CN117494779A - Training method of structure optimization model, structure optimization method and device - Google Patents

Abstract

The present disclosure provides a training method of a structural optimization model, a structural optimization method and a device, and relates to the field of artificial intelligence technology. The specific implementation plan is: sampling the component structure of sample engineering components in the engineering field to obtain multiple sampling points of the sample engineering components; training the stress solution network in the structural optimization model based on the location information of the sampling points to obtain the sampling points Predict stress; train the density solving network in the structural optimization model based on the predicted stress to obtain the predicted density of the sampling point; based on the location information, predicted stress and predicted density of the sampling point, the network parameters of the stress solving network and density solving network are respectively Make adjustments and continue training the adjusted network until the end conditions of model training are met, and the target structure optimization model is obtained.

Description

Training method of structural optimization model, structural optimization method and device

Technical field

The present disclosure relates to the field of artificial intelligence technology, and in particular to a training method of a structural optimization model, a structural optimization method and a device.

Background technique

In order to achieve gridlessness in the structural topology optimization process and avoid the waste of resources and time, deep learning networks can be used for structural topology optimization. However, deep learning networks cannot take into account both physical information and structural space information, and have large data requirements. The results are difficult to understand and have obvious flaws.

Contents of the invention

The present disclosure provides a training method for a structural optimization model, a structural optimization method, a device, an electronic device, a storage medium and a computer program product.

According to one aspect of the present disclosure, a training method for a structural optimization model is provided, including: sampling the component structure of a sample engineering component in the engineering field to obtain multiple sampling points of the sample engineering component; based on the sampling points The stress solving network in the structural optimization model is trained with the position information to obtain the predicted stress of the sampling point; the density solving network in the structural optimization model is trained based on the predicted stress to obtain the sampling point The predicted density; based on the position information of the sampling point, the predicted stress and the predicted density, adjust the network parameters of the stress solving network and the density solving network respectively, and continue to adjust the adjusted network Train until the end conditions of model training are met, and the target structure optimization model is obtained.

According to another aspect of the present disclosure, a structural optimization method is provided, including: obtaining the component structure of a target engineering component to be optimized; performing density prediction on the component structure of the target engineering component based on a target structure optimization model to obtain the Density distribution data of the component structure of the target engineering component, wherein the target structure optimization model is a model trained using the training method of the present disclosure.

According to another aspect of the present disclosure, a training device for a structural optimization model is provided, including: a sampling module for sampling the component structure of a sample engineering component in the engineering field to obtain multiple sampling points of the sample engineering component ; The first training module is used to train the stress solution network in the structural optimization model based on the position information of the sampling point to obtain the predicted stress of the sampling point; the second training module is used to train the stress solution network in the structural optimization model based on the location information of the sampling point; the second training module is used to train the stress solution network in the structural optimization model based on the location information of the sampling point; The stress trains the density solution network in the structural optimization model to obtain the predicted density of the sampling point; the adjustment module is used to respectively calculate the density based on the position information of the sampling point, the predicted stress and the predicted density. The network parameters of the stress solving network and the density solving network are adjusted, and the adjusted network is continued to be trained until the end condition of the model training is met, and the target structure optimization model is obtained.

According to another aspect of the present disclosure, a structure optimization device is provided, including: an acquisition module for acquiring the component structure of a target engineering component to be optimized; and a prediction module for predicting the target engineering component based on a target structure optimization model. Density prediction is performed on the component structure of the target engineering component to obtain density distribution data of the component structure of the target engineering component, wherein the target structure optimization model is a model trained using the training method of the present disclosure.

According to another aspect of the present disclosure, an electronic device is provided, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores information that can be used by the at least one processor. Execution instructions, the instructions are executed by the at least one processor, so that the at least one processor can execute the training method and the structural optimization method of the structural optimization model described in the above-mentioned aspect embodiment.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, on which computer programs/instructions are stored. The computer instructions are used to cause the computer to execute the above-mentioned embodiments of the aspect. The training method and structure optimization method of the structural optimization model described above.

According to another aspect of the present disclosure, a computer program product is provided, including a computer program/instructions that, when executed by a processor, implements the training method and structure of the structural optimization model described in the embodiments of the above aspect. Optimization.

It should be understood that what is described in this section is not intended to identify key or important features of the embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.

Description of the drawings

The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure. in:

Figure 1 is a schematic flowchart of a training method for a structural optimization model provided by an embodiment of the present disclosure;

Figure 2 is a schematic flowchart of another training method for a structural optimization model provided by an embodiment of the present disclosure;

Figure 3 is a schematic flowchart of training a structural optimization model provided by an embodiment of the present disclosure;

Figure 4 is a schematic flowchart of another training method for a structural optimization model provided by an embodiment of the present disclosure;

Figure 5 is a schematic flow chart of a structure optimization method provided by an embodiment of the present disclosure;

Figure 6 is a schematic structural diagram of a training device for a structural optimization model provided by an embodiment of the present disclosure;

Figure 7 is a schematic structural diagram of a structure optimization device provided by an embodiment of the present disclosure;

FIG. 8 is a block diagram of an electronic device used to implement a training method of a structural optimization model according to an embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the present disclosure are included to facilitate understanding and should be considered to be exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

Artificial Intelligence (AI) is a technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. Currently, AI technology has the advantages of high automation, high accuracy, and low cost, and has been widely used.

