A model, method, apparatus, medium, and computer for generating a cooling plate topology.

Abstract

This invention discloses a model, method, apparatus, medium, and computer for generating cooling plate topologies. The model includes a conditional control module, a convolution module, a diffusion model, and a U-shaped Mamba model. The conditional control module acquires structural requirement feature maps and physical field cloud maps of multiple existing cooling plate topologies, and concatenates and encodes them to obtain a conditional feature tensor. The convolution module performs convolution operations on the conditional feature tensor to obtain detailed and macroscopic features. The diffusion model acquires cooling plate operating condition features and structural requirement feature maps, and iteratively denoises the structural requirement feature maps based on the cooling plate operating condition features, detailed features, and macroscopic features to obtain a cross-attention map and a multi-scale feature map. The U-shaped Mamba model concatenates the cross-attention map and the multi-scale feature map, and models and generates the cooling plate topology based on the concatenated feature tensor. This approach improves the model's ability to generate cooling plate topologies.

Description

Technical Field

[0001] This invention relates to the field of data-driven generation strategy technology, and in particular to a generation model, method, apparatus, medium and computer for a cooling plate topology. Background Technology

[0002] Bioprinting technology has significant potential applications in tissue engineering and regenerative medicine. Extrusion printing has become the mainstream 3D printing method due to its simple process and wide material adaptability. However, the medical hydrogel materials used in printing are highly temperature sensitive, and their rheological properties depend on precise thermal environment control. Insufficient temperature control will damage the fluidity and forming stability of the hydrogel, directly affecting the geometric accuracy and biological function of the printed structure. Therefore, temperature management inside the printing nozzle is crucial to the quality of bioprinting, and the design of the cooling plate channel used to cool the nozzle is the core to achieving efficient heat dissipation and precise temperature control.

[0003] Current cooling plate topology designs often rely on experience or simplified geometry, making it difficult to balance heat transfer efficiency, flow resistance, and fabrication feasibility under complex operating conditions. Currently, deep learning models are frequently used to rapidly generate cooling plate topologies by learning the heat-flow-structure mapping relationship. However, existing discriminative modeling based on convolutional neural networks and graph neural networks lacks flexibility in structure generation and generalization ability across operating conditions. Meanwhile, generative methods based on generative adversarial networks and variational autoencoders face problems such as unstable convergence, insufficient sample diversity, and difficulty in incorporating physical constraints, resulting in low generation capability for existing deep learning models in generating cooling plate topologies. Summary of the Invention

[0004] In view of this, this application provides a model, method, apparatus, medium and computer for generating cooling plate topology, with the main purpose of solving the technical problem that existing deep learning models have low ability to generate cooling plate topology.

[0005] According to a first aspect of the present invention, a system is provided that includes a generative model of a cooling plate topology, the generative model of the cooling plate topology including a conditional control module, a convolution module, a diffusion model and a U-shaped Mamba model, wherein the U-shaped Mamba model is an encoder-decoder architecture;

[0006] The condition control module is used to acquire a preset structural requirement feature map and a physical field cloud map of multiple existing cooling plate topologies, and to stitch the structural requirement feature map and each of the physical field cloud maps into a condition feature map, and to perform feature encoding on the condition feature map to obtain a condition feature tensor.

[0007] The convolution module is used to perform convolution operations on the conditional feature tensor to obtain detailed conditional features and macroscopic conditional features.

[0008] The diffusion model is used to obtain the preset cooling plate operating condition features and the structural requirement feature map. Based on the cooling plate operating condition features, the detailed condition features, and the macroscopic condition features after cross-attention processing as spatial guidance information, the structural requirement feature map is iteratively denoised to obtain the cross-attention map and the multi-scale feature map.

[0009] The U-shaped Mamba model is used to concatenate the cross-attention map and the multi-scale feature map to obtain a topological feature tensor, and to model based on the topological feature tensor to generate the cooling plate topology.

[0010] Optionally, the convolution module performs convolution operations on the conditional feature tensor to obtain detailed conditional features and macroscopic conditional features, including: the convolution module performs a convolution operation on the conditional feature tensor based on a first convolution kernel with a first step amplitude to obtain the detailed conditional features; and performs a convolution operation on the conditional feature tensor based on a second convolution kernel with a second step amplitude to obtain the macroscopic conditional features; wherein the size of the first convolution kernel is smaller than that of the second convolution kernel, and the first step amplitude is larger than that of the second step amplitude.

[0011] Optionally, the method for determining the structural requirement feature map includes: obtaining a physical boundary matrix for characterizing the physical boundaries of the cooling plate topology, a flow channel boundary matrix for characterizing the positions of the inlet and outlet of the cooling plate topology, and a thermal domain matrix for characterizing the heat dissipation area of ​​the cooling plate topology; performing feature compression on the physical boundary matrix, the flow channel boundary matrix, and the thermal domain matrix respectively to obtain topological feature data, and performing latent space mapping on the topological feature data, using the obtained latent space image as the structural requirement feature map.

[0012] Optionally, the method for determining the physical field cloud map of the existing cooling plate topology includes: determining the initial temperature field cloud map, initial flow velocity field cloud map, initial pressure field cloud map, and initial volume fraction cloud map of the existing cooling plate topology based on finite element simulation; iterating the initial temperature field cloud map, initial flow velocity field cloud map, initial pressure field cloud map, and initial volume fraction cloud map of the existing cooling plate topology based on finite element simulation to obtain intermediate temperature field cloud map, intermediate flow velocity field cloud map, intermediate pressure field cloud map, and intermediate volume fraction cloud map after a preset number of iterations; and using the initial temperature field cloud map, initial flow velocity field cloud map, initial pressure field cloud map, initial volume fraction cloud map, intermediate temperature field cloud map, intermediate flow velocity field cloud map, intermediate pressure field cloud map, and intermediate volume fraction cloud map as the physical field cloud map of the existing cooling plate topology to characterize the physical constraint features of the existing cooling plate topology.

