CDS&E: Deep Spring: a Neural Network-based Approach to Design of Slender Structures

Slender structures are vital to our lives – from columns and beams that support buildings, to airplane wings and blood vessels. By understanding how they respond to strain and the factors that give rise to bending and buckling, engineers can design safer and more efficient systems with applications in robotics, civil engineering, and medical treatment.

Many systems with slender structures exhibit complex behavior that is non-linear and counter intuitive. This Computational and Data-Enabled Science and Engineering (CDS&E) project combines machine learning and physics-based modeling to better understand their behavior and how to harness it to provide new capabilities. The findings will enable new structures such as soft robots with new forms of motion and control for medical procedures or search and rescue activities.

The knowledge gained may also enable the design of products such as stretchable electronics for health care, bio sensing and a broad range of consumer and industrial products. The project will develop a website and software repository to share the simulation tools with the research community. Educational materials including instructional videos will be developed for graduate courses and K-12 classes.

This project will enable the design of new materials and structures like soft kirigami composites and compressive buckling-induced micro-sized 3D architecture materials. The researchers will advance numerical modeling of systems with slender structures through a combination of machine learning and discrete differential geometry, using “nonlinear springs” represented by neural networks to simulate the systems.

The enhanced computational speed of this approach will aid in the design and optimization of engineering systems such as deployable structures and soft robots. Developing predictive models for slender structures and metamaterials is challenging because of the inherent instabilities and many possible configurations. If successful, this development would be the first of its kind in the mechanics research community and would provide the twin capability to predict complex structural responses and design structures with an inverse problem solver.

While the target proof-of-concept examples relate to slender structures and metamaterials, the computational methods would be generalizable to a broad class of material-structure systems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.