Deep Learning (DL) is a new research direction in the field of Machine Learning (ML). It is to learn the inherent laws and representation levels of sample data, so that the machine can analyze and learn like humans, and can identify A science of data such as text, images, and sounds, widely used in speech and image recognition.

Structural topology optimization is an important engineering design task and is currently being widely used in engineering fields such as aerospace, automotive manufacturing, architectural design, and mechanical engineering to improve product design, enhance performance, and reduce material and energy waste. It can be used to design the structure of new materials to improve the strength and durability of the material, or to design the structure of certain specific components to reduce the weight of the structure while ensuring the strength of the structure.

The training method of the structural optimization model provided by the embodiment of the present disclosure can be applied to engineering fields such as aerospace, automobile manufacturing, and architectural design, and is mainly aimed at the structural design and optimization of engineering components.

Figure 1 is a schematic flowchart of a training method for a structural optimization model provided by an embodiment of the present disclosure. As shown in Figure 1, the training method of this structural optimization model may include:

S101: Sampling the component structure of the sample engineering component in the engineering field to obtain multiple sampling points of the sample engineering component.

It should be noted that the execution subject of the training method of the structural optimization model in the embodiment of the present disclosure may be a hardware device with data information processing capabilities and/or the necessary software required to drive the hardware device to work. Optionally, execution subjects may include servers, computers, user terminals and other intelligent devices. The embodiments of this disclosure are not specifically limited.

It can be understood that the component structure of engineering components refers to the structure of machine learning or deep learning designed and built to solve specific problems of the component in the engineering field. The component structure of the engineering component may be a network topology structure.

In some implementations, multiple sampling points of the sample engineering component are obtained by sampling the component structure of the sample engineering component. Among them, the sampling points collect location information. For example, the sampling point is (x, y, z) in the spatial coordinate system. Optionally, the component structure of the sample engineering component can be randomly sampled, or the component structure of the sample engineering component can be sampled evenly to obtain multiple sampling points. For example, the center area, edge area, corner area, etc. on the component structure of the sample engineering component can be sampled.

S102: Train the stress solving network in the structural optimization model based on the position information of the sampling point to obtain the predicted stress of the sampling point.

In some implementations, the stress solving network in the structural optimization model can be built in advance, and the position information of the sampling points is input into the stress solving network for training to obtain the predicted stress of the sampling points. Among them, the stress solution network can be a linear neural network, and the predicted stress is expressed as (u, v, w).

Optionally, the number of training times of the stress solution network can be set. When the number of training times of the stress solution network reaches the set number, the training of the stress solution network is ended, and then the training of the density solution network is started.

S103. Train the density solution network in the structural optimization model based on the predicted stress to obtain the predicted density of the sampling point.

In some implementations, the density solving network in the structural optimization model can be pre-constructed, and the predicted stress is input into the density solving network for training to obtain the predicted density of the sampling point. Among them, the density solving network can be a convolutional neural network, and the predicted density is expressed as (d).

In some implementations, the stress solver network and the density solver network can be trained alternately. In other words, you can set the training times of the stress solving network and the training times of the density solving network respectively. The stress solving network is trained first. When the training times reach the set number, the density solving network starts to be trained. After the density solving network is trained, the density solving network is trained. When the number of training times of the network reaches the set number, the stress solving network continues to be trained. In the embodiment of the present disclosure, by alternately training two solving networks, the error accumulation of serial training of the two solving networks can be reduced, and the training accuracy of the two solving networks can be improved.

For example, the stress solution network is trained first. When the number of training times reaches 50, the density solution network is trained. When the number of training reaches 100 times, the stress solution network is returned to be trained.

S104, based on the position information of the sampling point, predicted stress and predicted density, adjust the network parameters of the stress solving network and the density solving network respectively, and continue to train the adjusted network until the end conditions of the model training are met, and the target structure is obtained Optimize the model.

In some implementations, the loss function of the stress solving network can be determined based on the location information of the sampling point, the predicted stress and the predicted density, the network parameters of the stress solving network can be adjusted based on the loss function, and the adjusted stress solving network can continue to be modified. train.

In some implementations, the loss function of the density solving network can also be determined based on the position information of the sampling point, predicted stress and predicted density, the network parameters of the density solving network can be adjusted based on the loss function, and the adjusted density solving network can continue to be used. Conduct training.

Optionally, since the stress solution network and the density solution network are trained alternately, the network parameters of the stress solution network can be adjusted based on the loss function of the stress solution network, and the adjusted stress solution network can continue to be trained until the set value is reached. A certain number of times. Further, based on the loss function of the density solving network, the network parameters of the density solving network are adjusted, and the adjusted density solving network continues to be trained.

Optionally, the loss function of the network can be solved based on density to determine the end condition of model training. Since the smaller the loss function is, the better the training effect of the model is. You can set the threshold of the loss function of the density solution network, and make the loss function of the density solution network less than or equal to the set threshold as the end condition of the model training.