[0013] Optionally, the method for determining the operating condition characteristics of the cooling plate includes: obtaining the operating condition parameters of the cooling plate topology to be generated, wherein the operating condition parameters include at least one of the inlet flow rate information of the cooling plate topology, the thermal conductivity of the coolant, the density of the coolant, and the heat flux density of the cooling plate; and performing feature extraction on the operating condition parameters to obtain cooling plate operating condition characteristics that characterize the operating requirements of the cooling plate topology.

[0014] Optionally, the U-shaped Mamba model is used to model the cooling plate topology based on the topological feature tensor, including: the U-shaped Mamba model performs noise filtering on the topological feature tensor to obtain the noise-filtered topological feature tensor, so as to retain the features in the topological feature tensor that are associated with the physical constraint features and the cooling plate operating condition features; the topological feature tensor is subjected to latent space mapping to obtain a topological density map, and the cooling plate topology is generated based on the density value of each grid cell in the topological density map.

[0015] According to a second aspect of the present invention, a method for generating a cooling plate topology is provided, the method comprising:

[0016] Obtain the preset structural requirement feature map, cooling plate operating condition features, and physical field cloud map of multiple existing cooling plate topologies;

[0017] The structural requirement feature map, the cooling plate operating condition features, and the physical field cloud map of each existing cooling plate topology are input into the cooling plate topology generation model in the system described above to obtain the cooling plate topology output by the cooling plate topology generation model.

[0018] According to a third aspect of the present invention, an apparatus for generating a cooling plate topology is provided, the apparatus comprising:

[0019] The data acquisition module is used to acquire preset structural requirement feature maps, cooling plate operating condition features, and physical field cloud maps of multiple existing cooling plate topologies;

[0020] The topology generation module is used to input the structural requirement feature map, the cooling plate operating condition features, and the physical field cloud map of each existing cooling plate topology into the cooling plate topology generation model in the system described above, so as to obtain the cooling plate topology output by the cooling plate topology generation model.

[0021] According to a fourth aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method for generating the above-described cooling plate topology.

[0022] According to a fifth aspect of the present invention, a computer is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for generating a cooling plate topology.

[0023] This invention provides a cooling plate topology generation model, method, apparatus, medium, and computer. Through the collaborative design of a diffusion model and a Mamba model, it effectively addresses the low topology generation capability of existing deep learning models. Specifically, a conditional control module concatenates structural requirement feature maps with the physical field cloud map of the existing cooling plate and encodes them to obtain a conditional feature tensor. This tensor integrates structural requirements and physical data related to the cooling plate, overcoming the deficiency of traditional models in introducing physical constraints. Furthermore, a convolutional module extracts detailed and macroscopic conditional features from the conditional feature tensor, providing hierarchical feature support for subsequent topology generation. The flexibility of structure generation is enhanced; furthermore, the diffusion model combines the working condition features and hierarchical condition features processed by cross-attention to iteratively denoise the structural requirement feature map, avoiding the problems of unstable convergence of the generative adversarial network and insufficient sample diversity of the variational autoencoder; furthermore, the U-shaped Mamba model splices the cross-attention map and multi-scale feature map and models and generates the topology, strengthens the cross-scale feature association and long-range dependency modeling, enhances the model's cross-working condition generalization ability, and finally realizes an efficient cooling plate topology generation scheme that takes into account physical consistency, structural rationality and engineering adaptability, significantly improving the ability to generate cooling plate topologies.

[0024] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0026] Figure 1 A schematic diagram of a cooling plate provided in an embodiment of the present invention is shown;

[0027] Figure 2 This illustration shows one of the architectural schematic diagrams of a cooling plate topology generation model provided by an embodiment of the present invention;

[0028] Figure 3 This diagram illustrates a design schematic of a cooling plate topology provided by an embodiment of the present invention.

[0029] Figure 4This diagram illustrates a flow channel boundary matrix, physical boundary matrix, and thermal domain matrix provided in an embodiment of the present invention.

[0030] Figure 5 The illustration shows a schematic diagram of an initial volume fraction cloud map, an initial temperature field cloud map, an initial flow velocity field cloud map, and an initial pressure field cloud map provided by an embodiment of the present invention;

[0031] Figure 6 This diagram illustrates an intermediate volume fraction cloud map, an intermediate temperature field cloud map, an intermediate flow velocity field cloud map, and an intermediate pressure field cloud map provided by an embodiment of the present invention.

[0032] Figure 7 This is a second schematic diagram of the architecture of a generation model of a cooling plate topology provided by an embodiment of the present invention;

[0033] Figure 8 A flowchart illustrating a method for generating a cooling plate topology according to an embodiment of the present invention is shown.

[0034] Figure 9 A schematic diagram of a cooling plate topology generation device provided in an embodiment of the present invention is shown. Detailed Implementation

[0035] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0036] Currently, bioprinting technology shows significant potential in tissue engineering and regenerative medicine. Extrusion printing, due to its simple process and wide material adaptability, has become the mainstream 3D printing method. However, the medical hydrogel materials used in printing are highly temperature sensitive, and their rheological properties depend on precise thermal environment control. Insufficient temperature control can damage the fluidity and forming stability of the hydrogel, directly affecting the geometric accuracy and biological function of the printed structure. Therefore, temperature management inside the printing nozzle is crucial to the quality of bioprinting, and the design of the cooling plate channel used to cool the nozzle is the core to achieving efficient heat dissipation and precise temperature control.

