Performance prediction method and device of equipment
The integration of point cloud data with a deep learning network, particularly PointNet, addresses the inefficiencies of CFD simulation by enabling fast and accurate equipment performance prediction, enhancing design efficiency and reducing resource requirements.
Patent Information
- Application Number
- CN202411267771.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-31
AI Technical Summary
Traditional equipment performance prediction methods, such as computational fluid dynamics (CFD) simulation, are resource-intensive and time-consuming, making it difficult to meet the needs of fast-iteration product design.
A performance prediction method combining point cloud data with a deep learning network, specifically using PointNet, to simulate and analyze equipment performance, allowing for quick and efficient predictions of parameters affecting noise and aerodynamic performance.
Enables rapid and accurate prediction of equipment performance, reducing the need for extensive computing resources and expert knowledge, thereby accelerating product design iterations and improving market response speed.
Smart Images
Figure CN119227525A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of equipment performance prediction, and in particular to an equipment performance prediction method and an equipment performance prediction device. Background Art
[0002] In modern industrial and consumer electronic products, the performance of equipment such as cooling fans is directly related to the cooling efficiency and noise level of the equipment. The performance of cooling equipment, such as aerodynamic performance and noise performance, is an important indicator of product quality. Traditional equipment performance prediction methods mainly rely on experimental testing and computational fluid dynamics (CFD) simulation. CFD, as a numerical simulation method, can provide relatively accurate performance predictions. However, CFD simulation requires a lot of computing resources and expertise, and the simulation process is time-consuming, with low prediction efficiency, making it difficult to meet the needs of fast-iteration product design. Summary of the invention
[0003] In order to solve one of the above technical problems, the present invention proposes the following technical solution.
[0004] The first aspect of the present invention proposes a performance prediction method for a device, comprising the following steps: obtaining point cloud data of the device; performing simulation analysis on the performance of the device based on the point cloud data to obtain device performance data, wherein, when performing the simulation analysis, parameters affecting performance prediction are determined and set; constructing a deep learning network whose input layer is an n×(3+k) matrix, wherein n is the number of point cloud data, and k is the number of types of parameters affecting performance prediction; training the deep learning network using the point cloud data and the parameters affecting performance prediction as input data, and the device performance data as output data; obtaining actual point cloud data of the device to be predicted and the actual parameters affecting performance prediction, and inputting them into the trained deep learning network for performance prediction.
[0005] In addition, the performance prediction method of the device according to the above embodiment of the present invention may also have the following additional technical features.
[0006] According to one embodiment of the present invention, the performance of the equipment is simulated and analyzed based on the point cloud data to obtain equipment performance data, including: constructing a three-dimensional model of the equipment according to the point cloud data; using CFD to construct a simulation model of the three-dimensional model, including the construction of a calculation domain, a grid, and a required simulation result output type; determining parameters that affect performance prediction, and setting parameters of the simulation model according to the parameters that affect performance prediction; and performing performance simulation analysis on the set simulation model to obtain equipment performance data.
[0007] According to one embodiment of the present invention, the parameter affecting the performance prediction includes at least one of a parameter affecting the equipment noise performance prediction and a parameter affecting the equipment aerodynamic performance prediction.
[0008] According to one embodiment of the present invention, the equipment performance data includes at least one of noise performance data and aerodynamic performance data.
[0009] According to one embodiment of the present invention, the parameter affecting the performance prediction includes the rotation speed of the device, the device performance data includes the sound pressure level spectrum data of the device, and the input layer of the deep learning network is an n×4 matrix.
[0010] According to one embodiment of the present invention, the parameters affecting performance prediction include the rotation speed and incoming flow velocity of the equipment, the equipment performance data include the wind volume data and wind pressure data of the equipment, and the input layer of the deep learning network is an n×5 matrix.
[0011] According to one embodiment of the present invention, the deep learning network is a PointNet network.
[0012] According to an embodiment of the present invention, the device is a heat dissipation device.
[0013] The second aspect of the present invention proposes a performance prediction device for an equipment, comprising: an acquisition module, used to acquire point cloud data of the equipment; a simulation module, used to perform simulation analysis on the performance of the equipment based on the point cloud data to obtain equipment performance data, wherein, when performing the simulation analysis, parameters affecting the performance prediction are determined and set; a construction module, used to construct a deep learning network whose input layer is an n×(3+k) matrix, wherein n is the number of point cloud data, and k is the number of types of parameters affecting the performance prediction; a training module, used to train the deep learning network using the point cloud data and the parameters affecting the performance prediction as input data, and the equipment performance data as output data; a prediction module, used to obtain the actual point cloud data of the equipment to be predicted and the actual parameters affecting the performance prediction, and input them into the trained deep learning network for performance prediction.
