S2E1

Abstract:

Optimized sensing is important for computational imaging in low-resource environments, when images must be recovered from severely limited measurements. In this paper, we propose a physics-constrained, fully differentiable, autoencoder that learns a probabilistic sensor-sampling strategy for optimized sensor design. The proposed method learns a system's preferred sampling distribution that characterizes the correlations between different sensor selections as a binary, fully-connected Ising model. The learned probabilistic model is achieved by using a Markov Chain Monte Carlo (MCMC) sampling inspired network architecture, and is trained end-to-end with a reconstruction network for efficient co-design. The proposed framework is applicable to sensor selection problems in a variety of computational imaging applications. In this paper, we demonstrate the approach in the context of a very-long-baseline-interferometry (VLBI) array design task, where sensor correlations and atmospheric noise present unique challenges. We demonstrate results broadly consistent with expectation, and draw attention to particular structures preferred in the telescope array geometry that can be leveraged to plan future observations and design array expansions.

Abstract:

Machine learning recently enjoyed remarkable attentions and developments in different branches of engineering sciences. One of the most fundamental challenges of its potential applications is the request for a large and good-quality database that “drives” the models. To overcome this issue, the concept of the physics-guided machine learning method was recently proposed, which incorporates the already-known physical laws into the training process to get a reasonable prediction with only a small observed dataset. In this study, we apply this tool to investigate the finite elastic deformation of plates and shells that is theoretically understood to be governed by the Föppl–von Kármán equations, a set of in-plane and out-of-plane equations that are notoriously difficult to solve by conventional theoretical or numerical methods. In particular, we will discuss different ways of defining the loss function and compare the predictions with numerical solutions in various stress conditions.

Abstract:

After a decade of periodic truss-, plate-, and shell-based architectures having dominated the design of metamaterials, we introduce the new non-periodic class of spinodoid topologies. Inspired by natural self-assembly processes, spinodoid metamaterials are a close approximation of microstructures observed during spinodal phase separation. Their theoretical parametrization is so intriguingly simple that one can bypass costly phase-field simulations and obtain a rich and seamlessly tunable property space as demonstrated, e.g., by their tailorable anisotropic elastic moduli. Counter-intuitively, breaking with the periodicity of classical metamaterials is the enabling factor to the large property space and the ability to introduce seamless functional grading.

Towards the creation of materials with as-designed properties, we address the inverse design question, i.e., how can we systematically and efficiently find a microstructural topology from the nearly infinite design space to achieve a sought combination of macroscale properties. We introduce an efficient and robust machine learning technique for the inverse design of (meta-)materials which, when applied to spinodoid topologies, enables us to generate uniform and functionally graded cellular mechanical metamaterials with tailored direction-dependent (anisotropic) stiffness and density. Despite the inverse problem being ill-posed (e.g., significantly different designs may yield the same desired stiffness), our algorithm, based on the integration of two neural networks for the forward and inverse structure- property maps, renders this challenge well-posed. We specifically present inverse-designed and biomimetic artificial bone architectures based on spinodoid topologies that not only reproduce the properties of trabecular bone accurately but also even geometrically resemble natural bone. With possible integration into multiscale topology optimization or as a standalone framework to explore the design space, this machine learning framework accelerates the design process of (meta-)materials with a wide range of tunable mechanical response (anisotropic stiffness only being the tip of the iceberg).