[0037] Here, cooling plate topology refers to the material distribution and geometric layout of the internal flow channels and solid regions of the cooling plate; as an example, such as Figure 1 As shown, the cooling plate topology includes a coolant inlet, a coolant outlet, a coolant channel, and a substrate. After obtaining the cooling plate topology, the actual cooling plate can be generated by 3D printing the specific structure of the cooling plate topology.

[0038] Currently, the design of existing cooling plate topologies largely relies on experience or simplified geometry, making it difficult to balance heat transfer efficiency, flow resistance, and fabrication feasibility under complex operating conditions. Current methods often utilize deep learning models to rapidly generate cooling plate topologies by learning the heat-fluid-structure mapping relationship. However, existing discriminative modeling based on convolutional neural networks and graph neural networks lacks flexibility in structure generation and generalization ability across operating conditions. Meanwhile, generative methods based on generative adversarial networks and variational autoencoders face problems such as unstable convergence, insufficient sample diversity, and difficulty in incorporating physical constraints, resulting in low generation capability for existing deep learning models in generating cooling plate topologies.

[0039] To address the above problems, in one embodiment, such as Figure 2 As shown, a system including a generative model of a cooling plate topology is provided. Taking the application of the model in this system to generate a cooling plate topology for cooling the nozzle of a bio-3D printer as an example, the generative model of the cooling plate topology includes a conditional control module 100, a convolution module 200, a diffusion model 300, and a U-shaped Mamba model 400. Among them, the diffusion model 300 can be a latent space diffusion model; the U-shaped Mamba model 400 is a composite model built with the Mamba state space architecture as its core. It adopts the encoder-decoder hierarchical structure of U-Net, and its core feature processing units are all built based on Mamba's selective state space model (SSM). Here, by leveraging Mamba's linear complexity sequence modeling and selective memory capabilities, the U-shaped Mamba model 400 can efficiently capture cross-scale feature associations, while maintaining the advantages of multi-scale processing of the U-shaped structure and strengthening the long-range dependency modeling capability.

[0040] Furthermore, the condition control module 100 (ControlNet) is a module that implements multimodal condition guidance by adding a trainable replica structure to the denoised U-Net of the diffusion model 300. In this scheme, it injects the conditions after splicing and encoding the structural requirement feature map and physical field cloud map of the cooling plate into the cross attention module and convolution replica of the diffusion model 300. Combined with zero convolution, it avoids early noise interference, ensures that the topology generation of the liquid cooling plate meets physical constraints such as heat flow path and boundary heat flux, and improves the physical consistency of the generated structure.

[0041] Specifically, the condition control module 100 is used to acquire a preset structural requirement feature map and a physical field cloud map of multiple existing cooling plate topologies, and to stitch the structural requirement feature map and each of the physical field cloud maps into a condition feature map, and to perform feature encoding on the condition feature map to obtain a condition feature tensor.

[0042] Here, when generating the structural requirement feature map, a physical boundary matrix representing the physical boundaries of the cooling plate topology, a flow channel boundary matrix representing the locations of the inlet and outlet of the cooling plate topology, and a thermal domain matrix representing the heat dissipation area of ​​the cooling plate topology can be obtained; here, as... Figure 3 As shown, a design drawing (top view) of the required cooling plate topology can be pre-designed, and the symmetry boundaries and non-slip boundaries of the cooling plate topology can be marked. Here, L represents the length of the cooling plate topology, w represents the width of the cooling plate topology, and d represents the width of the inlet and outlet. Indicates the inlet velocity. This indicates export pressure.

[0043] Furthermore, the rectangular outer contour and inlet / outlet structures in the design drawing can be discretized into a mesh, and the boundary positions can be marked to form a physical boundary matrix. Furthermore, the inlet, outlet, and internal flow channel regions in the design drawing can be identified, generating a flow channel boundary matrix representing the flow channel boundaries. Furthermore, based on a preset heat dissipation area range, areas requiring enhanced heat dissipation can be marked in the design drawing, and the heat domain positions can be discretized through a mesh to form a heat domain matrix representing the heat domain distribution. Further, such as... Figure 4 As shown, in the flow channel boundary matrix 21, the positions of the inlet and outlet of the cooling plate topology can be marked with the value "1", and the range of the aluminum alloy substrate in the cooling plate topology can be marked with "0". In the physical boundary matrix 22, the position of the inlet of the cooling plate topology can be marked with the value "1", the position of the outlet can be marked with the value "-1", the non-slip wall boundary of the cooling plate topology can be marked with the value "3", and the position where flow channels can be set in the cooling plate topology can be marked with the value "0". Furthermore, in the thermal domain matrix 23, the heat-loaded area that needs to dissipate heat can be marked with the value "1", and the adiabatic boundary that does not need to dissipate heat can be marked with the value "0". Thus, the flow channel boundary matrix 21, physical boundary matrix 22, and thermal domain matrix 23 are obtained. Here, the flow channel boundary matrix 21, physical boundary matrix 22, and thermal domain matrix 23 can be preset according to the geometry of the cooling plate topology to be generated, so as to reflect the physical structure requirements of the cooling plate topology.

[0044] Furthermore, feature compression is performed on the physical boundary matrix, the flow channel boundary matrix, and the thermal domain matrix respectively to obtain topological structure feature data. Then, latent space mapping is performed on the topological structure feature data, and the obtained latent space image is used as the structural requirement feature map.