In some implementations, when training the adjusted network, it is determined whether the loss function of the density solving network is less than or equal to the set threshold, and in response to the loss function of the density solving network being less than or equal to the set threshold, it is determined that the end condition of the model training is met. , end the training of the structural optimization model, and obtain the target structural optimization model.

According to the training method of the structural optimization model provided by the embodiment of the present disclosure, by sampling the component structure of the sample engineering component in the engineering field, multiple sampling points of the sample engineering component are obtained, and the location information of the sampling points is input into the structural optimization model. The stress solving network is trained to obtain the predicted stress of the sampling point. The predicted stress is then input into the density solution network in the structural optimization model for training, and the predicted density of the sampling point is obtained, which can improve the correlation between stress information and density information. Further, based on the position information, predicted stress and predicted density of the sampling points, the network parameters are adjusted, and the adjusted network continues to be trained until the structure optimization model meets the training end conditions, and the target structure optimization model is obtained, which improves the target structure optimization. The generalization ability of the model, which combines position information, stress and density, can improve the capture and application of spatial information by the target structure optimization model during the model optimization process, so as to help the target structure optimization model understand the spatial relationship and geometry of engineering components. structure, and solve the problem of insufficient interpretability of the target structure optimization model when solving physical problems.

FIG. 2 is a schematic flowchart of a training method for a structural optimization model provided by an embodiment of the present disclosure. As shown in Figure 2, the training method of this structural optimization model may include:

S201: Sampling the component structure of the sample engineering component in the engineering field to obtain multiple sampling points of the sample engineering component.

S202: Train the stress solving network in the structural optimization model based on the position information of the sampling point to obtain the predicted stress of the sampling point.

S203: Train the density solving network in the structural optimization model based on the predicted stress to obtain the predicted density of the sampling point.

The relevant content of steps S201-S203 can be found in the above embodiments and will not be described again here.

In some implementations, the stress solution network is a linear neural network, and the density solution network is a convolutional neural network. The stress solution network and the density solution network are connected in series, which can directly correlate stress information and density information, and can provide real and certain prediction results.

It should be noted that when training the network in the structural optimization model, the stress solving network and the density solving network are trained alternately, and the network parameters of the other network are maintained during the training of one network, which can promote stress information and density. Correlation and fusion of information to improve the overall performance of the model.

That is to say, whenever the number of training times for one network reaches the first set number, the training of the other network is started. Whenever the training parameters of another network reach the second set number of times, training of one of the networks is started.

For example, assume that the first set number of times for the stress solution network is 100, and the second set number for the density solution network is 80. When the number of training times for the stress solution network reaches 100 times, the training of the density solution network is started; when When the number of training times of the density solution network reaches 80 times, the training of the corresponding stress solution network is started. Among them, when training the stress solving network, the network parameters of the density solving network remain unchanged. While training the density solver network, the network parameters of the stress solver network are kept constant.

S204: Adjust the network parameters of the stress solution network based on the position information, predicted stress and predicted density of the sampling point.

In some implementations, the network parameters of the stress solving network can be adjusted based on the loss function of the stress solving network in order to optimize the performance and generalization capabilities of the stress solving network. Optionally, the loss function of the stress solving network can be determined based on the predicted stress and physical information of the sampling points.

Optionally, the physical information of the sampling point can be obtained based on the location information, predicted stress and predicted density of the sampling point, where the physical information at least includes the compliance parameter of the sampling point. Based on the compliance equation, the location information, predicted stress and predicted density of the sampling point can be input into the compliance equation and the compliance parameters of the sampling point can be output.

Furthermore, based on the predicted stress and physical information of the sampling points, the first loss function of the stress solution network is obtained, and the network parameters of the stress solution network are adjusted based on the first loss function, which can make the stress solution network move in a better direction. Adjustment allows the structural optimization model to be gradually optimized during the training process.

S205: Adjust the network parameters of the density solution network based on the position information, predicted stress and predicted density of the sampling point.

In some implementations, the network parameters of the density solving network can be adjusted based on the loss function of the density solving network in order to optimize the performance and generalization capabilities of the density solving network. Optionally, the loss function of the density solving network can be determined based on the predicted density and physical information of the sampling points.

Optionally, the physical information of the sampling point can be obtained based on the location information, predicted stress and predicted density of the sampling point. Among them, the physical information can be the compliance parameter of the sampling point. Based on the compliance equation, the location information, predicted stress and predicted density of the sampling point can be input into the compliance equation and the compliance parameters of the sampling point can be output.

Further, based on the predicted density and physical information of the sampling points, a second loss function of the density solution network is obtained, and the network parameters of the density solution network are adjusted based on the second loss function. The density solution network can be adjusted in a more optimal direction, allowing the structural optimization model to be gradually optimized during the training process.

Optionally, in order to retain a relatively simple and smooth structure, Gaussian filtering can be performed on the physical information of the sampling points to obtain the derivative information of the physical information, where the derivative information refers to the derivative of the compliance parameter. Based on the predicted density, physical information and derivative information of the sampling points, the second loss function of the density solving network is obtained. Combining Gaussian filtering with loss function design can improve the smoothness of the prediction results of the density solving network, and then optimize the structural optimization model. of smoothness.

