eCNNTO: A Highly Generalizable ConvNet for Accelera...

eCNNTO: A Highly Generalizable ConvNet for Accelera... | AI Research

Key Takeaways

eCNNTO: A Highly Generalizable ConvNet for Accelerating Topology Optimization Topology optimization is a powerful engineering tool used to determine the most...
This work proposes an element-based Convolutional Neural Network (CNN) to accelerate density-based Topology Optimization (TO), termed eCNNTO.
To address this limitation, eCNNTO is proposed to build upon Kallioras et al.
However, the method lacks spatial correlations among neighboring elements and may lead to disconnected features in the final structure.
The proposed method employs CNN with residual connections to address this issue.

Paper AbstractExpand

This work proposes an element-based Convolutional Neural Network (CNN) to accelerate density-based Topology Optimization (TO), termed eCNNTO. TO generally undergoes a large number of iterations, where finite element analysis is performed in every iteration, leading to the efficiency bottleneck especially when dense meshes are used to achieve high-resolution designs. To address this limitation, eCNNTO is proposed to build upon Kallioras et al. (2020), where a Deep Belief Network (DBN) was trained for every element to predict its near-optimal density from its early history, thereby skipping the great majority of iterations and significantly accelerating the TO procedure. However, the method lacks spatial correlations among neighboring elements and may lead to disconnected features in the final structure. The proposed method employs CNN with residual connections to address this issue. On top of it, a novel training strategy is introduced to further enhance the optimization efficiency, where the training dataset consists of the final stage density histories rather than early ones. This change can also help reduce the required training data size. eCNNTO requires only a small dataset to train and yet it can be generalized to problems with largely different boundary conditions, loading cases, design domain geometries, mesh resolutions, as well as non-design domains. In the end, the generalization capabilities and efficiency of eCNNTO are demonstrated through a variety of examples in two and three dimensions, achieving up to 90% and 97% reduction of iterations, respectively.

eCNNTO: A Highly Generalizable ConvNet for Accelerating Topology Optimization
Topology optimization is a powerful engineering tool used to determine the most efficient distribution of material within a design space, such as in aerospace or automotive parts. However, the process is computationally demanding because it requires repeated finite element analysis across many iterations to reach a high-resolution design. This paper introduces eCNNTO, a new framework that uses a Convolutional Neural Network (CNN) to drastically speed up this process by predicting near-optimal structural designs from early-stage data, effectively skipping the majority of the traditional optimization iterations.

Improving Structural Connectivity

Previous attempts to accelerate this process, such as the DLTOP method, analyzed elements individually. While this reduced computation time, it ignored the spatial relationships between neighboring elements, often resulting in disconnected or physically defective structures. eCNNTO improves upon this by using a CNN architecture with residual connections. By analyzing "element patches"—the target element and its immediate neighbors—the model captures spatial correlations, ensuring that the final generated structures are more coherent and free of isolated, non-functional pieces.

A More Efficient Training Strategy

The researchers introduced a novel training strategy that further enhances efficiency. Instead of training the model on early-stage density data, as was common in previous methods, eCNNTO is trained using density histories from the final stages of the optimization process. This shift not only accelerates the optimization procedure but also reduces the amount of training data required to achieve high performance. Because the model learns the underlying patterns of structural evolution, it requires only a small dataset to become functional.

Strong Generalization Capabilities

One of the most significant advantages of eCNNTO is its ability to generalize. Once trained on a limited set of simple benchmark problems, the model can be applied to entirely new scenarios without needing to be retrained. It successfully adapts to varying boundary conditions, different loading cases, diverse design domain geometries, and different mesh resolutions.

Significant Computational Savings

The effectiveness of eCNNTO is demonstrated through various two-dimensional and three-dimensional examples. By predicting the near-optimal design and then performing only a few final "fine-tuning" iterations, the method achieves a massive reduction in the total number of iterations required. Specifically, the researchers reported up to a 90% reduction in iterations for 2D problems and up to a 97% reduction for 3D problems, representing a substantial leap in efficiency for high-resolution structural design.