A supercomputer-powered design technique enables the discovery of efficient mechanical structures that have an unprecedented level of detail. The findings provide insights into both physical and biological structures. See Letter p.84
As robots gradually replace human labour in manufacturing tasks, a similar trend is emerging in product design. Computational design techniques developed over the past two decades can create efficient mechanical structures nearly unsupervised. Today's cars and aircraft already contain components created by such processes. On page 84, Aage et al.1 report a remarkable advance in the capabilities of computational design. Their work could lead to mechanical structures that have lower weight and reduced environmental impact compared to those currently used.
Aage and co-workers use a method known as topology optimization. In this method, a structural-design problem is approached by distributing material in a given volume, using mathematical optimization techniques2,3. The possible material arrangements are practically unrestricted, which allows the method to generate innovative design solutions. Notably, there is no need for a designer to provide concepts or guidance — the evolution of a particular design is driven directly by simulated structural performance.
Topology-optimization procedures have been developed for design problems across a wide variety of disciplines, including solid and fluid mechanics, thermodynamics, acoustics and combinations thereof4. However, because computation time grows rapidly with increased design resolution, there is a limit on the structural features that can be resolved using these techniques, restricting applications to single components or simple structures.
Aage et al. report a breakthrough in addressing this limitation. Using advances in the integration of high-performance computing, mathematical optimization algorithms, data transfer and visualization procedures, the authors extend structural-topology optimization to a supercomputer environment. They produce designs comprising about 200 times more data than current state-of-the-art techniques. Such designs are the solution to optimization problems that have more than one billion variables.
The authors' design framework allows for six times finer detail in each spatial direction than is produced with currently available techniques. Not only does this result in a smoother and clearer final material layout, but it also enables different structural details and connections to be resolved. The optimal structure can therefore be described over a wider scale, from the smallest details to the size of the part itself. In this way, increased resolution allows the computational design process to reveal relationships between local and global features.
Aage and colleagues applied their extreme-resolution design framework to an engineering problem that spans a wide range of length scales: the lightweight design of an aircraft wing. The resulting design contains structural details that are more than 3,000 times smaller than the wing itself, illustrating the capability of the design process to create and integrate features at vastly different scales.
The final design has structures that resemble those of conventional aircraft wings, but also includes intriguing aspects such as curved spars and ribs (structural components of the wing), and relatively fine supporting struts connected to the skin of the wing (Fig. 1). The authors present an analysis confirming the superiority of the curved features over conventional straight ones. Their intricate mechanical design yields a mass reduction of up to 5% compared to current designs. If used on an aircraft, this would result in improved fuel economy and reduced emissions of carbon dioxide and other pollutants5.
In addition to offering engineers substantially more-powerful computational-design capabilities, the authors' framework provides biologists with an instrument for analysing and interpreting structures found in nature. Its unprecedented resolution can reveal designs that have hierarchical architectures of features spanning multiple length scales, as observed in structures such as bone and bamboo6. As an example, Aage et al. point out the resemblance between bone structures seen in a bird's beak and the fine struts observed in their computational wing design. Whether created by age-long evolution or supercomputer-powered design, optimal structures are rooted in the same mechanical principles.
However, care should be exercised in drawing conclusions from visual similarity alone. Restrictions imposed by specific fabrication processes, whether technological or biological, can be an important factor in shaping a mechanical structure. Such restrictions were not included in Aage and colleagues' study. The architectures of natural materials are also known for their remarkable fracture toughness7 (a measure of resistance to crack propagation), but this is not yet a commonly used criterion in computational design. Another future challenge will be to enhance the authors' high-resolution framework with capabilities to handle design problems involving, for example, time-dependent behaviour, which requires additional computational effort.
The design of Aage and colleagues' aircraft wing involved 8,000 computer processors used over several days. Few people will have access to such resources in the foreseeable future. Therefore, to enable a wider community to reap the benefits of the authors' work, it is imperative to find ways to more efficiently produce high-resolution designs. For example, the fact that very small structures were seen only in specific areas of the designed wing could allow for adaptive adjustment of resolution, which would reduce computational cost. Nevertheless, the authors' work represents a leap forward in the capabilities of computational design. Without doubt, its unprecedented resolution provides the foundation for further discoveries.
About this article
A novel asymptotic-analysis-based homogenisation approach towards fast design of infill graded microstructures
Journal of the Mechanics and Physics of Solids (2019)