[0014] The technical solution of the embodiment of the present invention combines performance simulation analysis and deep learning network to achieve performance prediction, and can predict the performance of different devices quickly and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 The present invention is a flowchart of a method for predicting performance of a device according to an embodiment of the present invention.
[0016] Figure 2 Schematic diagram of the PointNet network input layer according to an embodiment of the present invention.
[0017] Figure 3 A schematic diagram showing the principle of fan noise performance prediction according to an example of the present invention.
[0018] Figure 4 A schematic diagram showing the principle of fan aerodynamic performance prediction according to an example of the present invention.
[0019] Figure 5 It is a block diagram of a device for predicting performance of a device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0021] Figure 1 The present invention is a flowchart of a method for predicting performance of a device according to an embodiment of the present invention.
[0022] like Figure 1 As shown, the performance prediction method of the device includes the following steps S1 to S5.
[0023] The performance prediction method of the device in the embodiment of the present invention can be applied to any device, such as heat dissipation devices such as fans, and can predict the aerodynamic performance, noise performance and other performance of the fan.
[0024] S1, obtain the point cloud data of the device.
[0025] The device may be a heat dissipation device, and the heat dissipation device may be a fan.
[0026] Specifically, high-precision 3D scanning technology can be used to obtain point cloud data (sufficient quantity) of heat dissipation equipment (equipment samples) to ensure that the data can accurately reflect the geometric shape and surface characteristics of the heat dissipation equipment. Specifically, taking a fan as an example, high-precision 3D scanning technology can be used to obtain point cloud data of fan blades.
[0027] S2, performing simulation analysis on the performance of the device based on the point cloud data to obtain device performance data, wherein when performing the simulation analysis, parameters affecting performance prediction are determined and set.
[0028] The equipment performance data refers to the data representing the equipment performance obtained through performance simulation analysis.
[0029] The performance of the equipment may include at least one of noise performance and aerodynamic performance. The parameters affecting performance prediction include at least one of parameters affecting equipment noise performance prediction and parameters affecting equipment aerodynamic performance prediction. Equipment performance data include at least one of noise performance data and aerodynamic performance data.
[0030] The parameters include parameters that affect the prediction of noise performance (such as the rotation speed of the heat dissipation device). Accordingly, the equipment performance data include data representing the noise performance of the equipment, namely, noise performance data (such as the sound pressure level spectrum of the equipment); the parameters include parameters that affect the prediction of aerodynamic performance (such as the rotation speed and incoming flow velocity of the heat dissipation device). Accordingly, the equipment performance data include data representing the aerodynamic performance of the equipment, namely, aerodynamic performance data (such as the air volume data and air pressure data of the equipment).
[0031] Specifically, CFD can be used to simulate and analyze the performance of equipment such as heat dissipation equipment. First, the simulation model is constructed and the parameters are set, and then simulation analysis is performed to obtain the performance data of the heat dissipation equipment as subsequent network training samples. The parameters used for parameter setting refer to specific values that affect the performance prediction of the heat dissipation equipment (such as noise performance, aerodynamic performance, etc.), and the type of parameters can be determined according to specific performance requirements.
[0032] S3, construct a deep learning network with an input layer of n×(3+k) matrix, where n is the number of point cloud data and k is the number of types of parameters that affect performance prediction.
[0033] The deep learning network may be a PointNet network. PointNet is a deep learning architecture for processing point cloud data. Compared with traditional methods, PointNet can directly learn global features from point cloud data without relying on complex point cloud preprocessing or feature engineering steps. This method not only simplifies the data processing process, but also greatly improves the efficiency and accuracy of point cloud processing technology, and promotes the development of the entire field.
[0034] Specifically, a deep learning network based on PointNet can be constructed. Figure 3 As shown, first, an n×3 (n-row and 3-column matrix, n is the number of point clouds) network input layer is constructed according to the most basic number of point clouds and the spatial coordinates (x, y, z) of each point. Then, according to the number k of types of parameters affecting fan performance prediction determined and set in step S2, k column vectors are added to the n×3 matrix for inputting the parameters as additional input items to realize the combination of parameters and point cloud data. Based on this, an n×(3+k) network input layer is constructed, where k is the number of types of additional parameter items. For example, when the number of parameter types is one, k is equal to 1, and when the number of parameter types is two, k is equal to 2.
