A method and system for designing and optimizing a cylinder block reinforcement structure based on generative AI

Abstract

The application relates to the technical field of mechanical mechanism design, in particular to a cylinder block reinforcing rib structure design optimization method and system based on generative AI, which comprises the following steps: constructing a data set; training a conditional diffusion model with design constraints as conditions; obtaining target design constraints and inputting the target design constraints into the trained conditional diffusion model; generating a plurality of different reinforcing rib layout probability maps through random sampling, and performing binary processing on the probability maps to obtain a plurality of candidate layout schemes; performing rapid performance evaluation on each candidate layout scheme by using a pre-trained performance prediction agent model, and screening candidate schemes meeting preset performance constraints according to the evaluation results; and performing finite element analysis review on the screened candidate schemes to determine an optimal reinforcing rib layout scheme. The application can automatically and efficiently generate diversified cylinder block reinforcing rib layout candidate schemes according to design constraints.

Description

Technical Field

[0001] This application relates to the field of mechanical mechanism design technology, and in particular to a method and system for optimizing the design of cylinder block reinforcing rib structure based on generative AI. Background Technology

[0002] The cylinder block is the core load-bearing component of the engine, and its internal reinforcing rib design directly affects the overall rigidity, vibration and noise levels, and weight of the engine. A reasonable reinforcing rib layout can effectively reduce weight while ensuring structural strength and improving the engine's NVH (noise, vibration, and harshness) performance.

[0003] Currently, traditional stiffener layout design mainly relies on engineers' design experience and finite element iterative optimization. Engineers initially design the direction and location of the stiffeners based on experience, then perform simulation analysis using finite element software. Based on the analysis results, the model is manually modified and simulation is repeated until the solution meets the requirements. This traditional method suffers from a long design cycle because each round of modification requires re-modeling and finite element simulation, with each iteration often taking hours or even days. Furthermore, the manually defined topology optimization region is limited by the engineer's habitual thinking, making it difficult to explore unexpected, unconventional, high-performance layout configurations and easily leading to local optima. In addition, stiffener design needs to balance multiple conflicting objectives such as stiffness, weight, and casting processability. Manually finding multi-objective optimal solutions is extremely difficult, usually only yielding compromise solutions and failing to achieve global optima.

[0004] In recent years, generative artificial intelligence technologies, represented by generative adversarial networks and diffusion models, have demonstrated powerful capabilities in areas such as 2D image generation and 3D structural topology optimization, enabling them to learn complex distributions from large amounts of data and generate novel samples. However, these technologies have not yet been applied to the automated design of complex 3D structures such as cylinder block reinforcing ribs. How to effectively combine generative AI with cylinder block reinforcing rib design to automatically generate diverse layout schemes that meet multiple objective constraints is a pressing technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing the design of cylinder block reinforcing rib structure based on generative AI, so as to solve the problems of traditional design methods in the prior art relying on experience iteration, having a long cycle, and being difficult to explore the global optimal layout.

[0006] This invention provides the following solution:

[0007] A method for optimizing the design of cylinder block reinforcing rib structures based on generative AI, comprising:

[0008] Construct a dataset containing three-dimensional voxel models, binarized stiffener layouts, and corresponding performance parameters of multiple cylinder block samples;

[0009] The conditional diffusion model is trained using design constraints as conditions.

[0010] Obtain the target design constraints and input them into the trained conditional diffusion model. Generate multiple different stiffener layout probability maps through random sampling, and perform binarization on the probability maps to obtain multiple candidate layout schemes.

[0011] Using a pre-trained performance prediction proxy model, the performance of each candidate layout scheme is quickly evaluated, and candidate schemes that meet preset performance constraints are selected based on the evaluation results.

[0012] Finite element analysis was performed on the selected candidate schemes to verify the optimal stiffener layout scheme.

[0013] Furthermore, the training conditional diffusion model includes:

[0014] A U-Net structure in three-dimensional voxel space is used as a denoising network. The input of the U-Net is the noisy three-dimensional voxel data, and the output is the predicted noise.

[0015] In the cross-attention layer of the U-Net, the cylinder block outer contour feature map is embedded as a geometric constraint mask to limit the generation area. In the geometric constraint mask, the voxel values ​​corresponding to the water jacket area, bolt hole area and cylinder bore area are set to zero.

[0016] Furthermore, the geometric constraint mask is a three-dimensional mask voxel map pre-constructed based on the outer contour, water jacket, bolt holes, and cylinder bore position information of the cylinder block sample.