[0045] Specifically, firstly, the discrete parameters of the physical boundary matrix, flow channel boundary matrix, and thermal domain matrix can be transformed into continuous features perceptible to the model through tensor mapping. Then, dimensionality compression and feature extraction are performed on these matrices to retain core topological features such as the geometric contours of the physical boundaries, the connectivity of the flow channel boundaries, and the hot-loaded areas of the thermal domain, resulting in topological structure feature data. Subsequently, latent space mapping is performed on this topological structure feature data, converting it into a latent space image as a feature map representing structural requirements. Here, the latent space mapping process can be understood as projecting high-dimensional topological structure feature data into a low-dimensional latent space through specific parametric transformations, forming a visualized latent space image. This latent space image can serve as a clear structural constraint in the subsequent topology generation process, guiding the generation direction of the cooling plate topology and ensuring that the generated topology conforms to preset structural design specifications such as physical boundary constraints, flow channel layout requirements, and thermal domain heat dissipation requirements.

[0046] Furthermore, regarding the physical field cloud map, it can be determined that the existing cooling plate topology is based on finite element simulation, and the initial temperature field cloud map, initial flow velocity field cloud map, initial pressure field cloud map, and initial volume fraction cloud map of the cooling plate topology are simulated.

[0047] Specifically, multiple existing cooling plate topologies can be acquired, and finite element simulations can be used to simulate the velocity field, pressure field, temperature field, and volume fraction field of the coolant flow for each topology. Here, for each cooling plate topology, finite element simulations are used to generate initial velocity field contour maps, initial temperature field contour maps, initial pressure field contour maps, and initial volume fraction field contour maps to characterize the velocity field, temperature field, pressure field, and volume fraction field.

[0048] Furthermore, based on finite element simulation, the initial temperature field cloud map, the initial velocity field cloud map, the initial pressure field cloud map, and the initial volume fraction cloud map are iterated to obtain intermediate temperature field cloud map, intermediate velocity field cloud map, intermediate pressure field cloud map, and intermediate volume fraction cloud map after a preset number of iterations.

[0049] Specifically, finite element iterations are performed on the initial temperature field contour maps, initial velocity field contour maps, initial pressure field contour maps, and initial volume fraction contour maps to obtain the multi-stage physical field characteristics of the cooling plate topology during the dynamic heat-fluid coupling process. Here, finite element simulation software (such as ANSYS, multiphysics simulation software COMSOL, etc.) can be used to iteratively calculate the initial temperature field contour maps, initial velocity field contour maps, initial pressure field contour maps, and initial volume fraction contour maps based on physical models such as the heat conduction equation and the Navier-Stokes equation. In each iteration, the heat transfer and fluid flow state within the cooling plate are updated according to the current physical field distribution, thereby obtaining the temperature, velocity, pressure, and volume fraction distribution contour maps for the next stage. By setting a preset number of iterations, intermediate physical field cloud maps are finally obtained, including intermediate temperature field cloud maps, intermediate flow velocity field cloud maps, intermediate pressure field cloud maps, and intermediate volume fraction cloud maps. Here, the obtained intermediate physical field cloud maps can comprehensively characterize the physical constraint characteristics of the cooling plate topology at different working stages, providing rich historical physical evolution information for subsequent topology generation work, and ensuring that the generated cooling plate topology can meet the thermal flow performance requirements throughout the entire working cycle.

[0050] Furthermore, the initial temperature field cloud map, the initial flow velocity field cloud map, the initial pressure field cloud map, the initial volume fraction cloud map, the intermediate temperature field cloud map, the intermediate flow velocity field cloud map, the intermediate pressure field cloud map, and the intermediate volume fraction cloud map are used as the physical field cloud maps of the existing cooling plate topology to characterize the physical constraint features of the existing cooling plate topology.

[0051] Specifically, such as Figure 5 and Figure 6 As shown, the initial volume fraction cloud map 31, initial temperature field cloud map 32, initial flow velocity field cloud map 33, and initial pressure field cloud map 34 of the initial stage, as well as the intermediate volume fraction cloud map 35, intermediate temperature field cloud map 36, intermediate flow velocity field cloud map 37, and intermediate pressure field cloud map 38 after a preset number of iterations, are used together as the physical field cloud map of the existing cooling plate topology. This can comprehensively characterize the physical constraint characteristics of the existing cooling plate topology and present the thermal field, flow field, pressure field, and fluid volume distribution of the cooling plate during the working process from multiple dimensions and stages, covering the physical information of the cooling plate from the initial state to the intermediate dynamic process.

[0052] Furthermore, the initial temperature field cloud map, initial velocity field cloud map, initial pressure field cloud map, initial volume fraction cloud map, intermediate temperature field cloud map, intermediate velocity field cloud map, intermediate pressure field cloud map, and intermediate volume fraction cloud map can be obtained for each existing cooling plate topology. Here, by integrating these physical field cloud maps, rich and accurate physical constraint references can be provided for the subsequent generation of cooling plate topologies, enabling the model to obtain the thermodynamic parameters of different flow channels to design the flow channels of the cooling plate topology, ensuring that the newly generated cooling plate topology can meet the physical performance requirements of actual engineering in terms of heat transfer efficiency, fluid flow characteristics, and pressure loss control.

[0053] Furthermore, such as Figure 2 As shown, the condition control module 100 can concatenate the structural requirement feature map and each physical field cloud map into a condition feature map, and perform feature encoding on the condition feature map to obtain a condition feature tensor. Specifically, the condition control module 100 first concatenates the preset structural requirement feature map with the physical field cloud maps of multiple existing cooling plate topologies according to channel or spatial dimensions to form a condition feature map that integrates structural constraints and physical constraints. Subsequently, the condition control module 100 uses the feature encoding network (such as an encoding structure containing convolutional layers and attention layers) within the ControlNet architecture to extract and compress features from the condition feature map, converting it into a high-dimensional and semantically rich condition feature tensor. Here, through the condition injection mechanism of the ControlNet architecture, it is ensured that the structural requirements represented by the structural requirement feature map and the physical field information represented by the physical field cloud map are effectively encoded and transmitted to subsequent modules, providing accurate multimodal condition guidance for the generation of cooling plate topologies.