[0035] S4, uses point cloud data and parameters affecting performance prediction as input data and equipment performance data as output data to train the deep learning network.
[0036] Specifically, the point cloud data and the parameters affecting the performance prediction are used as input data and input into the n×(3+k) input layer of the PointNet deep learning network. The device performance data simulated in step S2 is used as output data and input into the deep learning network. The deep learning network is trained and the model is optimized based on a large number of sample data.
[0037] Reference Figure 2 ,The PointNet deep learning network contains multiple layers, including local feature extraction and recognition layer, ,feature conversion layer, feature aggregation layer, global feature learning layer and model ,training optimization layer. The deep learning network constructed and trained ,through these layers can accurately predict the required data.
[0038] S5, obtaining the actual point cloud data of the device to be predicted and the actual parameters that affect the performance prediction, and inputting them into the trained deep learning network for performance prediction.
[0039] Specifically, for devices to be predicted, such as fans, high-precision three-dimensional scanning technology can be used to obtain the actual point cloud data of fan blades and the actual parameters that affect the performance prediction of the fan, such as the actual parameters that affect the aerodynamic or noise performance. The actual point cloud data and actual parameters are input into the trained deep learning network for noise or aerodynamic performance prediction to output the required noise or aerodynamic performance prediction data.
[0040] Compared with the related art, the embodiment of the present invention first performs performance simulation analysis, builds and trains a deep learning network on this basis, and realizes performance prediction based on the deep learning network. It does not require a large amount of computing resources and corresponding professional knowledge, is fast and efficient, and can meet the needs of fast iteration of product design. In addition, the parameters that affect the predicted performance are used as additional input items and combined with point cloud data for network training, which improves the network's ability to fully capture the factors affecting fan performance and can improve the prediction accuracy.
[0041] Applying the PointNet network to performance prediction can make full use of its powerful modeling capabilities to quickly and accurately predict the aerodynamic performance and noise characteristics of the equipment. This prediction can not only help manufacturers evaluate the performance of different design schemes during the design phase, but also provide efficient support for the evaluation of the heat dissipation and noise performance of the whole machine. Through the rapid iteration and optimization of the deep learning network, the process of equipment design and the evaluation of the heat dissipation and noise performance of the whole machine can be significantly accelerated, thereby shortening the product development cycle and improving market response speed.
[0042] Therefore, the performance prediction method of the device in the embodiment of the present invention combines performance simulation analysis and deep learning network to achieve performance prediction, and can predict the performance of different devices quickly and efficiently.
[0043] In one embodiment of the present invention, step S2 is to simulate and analyze the performance of the equipment based on the point cloud data to obtain the equipment performance data, which may include: constructing a three-dimensional model of the equipment according to the point cloud data; using CFD to construct a simulation model of the three-dimensional model, including the construction of a calculation domain, a grid, and a desired simulation result output type; determining parameters that affect performance prediction, and setting parameters of the simulation model according to the parameters that affect performance prediction; and performing performance simulation analysis on the set simulation model to obtain the equipment performance data.
[0044] Among them, the parameters affecting the performance prediction are multiple different values set in advance, which can be set according to actual conditions.
[0045] Specifically, taking the fan as an example, first, the fan blade point cloud data obtained by scanning in step S1 is constructed into a three-dimensional (mathematical) model through reverse modeling technology, and the three-dimensional model is simulated by CFD, including the construction of the calculation domain, grid and the required simulation result output type, that is, the three-dimensional model is divided into the calculation domain and grid, and the required simulation result output type is set to obtain the simulation model, and the simulation model is parameterized, which is a parameter that affects the prediction of fan performance, including but not limited to the fan speed, a parameter that affects the prediction of fan aerodynamic noise, and the incoming flow velocity, a parameter that affects the prediction of fan aerodynamic performance. Then, the constructed and set simulation model is simulated and analyzed (such as a simulated wind tunnel test) to obtain the fan performance data under different parameters, including but not limited to the velocity field, pressure field and far-field sound field of the calculated fluid. Among them, the RANS (Reynolds averaged NS equations) physical model in CFD can be used to solve the fan aerodynamic performance, and the DES (Detached eddy simulation) coupled FW-H physical model (FfowcsWilliams-Hawkings Model) or CFD coupled CAA (Computational aeroacoustics) method can be used to solve the fan far-field noise performance. The wind volume data and wind pressure data representing the fan aerodynamic performance, as well as the far-field measurement point sound pressure level spectrum data representing the fan noise characteristics are obtained through the solution, and the obtained results are used as output data samples for subsequent deep learning network training.