[0017] Furthermore, the probability map is binarized to obtain multiple candidate layout schemes, including:

[0018] A preset threshold is set, and voxels with probability values ​​greater than or equal to the preset threshold in the probability graph are set as a first value indicating the presence of material, while voxels with probability values ​​less than the preset threshold are set as a second value indicating the absence of material.

[0019] Furthermore, the step of using a pre-trained performance prediction proxy model to quickly evaluate the performance of each candidate layout scheme includes:

[0020] A three-dimensional convolutional neural network is used as the performance prediction proxy model. The binarized voxel map of the stiffener layout is used as input, and the predicted quality, first-order bending mode frequency, first-order torsional mode frequency and maximum stress are output simultaneously.

[0021] Furthermore, the step of selecting candidate solutions that meet preset performance constraints based on the evaluation results includes:

[0022] Based on the predicted values ​​output by the performance prediction proxy model, schemes that do not meet the preset performance constraints are eliminated, and the top K candidate schemes are selected by sorting them according to the preset comprehensive performance index, where K is a preset positive integer.

[0023] Furthermore, the step of performing finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout scheme includes:

[0024] From the selected candidate schemes, an optimal topology layout that meets the mechanical performance requirements is determined through finite element analysis;

[0025] While keeping the optimal topology layout unchanged, a genetic algorithm is used to fine-tune and optimize the stiffener thickness parameters in the optimal topology layout.

[0026] Furthermore, after determining the optimal stiffener layout scheme, the process also includes:

[0027] The optimal reinforcing rib layout scheme is subjected to casting feasibility verification to check whether it meets the preset casting process requirements, which include at least the draft direction and minimum fillet radius.

[0028] If the conditions are not met, additional casting constraints are generated based on the verification results. These additional casting constraints are then merged with the target design constraints and input back into the condition diffusion model to regenerate candidate layout schemes.

[0029] Furthermore, the target design constraints are input in the form of a condition vector, which includes at least: engine displacement, cylinder center distance, target modal frequency range, maximum allowable stress, and minimum casting rib thickness.

[0030] On the other hand, the present invention also provides a cylinder block reinforcing rib structure design optimization system based on generative AI, comprising:

[0031] The data acquisition module is used to construct a dataset, which includes three-dimensional voxel models, binarized reinforcing rib layouts, and corresponding performance parameters of multiple cylinder block samples.

[0032] Generative model training module, used to train conditional diffusion model with design constraints as conditions;

[0033] The candidate generation module is used to obtain the target design constraints and input them into the trained conditional diffusion model. It generates multiple different stiffener layout probability maps through random sampling and performs binarization processing on the probability maps to obtain multiple candidate layout schemes.

[0034] The rapid evaluation module is used to perform rapid performance evaluation on each of the candidate layout schemes using a pre-trained performance prediction proxy model, and to select candidate schemes that meet the preset performance constraints based on the evaluation results.

[0035] The fine optimization module is used to perform finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout scheme;

[0036] The verification module is used to verify the casting feasibility of the optimal reinforcing rib layout scheme.

[0037] The above-described scheme achieves the following beneficial technical effects: The cylinder block stiffener structure design optimization method and system based on generative AI provided by this invention can significantly shorten the design cycle compared to existing technologies. Through trained conditional diffusion and surrogate models, dozens of candidate stiffener layouts satisfying design constraints can be automatically generated and screened within seconds to minutes, avoiding the time-consuming process of traditional iterative modeling and simulation. Simultaneously, the diffusion model introduces diversity through random sampling, generating unconventional layout schemes that transcend engineers' habitual thinking, avoiding getting trapped in local optima, and helping to explore and discover design configurations with superior performance. This invention injects performance targets such as stiffness, modal frequency, mass, and stress as conditional vectors into the generation process, and combines this with progressive screening using surrogate models and finite element analysis, efficiently achieving optimal balance among multiple conflicting design objectives. Furthermore, by embedding geometric constraint masks in the generative model and introducing casting feasibility verification and iterative feedback in the post-processing stage, it ensures that the final output design scheme is not only optimal in mechanical performance but also meets casting process requirements such as draft direction and minimum fillet radius, making it directly applicable to production. The final solution can also be automatically converted into a parametric CAD model, which is convenient for seamless integration into the existing engineering design process and has strong practical engineering value. Attached Figure Description

[0038] Figure 1 This is a flowchart of the cylinder block reinforcing rib structure design optimization method based on generative AI, according to an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the cylinder block reinforcing rib structure design optimization system based on generative AI, according to an embodiment of the present invention. Detailed Implementation

[0040] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.