In some implementations, the network parameters of any solution network can also be adjusted based on the gradient information of the loss function to optimize the performance and generalization ability of the solution network and the structural optimization model. Optionally, for any solution network among the stress solution network and the density solution network, perform gradient calculation on the loss function of any solution network to obtain the gradient information of any solution network, and then calculate the gradient information of any solution network based on the gradient information. Network parameters are adjusted.

S206: Continue training the adjusted network until the end conditions of model training are met, and obtain the target structure optimization model.

In some implementations, after each training of the density solving network ends, the end condition of the model training can be determined based on the second loss function of the density solving network. Since the smaller the loss function is, the better the training effect of the model is. You can set a set threshold for the second loss function, and make the second loss function less than or equal to the set threshold as the end condition of the model training.

Optionally, after each training of the density solution network is completed, the second loss function of the density solution network is compared with the set threshold. If the second loss function is less than or equal to the set threshold, it is determined that the end condition of the model training is met.

According to the training method of the structural optimization model provided by the embodiment of the present disclosure, by sampling the component structure of the sample engineering component in the engineering field, multiple sampling points of the sample engineering component are obtained, and the location information of the sampling points is input into the structural optimization model. The stress solving network is trained to obtain the predicted stress of the sampling point. The predicted stress is then input into the density solution network in the structural optimization model for training, and the predicted density of the sampling point is obtained, which can improve the correlation between stress information and density information. Further, based on the position information, predicted stress and predicted density of the sampling point, the first loss function and the second loss function are determined, and the gradient information of the loss function is calculated, the network parameters are adjusted based on the gradient information, and the adjusted network parameters are continued to be trained. network until the structure optimization model meets the training end conditions, and the target structure optimization model is obtained, which improves the generalization ability of the target structure optimization model. Combining position information, stress and density can improve the target structure optimization model during the model optimization process. Capture and apply spatial information to help the target structure optimization model understand the spatial relationship and geometric structure of engineering components, and solve the problem of insufficient interpretability when the target structure optimization model solves physical problems.

The flow chart for training the structural optimization model is shown in Figure 3. By obtaining multiple sampling points of the component structure of the sample engineering component and inputting them into the pre-built stress solving network, training is performed to obtain the predicted stress of the sampling points, and then the predicted stress values are input into the pre-built density solving network for training. Get the predicted density of the sampling point. Based on the predicted stress, predicted density, and location information of the sampling point, the compliance parameters of the sampling point are determined as physical information, and the first loss function of the stress solution network is calculated based on the physical information and the predicted stress, and the gradient calculation of the first loss function is performed , determine the gradient information of the first loss function, update and adjust the network parameters of the stress solution network based on the gradient information, and obtain the adjusted stress solution network. While training the stress solver network and adjusting network parameters, keep the network parameters of the density solver network unchanged.

Further, after training the stress solving network for a set number of times, the training of the density solving network is started. Input the predicted stress output from the last training of the stress solving network into the density solving network, train the predicted density of the sampling point, and determine the compliance parameters of the sampling point as physical information based on the predicted stress, predicted density and location information of the sampling point, and Perform Gaussian filtering on the physical information of the sampling points to obtain the derivative information of the physical information. Then based on the predicted density, physical information and derivative information, the second loss function of the density solution network can be determined. By performing gradient calculation on the second loss function, the gradient information of the second loss function is determined, and the network parameters of the density solution network are updated and adjusted based on the gradient information to obtain an adjusted density solution network. When training the density solver network and adjusting network parameters, keep the network parameters of the stress solver network unchanged. After each density solving network training is completed, the second loss function of the density solving network is compared with the set threshold. If the second loss function is less than or equal to the set threshold, it is determined that the end conditions of the model training are met, and the model training is ended. Obtain the target structure optimization model.

Figure 4 is a schematic flowchart of a training method for a structural optimization model provided by an embodiment of the present disclosure. As shown in Figure 4, the training method of the structural optimization model may include:

S401: Sampling the component structure of the sample engineering component in the engineering field to obtain multiple sampling points of the sample engineering component.

S402: Train the stress solving network in the structural optimization model based on the position information of the sampling point to obtain the predicted stress of the sampling point.

S403: Train the density solution network in the structural optimization model based on the predicted stress to obtain the predicted density of the sampling point.

S404: Obtain the physical information of the sampling point based on the position information, predicted stress and predicted density of the sampling point, where the physical information at least includes the compliance parameter of the sampling point.

S405. Based on the predicted stress and physical information of the sampling points, obtain the first loss function of the stress solution network.

S406: Adjust the network parameters of the stress solution network based on the first loss function.

S407: Obtain the physical information of the sampling point based on the position information, predicted stress and predicted density of the sampling point.

S408: Based on the predicted density and physical information of the sampling points, obtain the second loss function of the density solution network.

S409: Adjust the network parameters of the density solution network based on the second loss function.

S410: Continue training the adjusted network until the end conditions of model training are met, and obtain the target structure optimization model.