[0054] Furthermore, such as Figure 2 As shown, the convolution module 200 is used to perform convolution operations on the conditional feature tensor to obtain detailed conditional features and macroscopic conditional features. Specifically, the convolution module 200 performs a convolution operation on the conditional feature tensor based on a first convolution kernel with a first step amplitude to obtain the detailed conditional features, and performs a convolution operation on the conditional feature tensor based on a second convolution kernel with a second step amplitude to obtain the macroscopic conditional features; wherein, the size of the first convolution kernel is smaller than that of the second convolution kernel, and the first step amplitude is larger than that of the second step amplitude.

[0055] Here, the conditional feature tensor integrates the geometric constraints of the structural requirement feature map with the thermal-fluid physical features of the physical field cloud map. It can be used to separate key information of different granularities through convolution operations. For detailed conditional feature extraction, a smaller first convolutional kernel (e.g., a 3×3 kernel) is used. Its receptive field is limited to a local region, accurately capturing fine features such as flow channel edge contours and small thermal domain boundaries. Combined with a larger first stride, computational efficiency is optimized while maintaining the accuracy of detail extraction. For macroscopic conditional feature extraction, a larger second convolutional kernel (e.g., a 5×5 kernel) is used. Its larger receptive field aggregates global information, effectively capturing macroscopic features such as the overall flow channel layout and large-scale temperature gradient distribution. Combined with a smaller second stride, the integrity and spatial correlation of macroscopic features are ensured.

[0056] Furthermore, the diffusion model 300 is used to obtain preset cooling plate operating condition features and the structural requirement feature map, and based on the cooling plate operating condition features, the detailed condition features, and the macroscopic condition features after cross-attention processing as spatial guidance information, iteratively denoises the structural requirement feature map to obtain a cross-attention map and a multi-scale feature map.

[0057] The method for determining the operating characteristics of the cooling plate includes: First, obtaining the operating condition parameters of the cooling plate topology to be generated, wherein the operating condition parameters include at least one of the following: inlet flow velocity information of the cooling plate topology, coolant thermal conductivity, coolant density, and cooling plate heat flux density. Here, the inlet flow velocity requirements of the required cooling plate topology can be pre-designed to obtain the inlet flow velocity information, and the coolant thermal conductivity to be used in the cooling plate can be determined. Furthermore, the amount of heat passing through a unit area of ​​the cooling plate per unit time can be pre-designed as the cooling plate heat flux density according to actual needs to reflect the required heat dissipation intensity of the cooling plate. Then, feature extraction is performed on the operating condition parameters to obtain the cooling plate operating characteristics used to characterize the operating requirements of the cooling plate topology. Specifically, by using feature extraction networks with structures such as fully connected layers and embedded layers, the working condition parameters are numerically encoded and semantically mapped to extract key features that reflect the flow rate of the cooling plate inlet, the thermal conductivity of the coolant, the density of the coolant, and the heat flux density of the cooling plate. The resulting features are presented as feature tensors, providing precise guidance for subsequent topology generation and ensuring that the generated topology is adapted to the target working scenario.

[0058] Furthermore, the diffusion model 300 introduces an effective cross-attention layer in the denoising U-Net. This layer processes the cooling plate operating condition features and structural requirement feature maps using cross-attention, deeply fusing them with the local fine structural information represented by detailed condition features and the global layout information represented by macroscopic condition features. This results in spatial guidance information that reflects operating condition constraints and multi-scale structural guidance. Further, the diffusion model 300 can first add Gaussian noise to the structural requirement feature map. Then, based on the spatial guidance information, the diffusion model 300 performs an iterative denoising process on the structural requirement feature map. In this process, starting from the noisy initial feature map, each step gradually eliminates noise based on the guidance information, while simultaneously strengthening features that meet the operating condition and structural requirements. The final output is a cross-attention map that fuses the operating condition and structural relationships, and a multi-scale feature map containing hierarchical topological features from micro to macro levels.

[0059] Furthermore, the U-shaped Mamba model 400 is used to concatenate the cross-attention map and the multi-scale feature map to obtain a topological feature tensor, and to model based on the topological feature tensor to generate the cooling plate topology. Here, the U-shaped Mamba model 400 has an encoder-decoder architecture of Unified Semantic Mamba (USM), which can achieve topology generation with both physical consistency and engineering manufacturability through a progressive process of feature concatenation, noise filtering, and latent space mapping. Specifically, the U-shaped Mamba model 400 can first receive the cross-attention map and multi-scale feature map output by the diffusion model, and then fuse the two types of features into a topological feature tensor with unified dimensions through a channel-level concatenation operation.

[0060] Here, we assume the dimensions of the cross-attention map are: × × ,in , These represent the spatial height and width of the cross-attention map, respectively. The number of attention weight channels; furthermore, the multi-scale feature map contains multi-level features at both micro and macro levels, with the micro-level feature map having the following dimensions: × × ,in , These represent the spatial height and width of the micro-level feature map, respectively. The number of channels in the detail layer of the multi-scale feature map; the dimension of the macro-level feature map is: × × ,in , These represent the spatial height and width of the macroscopic hierarchical feature map, respectively. This represents the number of channels in the macroscopic layer of the multi-scale feature map. Further, using the downsampling or upsampling module of the USM encoder, a convolutional layer with a kernel size of 2×2 and a stride of 2 is employed to unify the spatial dimensions of the cross-attention map and the multi-scale feature map to match the input dimension requirements of the U-shaped Mamba model 400. Furthermore, the two types of features are fused using a channel-dimensional concatenation method. The cross-attention map with unified dimensions and the features at each level of the multi-scale feature map are superimposed along the channel direction to obtain the topological feature tensor.