[0042] In engine design, the cylinder block is a core load-bearing component, and the layout of its internal stiffeners directly affects the overall engine's rigidity, vibration and noise performance, and weight. Traditional stiffener layout design mainly relies on engineers' experience and finite element iterative optimization. This method suffers from long design cycles because each round of modifications requires re-modeling and simulation calculations in 3D. Furthermore, artificially defined topology optimization regions are prone to getting trapped in local optima, making it difficult to explore unconventional high-performance layouts. In addition, achieving a rapid balance between multiple objectives such as stiffness, weight, and casting manufacturability is a common engineering challenge.

[0043] In recent years, generative artificial intelligence technologies, such as generative adversarial networks and diffusion models, have shown great potential in the fields of image generation and structural topology optimization, but they have not yet been effectively applied in the automated design of complex three-dimensional structures such as cylinder block reinforcing ribs.

[0044] In view of this, in order to solve at least one of the above technical problems, this invention proposes a method and system for optimizing the design of cylinder block reinforcing rib structures based on generative AI. By constructing a dataset and training a conditional diffusion model guided by design constraints, it can automatically and quickly generate diverse candidate schemes for reinforcing rib layouts. Combined with surrogate models and finite element analysis, it can efficiently screen and optimize, thereby shortening the design cycle and effectively balancing multi-objective design requirements.

[0045] The present invention will now be described in detail with reference to the embodiments and accompanying drawings.

[0046] The first aspect of this invention provides a method for optimizing the design of cylinder block reinforcing rib structures based on generative AI, referring to... Figure 1 The method includes steps S10 to S60.

[0047] Step S10: Construct a dataset, which includes three-dimensional voxel models of multiple cylinder block samples, binarized reinforcing rib layouts, and corresponding performance parameters.

[0048] In this embodiment of the invention, a mapping library needs to be established first in order to train the subsequent generative and surrogate models. The collected data sources include existing historical cylinder block design schemes, and may also include a large amount of synthetic design data generated by traditional topology optimization methods (such as the variable density method) to enrich the diversity of the dataset and cover a wider design space.

[0049] For each cylinder block sample, a uniform voxel representation is applied. Specifically, the three-dimensional geometry of the cylinder block is converted into a three-dimensional matrix with a resolution of 128×128×128. The layout of the reinforcing ribs is represented using a binary voxel map: "1" indicates that the voxel location contains reinforcing material, and "0" indicates that the voxel location is a cavity region without material. This binary layout serves as the standard supervision signal for subsequent model training.

[0050] Simultaneously, each binarized layout sample is associated with its corresponding performance parameters. These performance parameters include at least: mass, first-order bending mode frequency, first-order torsional mode frequency, and maximum stress. These parameters are obtained through simulation calculations using standard finite element analysis procedures under uniform load conditions and boundary conditions.

[0051] Step S20: Train the conditional diffusion model using design constraints as conditions.

[0052] It's important to note that diffusion models are a type of generative model, with two core processes: forward diffusion and reverse denoising. The forward process gradually adds Gaussian noise to the real data samples until they become pure random noise. The reverse process trains a denoising network to learn how to remove noise step by step from the pure noise, recovering the original data. Conditional diffusion models, on the other hand, introduce conditional variables into this reverse denoising process to guide the generated results to meet specific design requirements.

[0053] In this embodiment of the invention, a conditional diffusion model is used as the core generator. During training, a binary reinforcement layout sample is extracted from the dataset, and noise is gradually added to it to generate a noisy layout. Then, the noisy layout and design constraints are input into the denoising network, allowing the network to learn to predict the added noise, thereby achieving a mapping from noise to layout.

[0054] Specifically, the denoising network employs a U-Net structure operating in three-dimensional voxel space. The U-Net extracts three-dimensional features step by step through an encoder, then restores the features to the original spatial resolution through a decoder, and retains detailed information using skip connections.

[0055] This invention introduces a pre-constructed geometric constraint mask as a geometric prior in the cross-attention layer of U-Net. This geometric constraint mask is a three-dimensional mask voxel map of the same size as the input voxel map (128×128×128). In this mask, the voxel values ​​corresponding to "forbidden zones" where reinforcement is prohibited, such as the water jacket region, bolt hole region, and cylinder bore region, are set to "0," while the voxel values ​​corresponding to the remaining permitted reinforcement regions inside the cylinder block's outer contour are set to "1." During the cross-attention calculation process, this mask acts as a strong constraint, causing the network to learn that any attention weight should tend to zero at the forbidden zone locations, thereby fundamentally preventing the generation of unreasonable layouts that infringe on functional areas. This geometric constraint mask is a pre-constructed three-dimensional mask voxel map based on the positional information of the cylinder block sample's outer contour, water jacket, bolt holes, and cylinder bores.

