding unit in the input sequence output by the neural network model, so as to output the predicted classification label corresponding to the first coding unit to perform classification fine-tuning training on the pre-trained neural network model to obtain a trained deep learning model.

8. An electronic component classification prediction device, characterized in that: The device comprises: A three-dimensional circuit board model acquisition module is used to obtain a simplified three-dimensional circuit board model of a printed circuit board to be subjected to electronic heat dissipation simulation; wherein the simplified three-dimensional circuit board model includes solid models of electronic components; and the geometric features of the simplified three-dimensional circuit board model are fewer than the geometric features of the unsimplified three-dimensional circuit board model; A classification label acquisition module is used to input the simplified three-dimensional circuit board model into a trained deep learning model to obtain the classification label of the electronic component physical model in the three-dimensional circuit board model output by the deep learning model; the deep learning model extracts geometric features and arrangement position information from the boundary representation data features of the electronic component physical model in the simplified three-dimensional circuit board model to predict the classification label of the electronic component physical model in the simplified three-dimensional circuit board model based on the geometric features and the arrangement position information; the deep learning model is a deep learning model combined with a graph neural network.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; A processor, configured to implement the method according to any one of claims 1 to 7 when executing a program stored in a memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images