[0046] It should be noted that the training of the deep learning network requires a certain number of output data samples. m output data samples are constructed for different output data. For example, for the output fan far-field measurement point sound pressure level spectrum data, the spectrum data to be output is a data set including the sound pressure level amplitude corresponding to each 5Hz interval from 100Hz to 4000Hz, which contains 780 data points. For the accuracy of subsequent deep learning network training, it is necessary to appropriately increase the number of output data samples m, and m must be greater than the number of output data points.
[0047] Therefore, the aerodynamic performance and sound pressure level spectrum data calculated by CFD are used as training samples to improve the accuracy and reliability of deep learning network training.
[0048] In one example, the parameters affecting performance prediction include the rotation speed of the device, the device performance data includes the sound pressure level spectrum data of the device, and the input layer of the deep learning network is an n×4 matrix.
[0049] For fan noise prediction, fan noise data needs to be output. Fan speed is the key parameter that affects the prediction of fan noise. The fan speed RPM is merged into the basic n×3 matrix as a single column matrix to form an n×4 network input layer.
[0050] Specifically, taking the prediction of the noise performance of a fan as an example, Figure 3 As shown in the figure, after obtaining the point cloud data, the three-dimensional model and the simulation model (such as the wind tunnel simulation model) in turn, the simulation model is parameterized, which is the parameter that affects the prediction of the fan noise performance, namely the fan speed, and then the simulation model is subjected to noise performance simulation analysis, and the fan noise performance is solved to obtain the far-field measurement point sound pressure level spectrum data of the fan. There is a parameter, so a PointNet network with an input layer of n×4 matrix is constructed, and the point cloud data and the fan speed are used as input data samples, and the sound pressure level spectrum data is used as output data samples. The n×4 PointNet network is trained and optimized to obtain the trained n×4 PointNet network.
[0051] During actual prediction, the actual point cloud data and actual speed of the fan to be predicted are obtained and input into the trained n×4 PointNet network for noise performance prediction, and the predicted value of the sound pressure level spectrum data of the fan to be predicted is output.
[0052] In another example, the parameters affecting performance prediction include the rotation speed and incoming flow velocity of the equipment, the equipment performance data include the wind volume data and wind pressure data of the equipment, and the input layer of the deep learning network is an n×5 matrix.
[0053] For the aerodynamic prediction of the fan, it is necessary to output the fan's air volume-pressure data. The fan speed and incoming flow velocity are the key parameters that affect the predicted aerodynamic performance. The speed RPM and incoming flow velocity V are merged into the basic n×3 matrix as two columns to form an n×5 network input layer.
[0054] Specifically, taking the prediction of the noise performance of a fan as an example, Figure 4 As shown in the figure, after obtaining the point cloud data, the three-dimensional model and the simulation model (such as the wind tunnel simulation model) in turn, the simulation model is parameterized, which is the parameter that affects the prediction of the fan's aerodynamic performance, namely the fan's rotation speed and the incoming flow velocity. Then, the simulation model is simulated and analyzed for aerodynamic performance, and the fan's aerodynamic performance is solved to obtain the fan's air volume data and air pressure data. There are two types of parameters, so a PointNet network with an input layer of n×5 matrix is constructed, and the point cloud data, fan rotation speed and incoming flow velocity are used as input data samples, and the air volume data and air pressure data are used as output data samples. The n×5 PointNet network is trained and optimized to obtain the trained n×5 PointNet network.
[0055] During actual prediction, the actual point cloud data, actual rotation speed, and actual incoming flow velocity of the fan to be predicted are obtained and input into the trained n×5 PointNet network for aerodynamic performance prediction, and the predicted values of the air volume data and wind pressure data of the fan to be predicted are output.
[0056] In summary, the embodiments of the present invention have the following advantages: (1) Based on the PointNet network structure, the fan speed and parameters affecting the fan prediction performance are taken as additional input items and combined with point cloud data, which improves the model's ability to fully capture the factors affecting fan performance; (2) Using aerodynamic performance and sound pressure level spectrum data calculated by CFD as training samples improves the accuracy and reliability of deep learning network training; (3) The trained PointNet network model can quickly predict the fan aerodynamic performance and noise spectrum under different fan blade designs and operating conditions; (4) It can significantly improve the speed and efficiency of fan aerodynamic performance and noise spectrum prediction while maintaining the accuracy of the prediction, providing a new tool for fan design and optimization, helping to shorten product development cycle, reduce R&D costs, and improve product performance.