[0056] After training, the conditional diffusion model learns a denoising function that maps pure random noise and a set of conditional vectors into a meaningful 3D stiffener layout probability map. Each voxel in this probability map has a continuous value between 0 and 1, representing the probability of the presence of stiffener material at that location.

[0057] Step S30: Obtain the target design constraints and input them into the trained conditional diffusion model. Generate multiple different reinforcing rib layout probability maps through random sampling, and perform binarization processing on the probability maps to obtain multiple candidate layout schemes.

[0058] In this embodiment of the invention, the target design constraints are input in the form of a condition vector, which includes at least: engine displacement, cylinder center distance, target modal frequency range, maximum allowable stress, and minimum casting rib thickness. When a user sets a new set of target design constraints, the maximum allowable stress can be directly specified by the user or automatically obtained by the system based on the yield strength of the material grade selected by the user, multiplied by a preset safety factor. The preset safety factor is an empirical value preset according to the material type and load characteristics, typically selected in the range of 1.2 to 3.0, and 1.5 to 2.5 for gray cast iron cylinder blocks. For example, when "a four-cylinder 1.5T engine, target first-order bending frequency > 220Hz, mass < 18kg" is specified, this type of design constraint is first encoded into a condition vector. This condition vector, along with a pure random three-dimensional noise tensor sampled from a standard Gaussian distribution, is input into the trained diffusion model. The diffusion model performs a multi-step inverse denoising process, gradually converting pure noise into a clear layout probability map.

[0059] The diversity of generation schemes stems from the randomness of the initial random noise. By sampling different random noises, N (e.g., 50) distinct and diverse stiffener layout probability maps can be generated in parallel or sequentially.

[0060] After obtaining the probability map, it needs to be converted into a binary layout scheme that can be processed in subsequent steps. Specifically, a preset threshold (e.g., 0.5) is set. For each voxel in the probability map, if its probability value is greater than or equal to the preset threshold, it is set to the first value indicating the presence of material, such as "1"; if its probability value is less than the preset threshold, it is set to the second value indicating the absence of material, such as "0". Through this binarization process, each probability map is converted into a definite candidate layout scheme.

[0061] It should be noted that, in the embodiments of the present invention, the "mechanical performance requirements" is a higher-level concept, which specifically includes two categories: one category is "mechanical performance constraints," which refers to the hard threshold indicators that the scheme must meet, such as "maximum stress < allowable stress of material," "mass < upper limit of mass," "first-order bending modal frequency > lower limit of frequency," etc. If these are not met, the scheme is eliminated; the other category is "mechanical performance optimization objectives," which refers to the ranking criteria used to measure the merits of multiple schemes when comparing them, such as "the lighter the mass, the better, under the premise of meeting the constraints" or "the higher the first-order modal frequency, the better." The optimization objective can be a single objective or a weighted combination of multiple objectives.

[0062] Step S40: Using a pre-trained performance prediction proxy model, perform a rapid performance evaluation on each candidate layout scheme, and select candidate schemes that meet the preset performance constraints based on the evaluation results.

[0063] In this embodiment of the invention, the surrogate model is a lightweight three-dimensional convolutional neural network (3D CNN). The network takes a binarized voxel map of the rib layout as input, extracts spatial features through multiple layers of three-dimensional convolution and pooling operations, and finally outputs four predicted values ​​simultaneously through a fully connected layer: mass, first-order bending mode frequency, first-order torsional mode frequency, and maximum stress.

[0064] The training data for this proxy model comes from the layout samples in the dataset from step S10 and their corresponding precise finite element calculation results. The model is trained by minimizing the mean square error between the predicted values ​​and the actual simulation values.

