Multi-fidelity Bayesian Data-Driven Design of Energy Absorbing Spinodoid Cellular Structures Journal: Applied Mathematical Modelling
Leo Guo, Hirak Kansara, Siamak F. Khosroshahi, GuoQi Zhang, Wei Tan · 2025
Read Multi-fidelity Bayesian Data-Driven Design of Energy Absorbing Spinodoid Cellular Structures Journal: Applied Mathematical Modelling by Leo Guo, Hirak Kansara, Siamak F. Khosroshahi, GuoQi Zhang, Wei Tan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-22079.txt.
Introduction
Leo Guo and co-authors compare two Bayesian design strategies for cellular structures that absorb energy, in a 2025 paper. The structures, called spinodoids, are sponge-like metamaterials tuned for crash protection. Computer simulations of such designs grow ever more exact but also more costly, while data-driven design wants many samples.
Bayesian optimization is a data-thrifty way to tune costly functions. Its multi-fidelity cousin exploits cheap, coarse simulations alongside dear, fine ones. The authors apply both to the same real design task, a head-to-head test that was missing. They also use variance-based sensitivity analysis on well spread samples to shrink the design space before optimizing.
Catalog record
| CREATOR | Leo Guo, Hirak Kansara, Siamak F. Khosroshahi, GuoQi Zhang, Wei Tan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-22079
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Multi-fidelity Bayesian Data-Driven Design of Energy Absorbing Spinodoid Cellular Structures Journal: Applied Mathematical Modelling Leo Guo, Hirak Kansara, Siamak F. Khosroshahi, GuoQi Zhang, Wei Tan Abstract Finite element (FE) simulations of structures and materials are getting increasingly more accurate, but also more computationally expensive as a collateral result. This development happens in parallel with a growing demand of data-driven design. To reconcile the two, a robust and data-efficient optimization method called Bayesian optimization (BO) has been previously established as a technique to optimize expensive objective functions. In parallel, the mesh width of an FE model can be exploited to evaluate an objective at a lower or higher fidelity (cost & accuracy) level. The multi-fidelity setting applied to BO, called multi-fidelity BO (MFBO), has also seen previous success. However, BO and MFBO have not seen a direct comparison with when faced with with a real-life engineering problem, such as metamaterial design for deformation and absorption qualities. Moreover, sampling quality and assessing design parameter sensitivity is often an underrepresented part of data-driven design. This paper aims to address these shortcomings by employing Sobol’ samples with variance-based sensitivity analysis in order to reduce design problem complexity. Furthermore, this work describes, implements, applies and compares the performance BO with that MFBO when maximizing the energy absorption (EA) problem of spinodoid cellular structures is concerned. The findings show that MFBO is an effective way to maximize the EA of a spinodoid structure and is able to outperform BO by up to 11% across various hyperparameter settings. The results, which are made open-source, serve to support the utility of multi-fidelity techniques across expensive data-driven design problems. 1 Introduction Energy-absorbing structures, such as crumple zones, are essential in applications like vehicle safety, where their primary function is to convert kinetic energy into plastic deformation to protect occupants. By minimizing the forces experienced during collisions, these structures play a crucial role in the crashworthiness of vehicles, a key consideration in design and manufacturing. Traditionally, materials like steel and aluminum have been used in these applications. With advancements in 3D metal printing [ 1 ] , this process of additive manufacturing allows for the creation of complex prototypes that enhance energy absorption efficiency. More recently, however, there has been a shift towards using advanced composite materials and architected metamaterials, such as cellular structures [ 2 ] and shell-based structures [ 3 , 4 ] , that offer superior strength-to-weight ratios and improved energy absorption. With the development of materials science, mechanical metamaterials have emerged as a promising solution for lightweight and efficient energy absorption (EA). Architected materials, including periodic structures like the aforementioned cellular solids, offer unique mechanical advantages but are often limited by stress concentrations that lead to premature failure. Innovations such as triply periodic minimal surfaces (TPMS) [ 5 ] introduced doubly curved, non-self-intersecting structures that distribute stress more evenly. However, while TPMS provides higher material