According to the training method of the structural optimization model provided by the embodiment of the present disclosure, by sampling the component structure of the sample engineering component in the engineering field, multiple sampling points of the sample engineering component are obtained, and the location information of the sampling points is input into the structural optimization model. The stress solving network is trained to obtain the predicted stress of the sampling point. The predicted stress is then input into the density solution network in the structural optimization model for training, and the predicted density of the sampling point is obtained, which can improve the correlation between stress information and density information. Further, based on the position information, predicted stress and predicted density of the sampling point, the first loss function and the second loss function are obtained, the network parameters are adjusted, and the adjusted network is continued to be trained until the structural optimization model meets the training end conditions, The target structure optimization model is obtained, which improves the generalization ability of the target structure optimization model. Combining position information, stress and density can improve the capture and application of spatial information by the target structure optimization model during the model optimization process, so as to help the target The structural optimization model understands the spatial relationship and geometric structure of engineering components, and solves the problem of insufficient interpretability when the target structural optimization model solves physical problems.

Figure 5 is a schematic flowchart of a structure optimization method provided by an embodiment of the present disclosure.

As shown in Figure 5, the structure optimization method may include:

S501: Obtain the component structure of the target engineering component to be optimized.

In some implementations, the component structure of the target engineering component can be generated by scanning and measuring the target engineering component to be optimized. The candidate component structure of the target engineering component can also be obtained from the model library as the component structure of the target engineering component. Among them, it is necessary to ensure that the component structure matches the shape and size of the target engineering component.

S502: Predict the density of the component structure of the target engineering component based on the target structure optimization model, and obtain the density distribution data of the component structure of the target engineering component.

It should be noted that the target structure optimization model can be obtained by using the training method of the structure optimization model shown in Figures 1 to 4, which will not be described again here.

In some implementations, by determining the target position on the component structure of the target engineering component and performing position acquisition on the target position, the position information of the target position on the component structure of the target engineering component is obtained, and based on the target structure optimization model, based on Using position information for density prediction, the density distribution data of the component structure can be obtained.

Optionally, input the position information of the target position into the stress solving network of the target structure optimization model to obtain the predicted stress at the target position, and input the predicted stress at the target position into the density solving network of the target structure optimization model to obtain the target position. Predict density.

Furthermore, using each position on the component structure of the target engineering component as a target position, based on the target structure optimization model, the predicted density based on each target position can be obtained, and then the density distribution data of the component structure of the target engineering component can be generated.

In some implementations, the component structure of the target engineering component can be determined based on the density distribution data to determine whether it meets the engineering requirements. For example, engineering requirements require that the component structure of the target engineering component has a uniform density. Based on the density distribution data, it can be determined whether the component structure of the target engineering component has a uniform density.

In some implementations, if the engineering requirements are not met, the optimization information of the component structure of the target engineering component can be generated based on the density distribution data, and based on the optimization information, the component structure of the target engineering component can be optimized. For example, based on density distribution data, abnormal areas with lower or higher density in the component structure of the target engineering component can be obtained, and structural optimization can be performed based on the abnormal density areas. In the embodiments of the present disclosure, the obtained density distribution data of the component structure of the target engineering component can help optimize structural performance, reduce production costs, and improve structural performance and reliability.

For example, taking the field of automobile manufacturing as an example, assuming that the target engineering component is a door, by measuring the door, the component structure of the door can be generated, and the target position on the component structure can be collected to obtain the component structure of the door. The position information of the target position is input into the target structure optimization model for density prediction, and the density distribution data of the door component structure can be obtained.

Furthermore, based on the density distribution data, it can be determined whether the component structure of the door meets the needs of automobile manufacturing, such as ensuring the strength and stiffness of the door. When the demand for automobile manufacturing is not met, the optimization information of the car door can be generated based on the density distribution data, and the car door can be optimized based on the optimization information.

According to the structural optimization method provided by the embodiment of the present disclosure, by obtaining the component structure of the target engineering component to be optimized, and inputting it into the target structure optimization model based on the position information of the target position on the component structure, the component structure of the target engineering component can be obtained density distribution data. Optimizing the structure of target engineering components based on density distribution data is beneficial to improving structural design, improving performance, and reducing waste of materials and energy.

Corresponding to the training methods of the structural optimization model provided by the above embodiments, one embodiment of the present disclosure also provides a training device of the structural optimization model. Since the training device of the structural optimization model provided by the embodiment of the present disclosure is consistent with the above-mentioned The training methods of the structural optimization model provided by several embodiments correspond to each other. Therefore, the implementation of the training method of the structural optimization model mentioned above is also applicable to the training device of the structural optimization model provided by the embodiments of the present disclosure, and will not be detailed in the following embodiments. describe.

FIG. 6 is a schematic structural diagram of a training device for a structural optimization model provided by an embodiment of the present disclosure.

As shown in FIG. 6 , the training device 600 of the structural optimization model according to the embodiment of the present disclosure includes: a sampling module 601 , a first training module 602 , a second training module 603 and an adjustment module 604 .

The sampling module 601 is used to sample the component structure of sample engineering components in the engineering field and obtain multiple sampling points of the sample engineering components.

