A Data-Driven Closed-Loop Platform for Optimal Design of Deployable Pin-Jointed Structures​​

Deployable pin-jointed (DPJ) structures are prevalent in contemporary engineering, notably in architectures and large spacecrafts. These structures can also model cytoskeleton cells. Broadly, they fall into three categories: those primarily made of cables, ones made of bars, and tensegrity structures incorporating both elements. With the recent technological demands, there’s an increasing emphasis on attributes like high shape precision and adjustable stiffness in DPJ structures. However, optimizing these structures presents challenges, especially with current modeling constraints and inadequate design tools. Traditional design methods tend to follow an “open-loop” approach; in contrast, my research aspires to transition to a data-driven “closed-loop” platform for more precise DPJ design. This innovative approach not only uses experimental results for model validation but also leverages them to refine the computational model, subsequently offering a foundation for robust initial structural guidelines.

Application of deployable pin-Jointed structures: (a) hanging roof of Olympiastadion; (b) Kurilpa bridge; (c) Astro-Mesh deployable mesh reflector; (d) cytoskeleton cells.
Flowchart of the closed-loop platform for optimal design of DPJ structures.

Form Finding

In the closed-loop platform for optimal structural design, two main sources are utilized to enhance DPJ structure performance. The first hinges on a novel stochastic method for form finding. Unlike traditional methods, which are constrained by the inability to position tensegrity structure nodes specifically and heavily depend on geometric simplicity, this new approach can precisely position nodes without member grouping or structural simplicity. This versatility makes it ideal for designing intricate, large-scale DPJ structures. The method seamlessly integrates solving linear equations with stochastic optimization and suits most DPJ topology designs without substantial alterations.

Form finding of DPJ structures: traditional methods vs the proposed method.

Data-Driven Computational Modeling Technique

In the optimal structural design’s closed-loop platform, a data-driven computational modeling technique is employed for precise mechanical behavior predictions of DPJ structures. Traditional constitutive modeling often oversimplifies structures, missing vital properties like inelastic behaviors. Especially with large or complex DPJ structures, these methods falter. Addressing this, the research introduces a new data-driven technique harnessing deep learning and nondestructive testing, such as 3D DIC and digital shearography. This approach, trained solely on experimental data, offers enhanced accuracy and efficiency without relying on constitutive models.

Development of a computational model by deep neural networks for DJP structures.
Example of ongoing research on nondestructive testing (a) 3D DIC measuremetn; (b) deformation of the object; (c) experimental setup of digital shearography; (d) measured shearogram phase map.

References

  1. Yuan, S.* and Zhu, W., 2021. Optimal self-stress determination of tensegrity structures. Engineering Structures, 238, p.112003. doi: 10.1016/j.engstruct.2021.112003