In topology optimization using deep learning, the load and boundary conditions represented as vectors or sparse matrices often miss the opportunity to encode a rich view of the design problem, leading to less than ideal generalization results. We propose a new data-driven topology optimization model called TopologyGAN that takes advantage of various physical fields computed on the original, unoptimized material domain, as inputs to the generator of a conditional generative adversarial network (cGAN). Compared to a baseline cGAN, TopologyGAN achieves a nearly 3 × reduction in the mean squared error and a 2.5 × reduction in the mean absolute error on test problems involving previously unseen boundary conditions. Built on several existing network models, we also introduce a hybrid network called U-SE(Squeeze-and-Excitation)-ResNet for the generator that further increases the overall accuracy. We publicly share our full implementation and trained network.

The demand for fast and accurate structural analysis is becoming increasingly more prevalent with the advance of generative design and topology optimization technologies. As one step toward accelerating structural analysis, this work explores a deep learning based approach for predicting the stress fields in 2D linear elastic cantilevered structures subjected to external static loads at its free end using convolutional neural networks (CNN). Two different architectures are implemented that take as input the structure geometry, external loads, and displacement boundary conditions, and output the predicted von Mises stress field. The first is a single input channel network called SCSNet as the baseline architecture, and the second is the multi-channel input network called StressNet. Accuracy analysis shows that StressNet results in significantly lower prediction errors than SCSNet on three loss functions, with a mean relative error of 2.04% for testing. These results suggest that deep learning models may offer a promising alternative to classical methods in structural design and topology optimization. Code and dataset are available at https: // github. com/ zhenguonie/ stress_ net

No abstract

Using deep learning to analyze mechanical stress distributions are gaining interest with the demand for fast stress analysis. Deep learning approaches have achieved excellent outcomes when utilized to speed up stress computation and learn the physical nature without prior knowledge of underlying equations. However, most studies restrict the variation of geometry or boundary conditions, making it difficult to generalize the methods to unseen configurations. We propose a conditional generative adversarial network (cGAN) model called StressGAN for predicting 2D von Mises stress distributions in solid structures. The StressGAN model learns to generate stress distributions conditioned by geometries, loads, and boundary conditions through a two-player minimax game between two neural networks with no prior knowledge. By evaluating the generative network on two stress distribution datasets under multiple metrics, we demonstrate that our model can predict more accurate stress distributions than a baseline convolutional neural network model, given various and complex cases of geometries, loads and boundary conditions.

Spatter is an inherent, unpreventable, and undesired phenomenon in laser powder bed fusion (L-PBF) additive manufacturing. Spatter behavior has an intrinsic correlation with the forming quality in L-PBF because it leads to metallurgical defects and the degradation of mechanical properties. This impact becomes more severe in the fabrication of large-sized parts during the multi-laser L-PBF process. Therefore, investigations of spatter generation and countermeasures have become more urgent. Although much research has provided insights into the melt pool, microstructure, and mechanical property, reviews of spatter in L-PBF are still limited. This work reviews the literature on the in situ detection, generation, effects, and countermeasures of spatter in L-PBF. It is expected to pave the way towards a novel generation of highly efficient and intelligent L-PBF systems.

This research presents a method of optimizing the consolidation of parts in an assembly using metal additive manufacturing (MAM). The method generates candidates for consolidation, filters them for feasibility and structural redundancy, finds the optimal build layout of the parts, and optimizes which parts to consolidate using a genetic algorithm. Results are presented for both minimal production time and minimal production costs, respectively. The production time and cost models consider each step of the manufacturing process, including MAM build, post-processing steps such as support structure removal, and assembly. It accounts for costs affected by part consolidation, including machine costs, material, scrap, energy consumption, and labor requirements. We find that developing a closed-loop filter that excludes consolidation candidates that are structurally redundant with others dramatically reduces the number of candidates, thereby significantly reducing convergence time. Results show that when increasing the number of parts that are consolidated, the production cost and time at first decrease due to reduced assembly steps, and then increase due to additional support structures needed to uphold the larger, consolidated parts. We present a rationale and evidence justifying that this is an important tradeoff of part consolidation that generalizes to many types of assemblies. Subsystems that are smaller, or can be oriented with very little support structures or have low material costs or fast deposition rates can have an optimum at full consolidation; for other subsystems, the optimum is less than 100%. The presented method offers a promising pathway to minimize production time and cost by consolidating parts using MAM. In our test-bed results for an aircraft fairing produced with powder-bed electron beam melting, the solution for minimizing production cost (time) is to consolidate 17 components into four (two) discrete parts, which leads to a 20% (25%) reduction in unit production cost (time).

In design for forming, it is becoming increasingly significant to develop surrogate models of high-fidelity finite element analysis (FEA) simulations of forming processes, to achieve effective component feasibility assessment as well as process and component optimizations. However, surrogate models using traditional scalar-based machine learning methods (SBMLMs) fall short on accuracy and generalizability. This is because SBMLMs fail to harness the location information available from the simulations. To overcome this shortcoming, the theoretical feasibility and practical advantages of innovatively applying image-based machine learning methods (IBMLMs) in developing surrogate models of sheet stamp forming simulations are explored in this study. To demonstrate the advantages of IBMLMs, the effect of the location information on both design variables and simulated physical fields is firstly proposed and analyzed. Based on a sheet steel stamping case study, a Res-SE-U-Net IBMLM surrogate model of stamping simulations is then developed and compared with a baseline multi-layer perceptron (MLP) SBMLM surrogate model. The results show that the IBMLM model is advantageous over the MLP SBMLM model in accuracy, generalizability, robustness, and informativeness. This paper presents a promising methodology in leveraging IBMLMs as surrogate models to make maximum use of information from stamp forming FEA results. Future prospective studies that are inspired by this paper are also discussed.

Using deep learning to analyze mechanical stress distributions has been gaining interest with the demand for fast stress analysis methods. Deep learning approaches have achieved excellent outcomes when utilized to speed up stress computation and learn the physics without prior knowledge of underlying equations. However, most studies restrict the variation of geometry or boundary conditions, making these methods difficult to be generalized to unseen configurations. We propose a conditional generative adversarial network (cGAN) model for predicting 2D von Mises stress distributions in solid structures. The cGAN learns to generate stress distributions conditioned by geometries, load, and boundary conditions through a two-player minimax game between two neural networks with no prior knowledge. By evaluating the generative network on two stress distribution datasets under multiple metrics, we demonstrate that our model can predict more accurate high-resolution stress distributions than a baseline convolutional neural network model, given various and complex cases of geometry, load and boundary conditions.