stiffness, the aperiodic spinodoid metamaterials [ 6 ] are more robust against imperfections in fabrication. Optimization of these complex energy-absorbing structures may require a strategic approach to balance multiple conflicting objectives, such as maximizing EA while minimizing peak forces (PF) experienced during impact. Conventional design methods often rely on trial-and-error or topology optimization [ 7 , 8 , 9 ] . Meanwhile, topology optimization iteratively redistributes material within a design domain but can produce geometries with sharp angles and irregularities that complicate fabrication. With both of these traditional methods having major drawbacks, data-driven design is a strong contender to address this type of optimization problem. One of the cornerstones of the data-driven design process is design space sampling. Popular examples include grid sampling [ 10 ] , uniform random sampling and Latin hypercube sampling. While these methods have proven to be effective in some use cases, Sobol’ sampling has proven to be more effective in gaining insight when data is scarce due to an expensive objective: by using Sobol’ samples, it is possible to perform variance-based sensitivity analysis of the objective with respect to the design parameters. However, sampling alone is not sufficient to provide a robust objective optimization scheme. In literature, data-driven optimization approaches leverage machine learning models as surrogates for costly simulations. Among them, Bayesian optimization (BO) with Gaussian process regression has gained traction for optimizing meta-material structure designs with high computational cost [ 11 , 12 , 13 ] , including spinodoid structures [ 14 ] . BO is a popular data-efficient objective optimization methodology. It aims to leverage the proven regression capabilities of Bayesian machine learning methods to assess the likelihood of the optimum location. This leads to a delicate balance between exploring new regions of high uncertainty where the black box function may have a global (or at least a better local) optimum, and exploiting regions where an optimum is already expected to be present. It trains a surrogate model on a set of training data, populated by evaluated design experiments, and subsequently optimizes a proxy belief model based on the surrogate regression model – also called an acquisition function. The design that is found through this process is then evaluated and augmented to the original training data, after which the cycle starts anew. See Figure 1 for a schematic overview of BO. Figure 1 : Flowchart diagram of BO. BO can be flexibly adapted into settings with data that adhere to particular input or output structures. Multi-fidelity BO (MFBO), which relies on surrogate models that handle the cost-accuracy trade-off between low- and high-fidelity data, allows for accelerated convergence by leveraging inexpensive approximations alongside selective high-fidelity evaluations. This approach enables more comprehensive exploration of the design space while maintaining the accuracy needed for realistic applications. Numerous successful mechanical engineering design optimizations have been performed by using MFBO. These cases include mathematical and physical phenomena with complex behavior such as inverted pendulums [ 15 ] , fractional advection-dispersion equations [ 16 ] , haemodynamics [ 17 ] and non-linear state-space models [ 18 ] . Also, engineering design processes have been developed with MFBO, including the design of antennas [ 19 ] , fixed-wing drones [ 20 ] , helicopter blades [ 21 ] , aerofoils [ 22 , 23 , 24 ] , flip-chip packages [ 25 ] , horizontal road alignment [ 26 ] , and analog circuits [ 27 ] . The multi-fidelity techniques has proven particularly useful when the objective functions are provided by finite element (FE) simulations, where the fidelity trade-off is intuitively encoded by the mesh size [ 28 , 29 ] . By incorporating elastic-plastic material behaviors into quasi-static FE compression simulations, we capture complex structural responses under impact. The positive track record of MFBO applications renders the EA optimization problem a suitable problem to tackle with a multi-fidelity optimization methodology. This work presents a novel framework that addresses two key gaps in current optimization practice: the lack of direct comparisons between MFBO and its single-fidelity counterpart, and the underexplored role of design space sampling in sensitivity analysis. We introduce a two-stage approach that first applies Sobol’ and Saltelli sampling to perform global, variance-based sensitivity analysis of the EA objective across fidelity levels—offering new insight into how parameter importance shifts with model accuracy. We then apply both BO and MFBO