[0065] The screening process is as follows: A set of preset performance constraints are pre-defined. The first type of constraint is an inherent constraint bound to material properties, such as "predicted maximum stress < yield strength of the selected material." The yield strength is an inherent physical property of the material; when the user selects the cylinder block material grade in the target design constraints, the system automatically retrieves the corresponding yield strength value as the upper limit of the constraint. The second type of constraint is a user-defined performance target constraint, such as "predicted mass < target mass upper limit." The surrogate model sequentially evaluates all N binary candidate schemes, eliminating schemes that do not satisfy any of the above constraints. Then, the schemes that satisfy the constraints are sorted according to preset comprehensive performance indicators (such as lightest mass or highest first-order modal frequency), and the top K candidate schemes are selected, where K is a preset positive integer; preferably, in some specific embodiments of this invention, K is 5.

[0066] Step S50: Perform finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout scheme.

[0067] In this embodiment of the invention, each candidate scheme is subjected to the exact same load conditions and boundary conditions as when the dataset was constructed, and simulation is performed to obtain its accurate mass, first-order bending / torsional modal frequencies, and maximum stress. Based on these accurate simulation results, engineers or algorithms automatically select a layout scheme with the best overall mechanical performance as the optimal topology layout.

[0068] Furthermore, even after determining the optimal topological layout, continuous parameters such as rib thickness still have room for optimization. This invention employs a genetic algorithm to fine-tune the rib thickness parameter. Specifically, in the finite element model, the rib thickness is parameterized, with rib thickness being a partitioned or independent variable for each rib. The optimization objective is to minimize mass or maximize modal frequency, constrained by the maximum stress not exceeding an allowable value. A genetic algorithm is used for iterative optimization. This fine-tuning process only changes the rib thickness dimension, without altering the presence or absence of the stiffeners, thus maintaining the topological layout unchanged.

[0069] Step S60: Casting feasibility verification.

[0070] In this embodiment of the invention, the optimal reinforcing rib layout scheme, determined after finite element verification and parameter fine-tuning, is converted into a computer-aided design (CAD) model. Within this CAD environment, it is automatically or manually checked whether it meets preset casting process requirements, which include at least the draft direction and minimum fillet radius. For example, it is checked whether the draft angle of the reinforcing rib's side is consistent with the mold parting surface to avoid failure to demold; it is checked whether the fillet radius at the intersection of the reinforcing rib and the outer wall of the cylinder body is greater than the minimum allowable value by the process to prevent casting cracks.

[0071] If the proposed design does not meet the casting feasibility requirements, additional casting constraints are generated based on the verification results. For example, if a rib wall angle is detected that prevents demolding, a local draft angle constraint is generated; if a corner radius is too small, a minimum corner radius constraint is generated. These additional casting constraints are merged with the target design constraints described in step S30 to form an updated condition vector, which is then input into the condition diffusion model to regenerate candidate layout schemes. The additional casting constraints are accumulated during the iteration process, meaning that the newly generated constraints in each round, together with the constraints from previous rounds, constitute the input conditions for the current round, gradually shrinking the generation space and guiding the model to generate a process-feasible layout.

[0072] The iteration stops when any of the following termination conditions are met: (1) there is at least one candidate solution generated in the current round that simultaneously satisfies all mechanical performance constraints and casting process requirements; (2) the number of iterations reaches the preset maximum number of iterations. After the iteration stops, the solutions that satisfy the mechanical performance constraints and casting process requirements from all historical rounds are sorted according to the preset mechanical performance optimization objective, and the solution with the highest ranking is selected as the final optimal design solution.

[0073] In some embodiments of the present invention, in order to seamlessly integrate the design results into the existing engineering process, the method further includes: converting the final optimal stiffener layout voxel model into a smooth STL surface format, and automatically establishing a solid model with parametric features through a secondary development script running in mainstream CAD software, which facilitates subsequent detailed engineering drawings and manufacturing preparation.

[0074] A second aspect of this invention provides a cylinder block reinforcing rib structure design optimization system based on generative AI, referring to... Figure 2 ,include:

[0075] The data acquisition module is used to construct a dataset, which includes three-dimensional voxel models, binarized reinforcing rib layouts, and corresponding performance parameters of multiple cylinder block samples.

[0076] Generative model training module, used to train conditional diffusion model with design constraints as conditions;

[0077] The candidate generation module is used to obtain the target design constraints and input them into the trained conditional diffusion model. It generates multiple different stiffener layout probability maps through random sampling and performs binarization processing on the probability maps to obtain multiple candidate layout schemes.

[0078] The rapid evaluation module is used to perform rapid performance evaluation on each of the candidate layout schemes using a pre-trained performance prediction proxy model, and to select candidate schemes that meet the preset performance constraints based on the evaluation results.

[0079] The fine optimization module is used to perform finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout scheme;

[0080] The verification module is used to verify the casting feasibility of the optimal reinforcing rib layout scheme.