[0061] Here, the weight information of the cross attention map is associated with the structural information of the multi-scale feature map. The heat source corresponding to the region with higher attention weight will be deeply bound to the flow channel details and temperature gradient features of that region in the multi-scale feature map. Key features such as reasonable flow channel structure of the region with high heat dissipation demand will be retained first in the topological feature tensor.

[0062] Furthermore, after obtaining the topological feature tensor, the U-shaped Mamba model 400 performs noise filtering on the topological feature tensor to obtain the noise-filtered topological feature tensor, so as to retain the features in the topological feature tensor that are associated with the physical constraint features and the cooling plate operating conditions. Specifically, the U-shaped Mamba model 400 can be constructed based on its main block structure of the USM, with built-in time step normalization (AdaLN) and Mamba Block units, to perform targeted noise filtering on the topological feature tensor, and through the state space modeling capability of the Mamba Block, to capture the effective features in the feature tensor that are strongly associated with the physical constraint features related to the continuity of the temperature field and the smoothness of the pressure field, as well as the cooling plate operating conditions related to the inlet flow rate and coolant characteristics.

[0063] Furthermore, latent space mapping is performed on the topological feature tensor to obtain a topological density map. Based on the density value of each grid cell in the topological density map, a model is created to generate the cooling plate topology. Specifically, the U-shaped Mamba model 400 can obtain the noise-filtered topological feature tensor based on its activation function module. Through latent space mapping, the high-dimensional topological feature tensor is projected onto the low-dimensional topological density map, obtaining the density value of each grid cell in the topological density map (the density value ranges from [0, 1]). Grids with density values ​​approaching 1 correspond to coolant flow channels, and grids with density values ​​approaching 0 correspond to solid heat-conducting regions, thus achieving a quantitative representation of the structural morphology. Further, the model performs binarization based on the density values ​​to clarify the boundaries between the flow channels and solid regions, ultimately generating a cooling plate topology with clear geometric contours and good flow channel connectivity to meet the cooling requirements of the bio-3D printing nozzle for the liquid cooling plate.

[0064] The cooling plate topology generation model provided in this embodiment, based on the TO-MambaLDM architecture model, combines topology optimization design priors with deep generative modeling. It encodes the temperature field, velocity field, and boundary conditions through a multimodal condition injection mechanism, achieving efficient modeling of heat-fluid-structure coupling constraints. It performs diffusion generation of the topology in the latent space, ensuring that the generated results have high-fidelity representation in terms of geometric details and physical consistency. At the same time, the Mamba module is introduced to enhance the model's ability to model long-range dependencies and cross-scale features, effectively improving the connectivity and engineering adaptability of the topology.

[0065] Furthermore, in combination Figure 7 The working process of the cooling plate topology generation model is explained. In the actual working process, a sub-network 80 can be generated based on the preset structural requirement feature map. The input matrix 71, which is composed of the preset physical boundary matrix, flow channel boundary matrix and thermal domain matrix, is compressed to obtain topological structure feature data. The topological structure feature data is then mapped to the latent space, and the resulting latent space image is used as the structural requirement feature map 72. Here, the structural requirement feature map generation sub-network 80 can be a preset convolutional neural network model.

[0066] Furthermore, multiple physical field cloud maps 73, including initial temperature field cloud maps, initial flow velocity field cloud maps, initial pressure field cloud maps, initial volume fraction cloud maps, intermediate temperature field cloud maps, intermediate flow velocity field cloud maps, intermediate pressure field cloud maps, and intermediate volume fraction cloud maps, are stitched together with the structural requirement feature map 72 to obtain a conditional feature map. The conditional feature map is then feature-encoded based on the conditional control module 100 to obtain a conditional feature tensor. Simultaneously, the operating condition parameters 74 of the cooling plate topology to be generated are obtained, and feature extraction is performed on these parameters to obtain cooling plate operating condition features 75 characterizing the operating conditions of the cooling plate topology.

[0067] Furthermore, the convolution module 200 performs convolution operations on the conditional feature tensor based on different convolution kernels to obtain detailed conditional features and macroscopic conditional features. Further, the diffusion model 300 obtains the cooling plate operating condition features 75 and the structural requirement feature map 72, and uses the cooling plate operating condition features 75 (processed with cross-attention), the detailed conditional features, and the macroscopic conditional features as spatial guidance information to iteratively denoise the structural requirement feature map 72, obtaining a cross-attention map 76 and a multi-scale feature map 77. Further, the U-shaped Mamba model 400 concatenates the cross-attention map 76 and the multi-scale feature map 77 to obtain a topological feature tensor 78, and models are performed based on the topological feature tensor 78 to generate the cooling plate topology 79.

[0068] Furthermore, the training method for the generative model of the cooling plate topology can be as follows: First, acquire multiple model training datasets; each model training dataset contains a preset structural requirement feature map 72 as model input data, physical field cloud maps 73 of multiple existing cooling plate topologies, and cooling plate operating condition features 75, as well as a target cooling plate topology as model output data. Here, for the physical field cloud map 73 of each existing cooling plate topology, multi-stage temperature field, flow velocity field, and other cloud maps can be obtained based on finite element simulation to construct a physical field database for training the condition control module 100.