The first training module 602 is used to train the stress solution network in the structural optimization model based on the position information of the sampling point to obtain the predicted stress of the sampling point.

The second training module 603 is used to train the density solving network in the structural optimization model based on the predicted stress to obtain the predicted density of the sampling point.

Adjustment module 604 is configured to adjust the network parameters of the stress solution network and the density solution network respectively based on the position information of the sampling point, the predicted stress and the predicted density, and continue to adjust the adjusted network parameters. The network is trained until the end conditions of model training are met, and the target structure optimization model is obtained.

In one embodiment of the present disclosure, the adjustment module 604 is further configured to: obtain the physical information of the sampling point based on the position information of the sampling point, the predicted stress and the predicted density, wherein the The physical information at least includes the compliance parameter of the sampling point; based on the predicted stress of the sampling point and the physical information, a first loss function of the stress solution network is obtained; based on the first loss function, the stress The network parameters of the solving network are adjusted.

In one embodiment of the present disclosure, the adjustment module 604 is further configured to: obtain the physical information of the sampling point based on the location information of the sampling point, the predicted stress and the predicted density; The predicted density of the sampling point and the physical information are used to obtain the second loss function of the density solving network; the network parameters of the density solving network are adjusted based on the second loss function.

In one embodiment of the present disclosure, the adjustment module 604 is also configured to: perform Gaussian filtering on the physical information of the sampling points to obtain derivative information of the physical information; based on the predicted density of the sampling points, the The physical information and the derivative information are used to obtain the second loss function of the density solution network.

In one embodiment of the present disclosure, the adjustment module 604 is further configured to: perform a gradient on the loss function of any one of the stress solution network and the density solution network. Calculate to obtain the gradient information of any solving network; adjust the network parameters of any solving network based on the gradient information.

In one embodiment of the present disclosure, the apparatus further includes: alternately training the stress solution network and the density solution network, and maintaining network parameters of the other network during the training of one network.

In one embodiment of the present disclosure, the device further includes: whenever the number of training times for one of the networks reaches a first set number of times, starting training for the other network; After the training parameters of the network reach the second set number of times, training of one of the networks is started.

In one embodiment of the present disclosure, the stress solution network is a linear neural network, the density solution network is a convolutional neural network, and the stress solution network is connected in series with the density solution network.

In one embodiment of the present disclosure, the adjustment module 604 is also used to: after each training of the density solution network, compare the second loss function of the density solution network with a set threshold, if If the second loss function is less than or equal to the set threshold, it is determined that the end condition of the model training is met.

According to the training device of the structural optimization model provided by the embodiment of the present disclosure, by sampling the component structure of the sample engineering component in the engineering field, multiple sampling points of the sample engineering component are obtained, and the position information of the sampling points is input into the structural optimization model. The stress solving network is trained to obtain the predicted stress of the sampling point. The predicted stress is then input into the density solution network in the structural optimization model for training, and the predicted density of the sampling point is obtained, which can improve the correlation between stress information and density information. Further, based on the position information, predicted stress and predicted density of the sampling points, the network parameters are adjusted, and the adjusted network continues to be trained until the structure optimization model meets the training end conditions, and the target structure optimization model is obtained, which improves the target structure optimization. The generalization ability of the model, which combines position information, stress and density, can improve the capture and application of spatial information by the target structure optimization model during the model optimization process, so as to help the target structure optimization model understand the spatial relationship and geometry of engineering components. structure, and solve the problem of insufficient interpretability of the target structure optimization model when solving physical problems.

According to an embodiment of the present disclosure, the present disclosure also provides a structural optimization device for implementing the above-mentioned structural optimization method.

Figure 7 is a schematic structural diagram of a structure optimization device according to the first embodiment of the present disclosure.

As shown in Figure 7, the structure optimization device 700 of the embodiment of the present disclosure includes: a first acquisition module 701 and a prediction module 702.

The acquisition module 701 is used to acquire the component structure of the target engineering component to be optimized.

The prediction module 702 is configured to perform density prediction on the component structure of the target engineering component based on the target structure optimization model, and obtain density distribution data of the component structure of the target engineering component.

In one embodiment of the present disclosure, the device further includes: based on the density distribution data, generating optimization information of the component structure of the target engineering component; based on the optimization information, generating optimization information on the component structure of the target engineering component. optimize.

In one embodiment of the present disclosure, the prediction module 702 is also used to: obtain the position information of the target position on the component structure of the target engineering component; input the position information of the target position into the target structure optimization In the stress solving network of the model, the predicted stress of the target position is obtained; the predicted stress of the target position is input into the density solving network of the target structure optimization model, and the predicted density of the target position is obtained; based on each The predicted density of the target location is used to generate density distribution data of the component structure of the target engineering component.