[0081] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A method for optimizing the design of cylinder block reinforcing rib structures based on generative AI, characterized in that, include: Construct a dataset containing three-dimensional voxel models, binarized stiffener layouts, and corresponding performance parameters of multiple cylinder block samples; The conditional diffusion model is trained using design constraints as conditions. Obtain the target design constraints and input them into the trained conditional diffusion model. Generate multiple different stiffener layout probability maps through random sampling, and perform binarization on the probability maps to obtain multiple candidate layout schemes. Using a pre-trained performance prediction proxy model, the performance of each candidate layout scheme is quickly evaluated, and candidate schemes that meet preset performance constraints are selected based on the evaluation results. Finite element analysis was performed on the selected candidate schemes to verify the optimal stiffener layout scheme.

2. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The training conditional diffusion model includes: A U-Net structure in three-dimensional voxel space is used as a denoising network. The input of the U-Net is the noisy three-dimensional voxel data, and the output is the predicted noise. In the cross-attention layer of the U-Net, the cylinder block outer contour feature map is embedded as a geometric constraint mask to limit the generation area. In the geometric constraint mask, the voxel values ​​corresponding to the water jacket area, bolt hole area and cylinder bore area are set to zero.

3. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 2, characterized in that, The geometric constraint mask is a three-dimensional mask voxel map pre-constructed based on the outer contour, water jacket, bolt holes, and cylinder bore position information of the cylinder block sample.

4. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The probability map is binarized to obtain multiple candidate layout schemes, including: A preset threshold is set, and voxels with probability values ​​greater than or equal to the preset threshold in the probability graph are set as a first value indicating the presence of material, while voxels with probability values ​​less than the preset threshold are set as a second value indicating the absence of material.

5. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The step of using a pre-trained performance prediction proxy model to quickly evaluate the performance of each candidate layout scheme includes: A three-dimensional convolutional neural network is used as the performance prediction proxy model. The binarized voxel map of the stiffener layout is used as input, and the predicted quality, first-order bending mode frequency, first-order torsional mode frequency and maximum stress are output simultaneously.

6. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The step of selecting candidate solutions that meet preset performance constraints based on the evaluation results includes: Based on the predicted values ​​output by the performance prediction proxy model, schemes that do not meet the preset performance constraints are eliminated, and the top K candidate schemes are selected by sorting them according to the preset comprehensive performance index, where K is a preset positive integer.

7. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The step of performing finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout includes: From the selected candidate schemes, an optimal topology layout that meets the mechanical performance requirements is determined through finite element analysis; While keeping the optimal topology layout unchanged, a genetic algorithm is used to fine-tune and optimize the stiffener thickness parameters in the optimal topology layout.

8. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to any one of claims 1 to 7, characterized in that, After determining the optimal stiffener layout scheme, the following steps are also included: The optimal reinforcing rib layout scheme is subjected to casting feasibility verification to check whether it meets the preset casting process requirements, which include at least the draft direction and minimum fillet radius. If the conditions are not met, additional casting constraints are generated based on the verification results. These additional casting constraints are then merged with the target design constraints and input back into the condition diffusion model to regenerate candidate layout schemes.

9. The method for optimizing the design of cylinder block reinforcing rib structure based on generative AI according to claim 1, characterized in that, The target design constraints are input in the form of a condition vector, which includes at least: engine displacement, cylinder center distance, target modal frequency range, maximum allowable stress, and minimum casting rib thickness.

10. A cylinder block reinforcing rib structure design optimization system based on generative AI, applied to the cylinder block reinforcing rib structure design optimization method based on generative AI as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to construct a dataset, which includes three-dimensional voxel models, binarized reinforcing rib layouts, and corresponding performance parameters of multiple cylinder block samples. Generative model training module, used to train conditional diffusion model with design constraints as conditions; The candidate generation module is used to obtain the target design constraints and input them into the trained conditional diffusion model. It generates multiple different stiffener layout probability maps through random sampling and performs binarization processing on the probability maps to obtain multiple candidate layout schemes. The rapid evaluation module is used to perform rapid performance evaluation on each of the candidate layout schemes using a pre-trained performance prediction proxy model, and to select candidate schemes that meet the preset performance constraints based on the evaluation results. The fine optimization module is used to perform finite element analysis to verify the selected candidate schemes in order to determine the optimal stiffener layout scheme; The verification module is used to verify the casting feasibility of the optimal reinforcing rib layout scheme.

Images