[0069] Subsequently, the structural requirement feature map 72 and the physical field cloud map 73 are input into the condition control module 100, and the structural requirement feature map 72, the physical field cloud map 73 and the cooling plate working condition feature 75 are input into the diffusion model 300 for collaborative training. By fusing multimodal conditions, the denoising U-Net of the diffusion model 300 is trained so that its output cross-attention map 76 and multi-scale feature map 77 can accurately reflect physical constraints and structural requirements.

[0070] Specifically, the structural requirement feature map 72 and the physical field cloud map 73 of the existing cooling plate can be used as training data. First, the structural requirement feature map 72 and the physical field cloud map 73 are concatenated to form a conditional feature map, which is then input into the encoding network of the conditional control module 100 for training. Here, through supervised learning, the conditional control module 100 learns the fusion law of structural constraints and physical field information. The difference between the generated conditional feature tensor and the real physical-structural correlation features is used as the loss to iteratively optimize the network parameters, ensuring that it can accurately encode multimodal conditional information. Further, the diffusion model 300 is trained with the noisy structural requirement feature map 72 as the initial input, combined with the conditional feature tensor output by the conditional control module 100 and the cooling plate operating condition features processed by cross-attention. Here, a loss function is constructed by calculating the difference between the feature maps before and after denoising, enabling the network of the diffusion model 300 to learn the law of gradually eliminating noise and strengthening effective topological features under multi-condition guidance, and finally outputting a denoised result containing a cross-attention map 76 and a multi-scale feature map 77. Furthermore, the condition control module 100 and the diffusion model 300 are trained collaboratively. Specifically, the condition control module 100 and the diffusion model 300 are connected in series and trained collaboratively in an end-to-end manner. The input structure requirement feature map 72, the physical field cloud map 73, and the cooling plate working condition feature 75 enable the condition control module 100 to generate a condition feature tensor. The diffusion model 300 performs denoising processing based on this tensor. The generated cross-attention map 76 and multi-scale feature map 77 are matched with the features of the real target cooling plate topology, and the physical performance of the topology generated by the subsequent U-shaped Mamba model 400 is used as the joint loss. At the same time, the parameters of the two modules are optimized to ensure their synergy in feature transfer and condition guidance. Finally, the diffusion model 300 can output a high-quality feature map that supports topology generation under the precise condition injection of the condition control module 100.

[0071] Simultaneously, the parameter prediction module and activation function module included in the U-shaped Mamba model 400 are also trained concurrently. The parameter prediction module learns to filter out topological feature tensor noise and retain effective features associated with physical and operational characteristics, while the activation function module learns latent space mapping to generate a topological density map. During model training, the physical performance and geometric consistency between the model-generated cooling plate topology and the real target cooling plate topology are used as the loss function to iteratively optimize the parameters of each module, ultimately enabling the model to stably output a cooling plate topology that conforms to thermal flow performance and structural constraints.

[0072] Furthermore, embodiments of this application also provide a method for generating a cooling plate topology, applied to a computer, such as... Figure 8 As shown, the method includes:

[0073] 101. Obtain the preset structural requirement feature map, cooling plate operating condition features, and physical field cloud map of multiple existing cooling plate topologies.

[0074] Here, the methods for obtaining the structural requirement feature map, cooling plate operating condition characteristics, and physical field cloud map of multiple existing cooling plate topologies are as described above, and will not be repeated here.

[0075] 102. Input the structural requirement feature map, the cooling plate operating condition feature, and the physical field cloud map of each existing cooling plate topology into the cooling plate topology generation model to obtain the cooling plate topology output by the cooling plate topology generation model.

[0076] The cooling plate topology generation method provided in this embodiment directly calls the generation model of the cooling plate topology, integrates structural requirement feature maps, operating condition features and existing physical field cloud maps, and can efficiently generate cooling plate topologies without additional complex processes. It can leverage the advantages of physical constraint integration and cross-scale feature processing of the model to ensure that the generated topology takes into account both thermal flow performance and manufacturability. At the same time, it simplifies the generation process, reduces operational complexity, and can adapt to the needs of rapid and accurate design of cooling plates in scenarios such as bio-3D printing nozzles, thereby improving the efficiency of engineering applications.

[0077] Furthermore, as Figure 8 The specific implementation of the method shown in this embodiment provides a device for generating a cooling plate topology, such as... Figure 9 As shown, the device includes a data acquisition module 91 and a topology generation module 92.

[0078] The data acquisition module 91 can be used to acquire preset structural requirement feature maps, cooling plate operating condition features, and physical field cloud maps of multiple existing cooling plate topologies.

[0079] The topology generation module 92 can be used to input the structural requirement feature map, the cooling plate operating condition features, and the physical field cloud map of each existing cooling plate topology into the cooling plate topology generation model to obtain the cooling plate topology output by the cooling plate topology generation model.

[0080] It should be noted that other corresponding descriptions of the functional units involved in the cooling plate topology generation apparatus provided in this embodiment can be found in [reference needed]. Figure 8 The corresponding descriptions in [the document] will not be repeated here.

[0081] Based on the above, Figure 8 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 8 The method for generating the cooling plate topology shown.

[0082] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer (such as a personal computer, server, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0083] Based on the above, Figure 8 The method shown, and Figure 9 The illustrated embodiment of the cooling plate topology generation apparatus, in order to achieve the above objectives, also provides a physical device for generating the cooling plate topology. Specifically, this device can be a personal computer, server, smartphone, tablet computer, smartwatch, or other network device, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-described... Figure 8 The method shown.

