AI-aided exploration of lunar arch forms under in-plane seismic loading

Abstract

Increasing computational power has led to the expansion of civil engineering research into using machine learning concepts for developing improved design strategies. These strategies are particularly useful for the design of extra-terrestrial habitats under uncertain environmental conditions. This paper focuses on building an unsupervised machine learning model (convolutional autoencoder) capable of detecting patterns in arch shapes and differentiating between their stress and displacement contours. Foremost, detailed discussions of the model’s architecture and input data are presented. The variation of arch shapes and contours between cluster centroids in the latent space is determined, proving the capability of optimisation by moving towards clusters with optimal contours. Finally, a regression model is built to investigate the relationship between the input geometric variables and the latent space representation. We prove that the autoencoder and regression models produce arch shapes with logical structural contours given a set of input geometric variables. The results presented in this paper provide essential tools for the development of an automated design strategy capable of finding optimal arch shapes for extra-terrestrial habitats.

Machine Learning in Structural Design: An Opinionated Review

“…The exploration of diverse design options brought about by AI was also exploited by Maqdah et al (2021) and Palmeri et al (2021) while studying the provision of structurally-efficient regolith-based arch forms for extraterrestrial construction. They built unsupervised machine learning models (Convolutional Autoencoders, CAE) capable of detecting patterns and differentiating between arch geometries and their stress and deformation contours (Figure 4).…”

“…Although the optimal configurations found resembled those obtained by more traditional approaches (McLean et al, 2021), the possibility of including a diversity of design actions (gravity, thermal, and seismic) and a substantial number of dimensions that are then reduced to a smaller latent space where a holistic search process can be used was featured as a clear contribution of AI. Moreover, Maqdah et al (2021) and Palmeri et al (2021) were able to elucidate some of the dependencies of the latent space (reduced) dimensions on geometric and structural parameters which can be helpful in making informed (partially explainable) searches. Alongside the CAE, regression models were used to allow the visualisation of the changes in the arch shape and stress fields when moving towards a certain direction in the design space.…”

“…Similarly, efforts to incorporate heuristic thinking into AI have been trialled in other branches of design (Nanda and Koder, 2010) and it may be beneficial to explore those in structural engineering also. At the end of the day, heuristics (intuition) is already routinely used by engineers to reduce the search space of potentially feasible designs, e.g., (Maqdah et al, 2021;Palmeri et al, 2021;Danhaive and Mueller, 2021). A perceived hurdle, however, comes form the fact that much of the progress of ML and AI has come from the formalization of mathematical and logical approaches aiming at well defined problems with clear goals.…”

Editorial to special issue “Recent mechanics-based developments in structural dynamics and earthquake engineering”

“…In particular, a feedforward neural network is used for inverse modeling of nonlinear restoring forces. On the other hand, Maqdah et al [10] build an unsupervised machine learning model capable of detecting patterns in arch forms under seismic loading and distinguishing between their stress and displacement contours. In one of the three contributions on structures under seismic loads, Zakian and Kaveh [11] provide a comprehensive review on seismic design optimization of engineering structures.…”