[0057] Corresponding to the device performance prediction method of the above embodiment, the present invention also proposes a device performance prediction apparatus.
[0058] Figure 5 It is a block diagram of a device for predicting performance of a device according to an embodiment of the present invention.
[0059] like Figure 5 As shown, the performance prediction device of the equipment includes an acquisition module 10, a simulation module 20, a construction module 30, a training module 40 and a prediction module 50.
[0060] Among them, the acquisition module 10 is used to acquire the point cloud data of the device; the simulation module 20 is used to perform simulation analysis on the performance of the device based on the point cloud data to obtain the device performance data, wherein, when performing the simulation analysis, the parameters affecting the performance prediction are determined and set; the construction module 30 is used to construct a deep learning network with an input layer of an n×(3+k) matrix, wherein n is the number of point cloud data, and k is the number of types of parameters affecting the performance prediction; the training module 40 is used to train the deep learning network using the point cloud data and the parameters affecting the performance prediction as input data and the device performance data as output data; the prediction module 50 is used to obtain the actual point cloud data of the device to be predicted and the actual parameters affecting the performance prediction, and input them into the trained deep learning network for performance prediction.
[0061] It should be noted that the specific implementation of the performance prediction device of the equipment can refer to the specific implementation of the performance prediction method of the above-mentioned equipment. In order to avoid redundancy, it will not be described in detail here.
[0062] The performance prediction device of the device in the embodiment of the present invention combines performance simulation analysis and deep learning network to achieve performance prediction, and can predict the performance of different devices quickly and efficiently.
[0063] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be construed as limitations on the present invention, and those of ordinary skill in the art can change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A method for predicting performance of a device, characterized in that: The following steps are involved: Get the point cloud data of the device; Performing simulation analysis on the performance of the device based on the point cloud data to obtain device performance data, wherein, when performing the simulation analysis, parameters affecting performance prediction are determined and set; Construct a deep learning network with an input layer of n×(3+k) matrix, where n is the number of point cloud data and k is the number of types of parameters that affect performance prediction; Using the point cloud data and the parameters affecting performance prediction as input data and the device performance data as output data, training the deep learning network; The actual point cloud data of the device to be predicted and the actual parameters that affect the performance prediction are obtained, and input into the trained deep learning network for performance prediction.
2. The performance prediction method of the device according to claim 1, characterized in that: The performance of the equipment is simulated and analyzed based on the point cloud data to obtain equipment performance data, including: constructing a three-dimensional model of the device according to the point cloud data; Using CFD to construct a simulation model of the three-dimensional model, including the construction of a calculation domain, a grid, and a desired simulation result output type; Determining parameters that affect performance prediction, and setting parameters of the simulation model according to the parameters that affect performance prediction; Perform performance simulation analysis on the set simulation model to obtain equipment performance data.
3. The performance prediction method of the device according to claim 1, characterized in that: The parameters affecting performance prediction include at least one of parameters affecting equipment noise performance prediction and parameters affecting equipment aerodynamic performance prediction.
4. The performance prediction method of the device according to claim 1, characterized in that: The equipment performance data includes at least one of noise performance data and aerodynamic performance data.
5. The performance prediction method of the device according to claim 1, characterized in that: The parameters affecting performance prediction include the rotation speed of the device, the device performance data include the sound pressure level spectrum data of the device, and the input layer of the deep learning network is an n×4 matrix.
6. The performance prediction method of the device according to claim 1, characterized in that: The parameters affecting performance prediction include the rotation speed and incoming flow velocity of the equipment, the equipment performance data include the wind volume data and wind pressure data of the equipment, and the input layer of the deep learning network is an n×5 matrix.
7. The performance prediction method of the device according to claim 1, characterized in that: The deep learning network is a PointNet network.
8. The performance prediction method of the device according to claim 1, characterized in that: The device is a heat dissipation device.
9. A device for predicting performance of equipment, characterized in that: include: Acquisition module, used to acquire point cloud data of the device; A simulation module, used for performing simulation analysis on the performance of the device based on the point cloud data to obtain device performance data, wherein when performing the simulation analysis, parameters affecting performance prediction are determined and set; A construction module is used to construct a deep learning network with an input layer of n×(3+k) matrix, where n is the number of point cloud data and k is the number of types of parameters that affect performance prediction; A training module, configured to train the deep learning network by taking the point cloud data and the parameters affecting performance prediction as input data and the device performance data as output data; The prediction module is used to obtain the actual point cloud data of the device to be predicted and the actual parameters that affect the performance prediction, and input them into the trained deep learning network for performance prediction.