[0084] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0085] Those skilled in the art will understand that the physical device structure generated by the cooling plate topology provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0086] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0087] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0088] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A system comprising a generated model of a cold plate topology, characterized by, The generated model of the cooling plate topology includes a conditional control module, a convolution module, a diffusion model, and a U-shaped Mamba model, wherein the U-shaped Mamba model is an encoder-decoder architecture; The conditional control module is configured to obtain a preset structure requirement feature map and a plurality of physical field cloud maps of existing cooling plate topologies, splice the structure requirement feature map and each physical field cloud map into a conditional feature map, and perform feature coding on the conditional feature map to obtain a conditional feature tensor; The convolution module is configured to perform convolution operation on the conditional feature tensor to obtain detailed conditional features and macro conditional features; The diffusion model is configured to obtain a preset cooling plate working condition feature and the structure requirement feature map, and perform iterative denoising on the structure requirement feature map based on the cooling plate working condition feature processed by cross-attention, the detailed conditional features, and the macro conditional features as spatial guidance information to obtain a cross-attention map and a multi-scale feature map; The U-shaped Mamba model is configured to splice the cross-attention map and the multi-scale feature map to obtain a topology feature tensor, and perform modeling based on the topology feature tensor to generate a cooling plate topology.

2. The system of claim 1, wherein the generated model comprising a cold plate topology is characterized by, The way in which the convolution module performs convolution operation on the conditional feature tensor to obtain detailed conditional features and macro conditional features includes: The convolution module performs convolution operation on the conditional feature tensor based on a first convolution kernel with a first step size to obtain the detailed conditional features; The convolution module performs convolution operation on the conditional feature tensor based on a second convolution kernel with a second step size to obtain the macro conditional features; Wherein, the size of the first convolution kernel is smaller than the size of the second convolution kernel, and the first step size is higher than the second step size.

3. The system of claim 1, wherein the generated model comprising a cold plate topology is further characterized by, The way in which the structure requirement feature map is determined includes: Obtaining a physical boundary matrix for representing the physical boundary of the cooling plate topology, a flow channel boundary matrix for representing the inlet and outlet positions of the cooling plate topology, and a thermal domain matrix for representing the heat dissipation area range of the cooling plate topology; Performing feature compression on the physical boundary matrix, the flow channel boundary matrix, and the thermal domain matrix respectively to obtain topology structure feature data, and performing latent space mapping on the topology structure feature data to obtain a latent space image as the structure requirement feature map.

4. The system of claim 1, wherein the generated model comprising a cold plate topology is further characterized by, The way in which the physical field cloud map of the existing cooling plate topology is determined includes: Determining the initial temperature field cloud map, the initial flow velocity field cloud map, the initial pressure field cloud map, and the initial volume fraction cloud map of the cooling plate topology simulated based on finite element simulation of the existing cooling plate topology; Performing iteration on the initial temperature field cloud map, the initial flow velocity field cloud map, the initial pressure field cloud map, and the initial volume fraction cloud map based on finite element simulation to obtain the intermediate temperature field cloud map, the intermediate flow velocity field cloud map, the intermediate pressure field cloud map, and the intermediate volume fraction cloud map after iteration for a preset number of times; The initial temperature field cloud map, the initial flow velocity field cloud map, the initial pressure field cloud map, the initial volume fraction cloud map, the intermediate temperature field cloud map, the intermediate flow velocity field cloud map, the intermediate pressure field cloud map, and the intermediate volume fraction cloud map are taken as the physical field cloud maps of the existing cooling plate topology, and are used to characterize the physical constraint features of the existing cooling plate topology.

5. The system of claim 4, wherein the generated model comprising a cold plate topology is characterized by, The determination manner of the cooling plate working condition feature comprises: obtaining working condition parameters of a cooling plate topology to be generated, wherein the working condition parameters comprise at least one of liquid inlet flow velocity information, cooling liquid thermal conductivity, cooling liquid density, and cooling plate heat flux density of the cooling plate topology; performing feature extraction on the working condition parameters to obtain a cooling plate working condition feature used to characterize working condition requirements of the cooling plate topology.

6. The system of claim 5, wherein the generated model comprising a cold plate topology is characterized by, The U-shaped manba model is based on the topology feature tensor to model and generate a cooling plate topology, comprising: The U-shaped manba model performs noise filtering processing on the topology feature tensor to obtain the topology feature tensor after noise filtering processing, so as to retain features in the topology feature tensor associated with the physical constraint features and the cooling plate working condition features; performing latent space mapping on the topology feature tensor to obtain a topology density map, and modeling based on a density value of each grid cell in the topology density map to generate the cooling plate topology.

7. A method of generating a cooling plate topology, characterized by, The method comprises: obtaining a preset structure requirement feature map, a cooling plate working condition feature, and physical field cloud maps of a plurality of existing cooling plate topologies; inputting the structure requirement feature map, the cooling plate working condition feature, and the physical field cloud map of each existing cooling plate topology into a cooling plate topology generation model in the system of any one of claims 1 to 6 to obtain a cooling plate topology generated by the cooling plate topology generation model.

8. A cooling plate topology generation apparatus, characterized by, The device comprises: a data acquisition module configured to obtain a preset structure requirement feature map, a cooling plate working condition feature, and physical field cloud maps of a plurality of existing cooling plate topologies; a topology generation module configured to input the structure requirement feature map, the cooling plate working condition feature, and the physical field cloud map of each existing cooling plate topology into a cooling plate topology generation model in the system of any one of claims 1 to 6 to obtain a cooling plate topology generated by the cooling plate topology generation model.

9. A storage medium having stored thereon a computer program, characterized in that The computer program, when executed by a processor, implements the steps of the method of claim 7.

10. A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program, when executed by a processor, implements the steps of the method of claim 7.

Images