According to the structure optimization device provided by the embodiment of the present disclosure, by obtaining the component structure of the target engineering component to be optimized, and inputting it into the target structure optimization model based on the position information of the target position on the component structure, the component structure of the target engineering component can be obtained density distribution data. Optimizing the structure of target engineering components based on density distribution data is beneficial to improving structural design, improving performance, and reducing waste of materials and energy.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

Figure 8 shows a schematic block diagram of an example electronic device 800 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to refer to various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are examples only and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 8 , the electronic device 800 includes a computing unit 801 that can operate according to a computer program/instructions stored in a read-only memory (ROM) 802 or loaded from a storage unit 806 into a random access memory (RAM) 803 /command to perform various appropriate actions and processing. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. Computing unit 801, ROM 802 and RAM 803 are connected to each other via bus 804. An input/output (I/O) interface 805 is also connected to bus 804.

Multiple components in the device 800 are connected to the I/O interface 805, including: input unit 806 such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as magnetic disk, optical disk, etc.; and a communication unit 809, such as a network card, modem, wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information/data with other devices through computer networks such as the Internet and/or various telecommunications networks.

Computing unit 801 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processing processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the training method of the structural optimization model. For example, in some embodiments, the training method of the structural optimization model can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as the storage unit 806. In some embodiments, part or all of the computer program/instructions May be loaded and/or installed onto device 800 via ROM 802 and/or communication unit 809. When the computer program/instructions are loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the training method of the structural optimization model described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the training method of the structural optimization model in any other suitable manner (eg, by means of firmware).

Various implementations of the systems and techniques described above may be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip implemented in a system (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or a combination thereof. These various implementations may include implementation in one or more computer programs/instructions executable and/or interpreted on a programmable system including at least one programmable processor, The programmable processor may be a special purpose or general purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device , and the at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that the program codes, when executed by the processor or controller, cause the functions specified in the flowcharts and/or block diagrams/ The operation is implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, laptop disks, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer. Other kinds of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and may be provided in any form, including Acoustic input, voice input or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., A user's computer having a graphical user interface or web browser through which the user can interact with implementations of the systems and technologies described herein), or including such backend components, middleware components, or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communications network). Examples of communication networks include: local area network (LAN), wide area network (WAN), the Internet, and blockchain networks.

Computer systems may include clients and servers. Clients and servers are generally remote from each other and typically interact over a communications network. The relationship of client and server is created by computer programs/instructions running on respective computers and having a client-server relationship with each other. The server can be a cloud server, a distributed system server, or a server combined with a blockchain.

It should be understood that various forms of the process shown above may be used, with steps reordered, added or deleted. For example, each step described in the disclosure may be executed in parallel, sequentially, or in a different order. As long as the desired results of the technical solution disclosed in the present disclosure can be achieved, there is no limitation here.

The above-mentioned specific embodiments do not constitute a limitation on the scope of the present disclosure. It will be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions are possible depending on design requirements and other factors. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this disclosure shall be included in the protection scope of this disclosure.

Claims (27)

1. A method of training a structural optimization model, wherein the method comprises:

sampling the component structure of a sample engineering component in the engineering field to obtain a plurality of sampling points of the sample engineering component;

training a stress solving network in the structure optimization model based on the position information of the sampling points to obtain predicted stress of the sampling points;

training a density solving network in the structure optimization model based on the predicted stress to obtain the predicted density of the sampling point;

Based on the position information of the sampling points, the predicted stress and the predicted density, respectively adjusting network parameters of the stress solving network and the density solving network, and continuously training the adjusted network until the end condition of model training is met, so as to obtain a target structure optimization model.

2. The method of claim 1, wherein adjusting network parameters of the stress solving network comprises:

obtaining physical information of the sampling point based on the position information of the sampling point, the predicted stress and the predicted density, wherein the physical information at least comprises compliance parameters of the sampling point;

based on the predicted stress of the sampling point and the physical information, a first loss function of the stress solving network is obtained;

and adjusting network parameters of the stress solving network based on the first loss function.

3. The method of claim 1, wherein adjusting network parameters of the density solution network comprises:

obtaining physical information of the sampling point based on the position information of the sampling point, the predicted stress and the predicted density;

Obtaining a second loss function of the density solving network based on the predicted density of the sampling points and the physical information;

and adjusting network parameters of the density solving network based on the second loss function.

4. A method according to claim 3, wherein said deriving a second loss function of said density solving network based on said predicted density of said sampling points and said physical information comprises:

carrying out Gaussian filtering on the physical information of the sampling points to obtain derivative information of the physical information;

and obtaining a second loss function of the density solving network based on the predicted density of the sampling point, the physical information and the derivative information.

5. A method according to claim 2 or 3, wherein the method further comprises:

aiming at any one of the stress solving network and the density solving network, carrying out gradient calculation on a loss function of the any one solving network to obtain gradient information of the any one solving network;

and adjusting network parameters of any solving network based on the gradient information.

6. The method of any of claims 1-4, wherein the method further comprises:

The stress solving network and the density solving network are trained alternately, and network parameters of one network are maintained during the training of the other network.

7. The method of claim 5, wherein the method further comprises:

after the training times of one network reach the first set times, training the other network is started;

and starting to train one of the networks after the training parameters of the other network reach the second set times.

8. The method of any of claims 1-4, wherein the stress solving network is a linear neural network, the density solving network is a convolutional neural network, and the stress solving network is connected in series with the density solving network.

