Graph-based reinforcement learning for discrete cross-section optimization of planar steel frames

Hayashi, Keiichi; Ohsaki, Makoto

doi:10.1016/j.aei.2021.101512

Cited by 27 publications

(4 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Otherwise, the global effort for the RL agent is significantly larger compared to the conventional optimizer. Compared to the approaches used in the reviewed literature concerning RL for structural optimization, the agent was directly calling the FE-model (Hayashi and Ohsaki, 2020;Brown et al, 2022;Hayashi and Ohsaki, 2022;. This works for small academic models, but not for industry standard models as used for occupant safety simulations in modern vehicle development.…”

Section: Discussionmentioning

confidence: 99%

“…Notable is that the training is computationally expensive, but only needs to be done once. In further research, Hayashi and Ohsaki (2022) applied the technique to the optimization of planar steel frames under short-and long-term elastic load conditions, whereby the agent aims to minimize the structural volume under several practical constraints. In each action, the agent specifies each member by choosing it from a prescribed list.…”

Section: Optimization Of Mechanical Systems With Reinforcement Learningmentioning

confidence: 99%

See 1 more Smart Citation

Minimizing occupant loads in vehicle crashes through reinforcement learning-based restraint system design: assessing performance and transferability

Mathieu,

Gupta,

Di Roberto

et al. 2024

Proc. Des. Soc.

View full text Add to dashboard Cite

The optimization of mechanical behavior in safety systems during crash scenarios consistently poses challenges in vehicle development. Hence, a reinforcement learning-based approach for optimizing restraint systems in frontal impacts is proposed. The trained agent, which adjusts five parameters simultaneously, is capable of minimizing loads on a seen and unseen anthropomorphic test device on the co-driver position and is thus able of transferring knowledge. A hundred times higher rate of convergence to reach a similar optimum compared to a global optimization algorithm has been achieved.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Optimization Of Mechanical Systems With Reinforcement Learningmentioning

confidence: 99%

Minimizing occupant loads in vehicle crashes through reinforcement learning-based restraint system design: assessing performance and transferability

Mathieu,

Gupta,

Di Roberto

et al. 2024

Proc. Des. Soc.

View full text Add to dashboard Cite

show abstract

“…To stabilize the training process, we further introduce mini-batch training, in which trainable parameters are updated using multiple samples at the same time [5]. During the sampling phase, prioritized experience replay [8] is further employed to preferentially use samples that are highly unexpected to the agent.…”

Section: Loss Minimization For Trainingmentioning

confidence: 99%

“…RL is a method to learn a sequential decision-making process to maximize rewards through a huge number of simulations. Although RL has been reported to perform well in tasks that are difficult to control by rule-based programming [3] and in solving optimization problems [4], RL has rarely been applied to the field of skeletal structures because of its complex connectivity [5].…”

Section: Introductionmentioning

confidence: 99%

Assembly Sequence Optimization of Spatial Trusses Using Graph Embedding and Reinforcement Learning

Hayashi

Ohsaki

Kotera

2022

Journal of the International Association for Shell and Spatial

View full text Add to dashboard Cite

We consider a truss as a graph consisting of nodes and edges, and combine graph embedding (GE) and reinforcement learning (RL) to develop an agent for generating a stable assembly path for a truss with arbitrary configuration. GE is a method of embedding the features of a graph into a vector space. By using GE, the agent can obtain numerical information on neighboring members and nodes considering their connectivity. Since the stability of a structure is strongly affected by the relative positions of members and nodes, feature extraction by GE should be effective in considering the stability of a truss. The proposed method not only can train agents using trusses with arbitrary connectivity but also can apply trained agents to trusses with arbitrary connectivity, ensuring the versatility of the trained agents' applicability. In the numerical examples, the trained agents are verified to find rational assembly sequences for various trusses more than 1000 times faster than metaheuristic approaches. The trained agent is further implemented as a user-friendly component compatible with 3D modeling software.

show abstract

Reinforcement learning based agents for improving layouts of automotive crash structures

Trilling,

Schumacher,

Zhou

2024

Appl Intell

View full text Add to dashboard Cite

The topology optimization of crash structures in automotive and aeronautical applications is challenging. Purely mathematical methods struggle due to the complexity of determining the sensitivities of the relevant objective functions and restrictions according to the design variables. For this reason, the Graph- and Heuristic-based Topology optimization (GHT) was developed, which controls the optimization process with rules derived from expert knowledge. In order to extend the collected expert rules, the use of reinforcement learning (RL) agents for deriving a new optimization rule is proposed in this paper. This heuristic is designed in such a way that it can be applied to many different models and load cases. An environment is introduced in which agents interact with a randomized graph to improve cells of the graph by inserting edges. The graph is derived from a structural frame model. Cells represent localized parts of the graph and delineate the areas where agents can insert edges. A newly developed shape preservation metric is presented to evaluate the performance of topology changes made by agents. This metric evaluates how much a cell has deformed by comparing its shape in the deformed and undeformed state. The training process of the agents is described and their performance is evaluated in the training environment. It is shown how the agents and the environment can be integrated as a new heuristic into the GHT. An optimization of the frame model and a vehicle rocker model with the enhanced GHT is carried out to assess its performance in practical optimizations.

show abstract

Graph-based reinforcement learning for discrete cross-section optimization of planar steel frames

Cited by 27 publications

References 32 publications

Minimizing occupant loads in vehicle crashes through reinforcement learning-based restraint system design: assessing performance and transferability

Minimizing occupant loads in vehicle crashes through reinforcement learning-based restraint system design: assessing performance and transferability

Assembly Sequence Optimization of Spatial Trusses Using Graph Embedding and Reinforcement Learning

Reinforcement learning based agents for improving layouts of automotive crash structures

Contact Info

Product

Resources

About