9. The method of any of claims 1-4, wherein the method further comprises:

and after each training of the density solving network is finished, comparing a second loss function of the density solving network with a set threshold value, and determining that the finishing condition of the model training is met if the second loss function is smaller than or equal to the set threshold value.

10. A method of structural optimization, wherein the method comprises:

Acquiring a component structure of a target engineering component to be optimized;

and carrying out density prediction on the component structure of the target engineering component based on a target structure optimization model to obtain density distribution data of the component structure of the target engineering component, wherein the target structure optimization model is trained by the method according to any one of claims 1-9.

11. The method of claim 10, wherein the method further comprises:

generating optimization information of the component structure of the target engineering component based on the density distribution data;

and optimizing the component structure of the target engineering component based on the optimization information.

12. The method according to claim 10 or 11, wherein the performing density prediction on the component structure of the target engineering component based on the target structure optimization model to obtain density distribution data of the component structure of the target engineering component includes:

acquiring position information of a target position on a component structure of the target engineering component;

inputting the position information of the target position into a stress solving network of the target structure optimizing model to obtain the predicted stress of the target position;

Inputting the predicted stress of the target position into a density solving network of the target structure optimizing model to obtain the predicted density of the target position;

and generating density distribution data of the component structure of the target engineering component based on the predicted density of each target position.

13. A training device for a structural optimization model, wherein the device comprises:

the sampling module is used for sampling the component structure of the sample engineering component in the engineering field to obtain a plurality of sampling points of the sample engineering component;

the first training module is used for training a stress solving network in the structure optimization model based on the position information of the sampling points to obtain predicted stress of the sampling points;

the second training module is used for training a density solving network in the structure optimization model based on the predicted stress to obtain the predicted density of the sampling point;

and the adjusting module is used for respectively adjusting network parameters of the stress solving network and the density solving network based on the position information of the sampling points, the predicted stress and the predicted density, and continuously training the adjusted network until the end condition of model training is met, so as to obtain a target structure optimization model.

14. The apparatus of claim 13, wherein the adjustment module is further configured to:

obtaining physical information of the sampling point based on the position information of the sampling point, the predicted stress and the predicted density, wherein the physical information at least comprises compliance parameters of the sampling point;

based on the predicted stress of the sampling point and the physical information, a first loss function of the stress solving network is obtained;

and adjusting network parameters of the stress solving network based on the first loss function.

15. The apparatus of claim 13, wherein the adjustment module is further configured to:

obtaining physical information of the sampling point based on the position information of the sampling point, the predicted stress and the predicted density;

obtaining a second loss function of the density solving network based on the predicted density of the sampling points and the physical information;

and adjusting network parameters of the density solving network based on the second loss function.

16. The apparatus of claim 15, wherein the adjustment module is further configured to:

carrying out Gaussian filtering on the physical information of the sampling points to obtain derivative information of the physical information;

And obtaining a second loss function of the density solving network based on the predicted density of the sampling point, the physical information and the derivative information.

17. The apparatus of claim 14 or 15, wherein the adjustment module is further configured to:

aiming at any one of the stress solving network and the density solving network, carrying out gradient calculation on a loss function of the any one solving network to obtain gradient information of the any one solving network;

and adjusting network parameters of any solving network based on the gradient information.

18. The apparatus of any of claims 13-16, wherein the apparatus further comprises:

the stress solving network and the density solving network are trained alternately, and network parameters of one network are maintained during the training of the other network.

19. The apparatus of claim 17, wherein the apparatus further comprises:

after the training times of one network reach the first set times, training the other network is started;

and starting to train one of the networks after the training parameters of the other network reach the second set times.

20. The apparatus of any of claims 13-16, wherein the stress solving network is a linear neural network, the density solving network is a convolutional neural network, and the stress solving network is connected in series with the density solving network.

21. The apparatus of any of claims 13-16, wherein the adjustment module is further to:

and after each training of the density solving network is finished, comparing a second loss function of the density solving network with a set threshold value, and determining that the finishing condition of the model training is met if the second loss function is smaller than or equal to the set threshold value.

22. A structure optimization apparatus, wherein the apparatus comprises:

the acquisition module is used for acquiring the component structure of the target engineering component to be optimized;

a prediction module, configured to perform density prediction on a component structure of the target engineering component based on a target structure optimization model, to obtain density distribution data of the component structure of the target engineering component, where the target structure optimization model is obtained by training an apparatus according to any one of claims 13-21.

23. The apparatus of claim 22, wherein the apparatus further comprises:

Generating optimization information of the component structure of the target engineering component based on the density distribution data;

and optimizing the component structure of the target engineering component based on the optimization information.

24. The apparatus of claim 22 or 23, wherein the prediction module is further configured to:

acquiring position information of a target position on a component structure of the target engineering component;

inputting the position information of the target position into a stress solving network of the target structure optimizing model to obtain the predicted stress of the target position;

inputting the predicted stress of the target position into a density solving network of the target structure optimizing model to obtain the predicted density of the target position;

and generating density distribution data of the component structure of the target engineering component based on the predicted density of each target position.

25. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-12.

26. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-12.

27. A computer program product comprising computer program/instructions which, when executed by a processor, implement the method steps of any one of claims 1 to 12.