AI for design
Australia faces a new wave of challenges in addressing the joint problems of modernising its industry whilst concurrently supporting Net Zero targets and transitioning to greener innovation models. There is a critical need to accelerate technology development to meet these demands across a range of sectors, and CSIRO, out of all Australian entities, is best placed to deliver those technologies. AI for design aims to drive this technology development. CSIRO is already working to produce solutions across a range of relevant industrial components, from static catalytic mixers to air capture devices, and from soft robotic grippers to new types of membranes. However, these devices are typically hand-designed, based on guesswork and intuition, rather than the result of any extensive design search. Because of the lack of ability to perform principled, comprehensive design explorations in these challenging problems, the component designs currently used in these processes are likely suboptimal. This has knock-on effects in terms of limiting the quality, efficiency, and yield of the associated industrial processes, and as well as the types of industrial processes that can be supported. In short, the solutions are unlikely to meet the rapid technological advancement required to revolutionise Australian industry.
The solution is to provide a comprehensive design tool, allowing researchers to generate provably optimal industrial componentry that can enable previously impossible industrial processes. AI for design will create next-generation high-performance industrial equipment via a combination of computational design and advanced modelling. AI for design aims to accelerate technology discovery to meet nationally and internationally relevant goals surrounding Net Zero targets as well as modernising Australian industry. AI for design is a portfolio, co-funded by AI For Missions and Future Digital Manufacturing Fund (FDMF), both Data61 APaIR initiatives. Our solution combines the real-world veracity of fabrication-based testing with the speed and efficiency of model-based testing, providing a scalable design framework for exploring a wide range of promising solutions that no human could design, and whose outputs (individual designs) work in reality the same way they do in the model. By applying this approach to a range of high-value industrial components, we will unlock new efficiency and performance, and discover unprecedented solutions to some of Australia’s most pressing issues. In AI For Missions, our Mission engagement focuses on next-generation equipment to help us push Towards Net Zero targets, as well as making Australia more Drought Resilient.
Our Research Methodology
Each project follows an innovative research methodology that represents a new way of performing science, transforming the way CSIRO researchers perform design-based research. The portfolio additionally aims to create a CSIRO-wide community of practice centred on this novel scientific method. Each AI for design project follows a research methodology combines advanced design algorithms with cutting edge computational modelling and real-world data acquisition. Our research methodology directly addresses two underpinning science challenges:
- The first science challenge is the Reality Gap – the fact that models are not perfect reflections of reality, and thus a design assessed in the model will behave differently than the same design assessed in reality. AI for design tackles this challenge through the use of experimentally coupled modelling to tune model behaviour to the observed experimental behaviour. Additionally, we focus on reducing fabrication noise (distortion, layer-based noise, and other effects), so our printed designs are closer to their ‘ideal’ CAD models, and the data they generate is less noisy. This strongly ties to wider Data61 efforts in digital twinning and modelling of additive manufacturing processes.
- Our second science challenge is design representation. Our goal is to find novel, previously undiscovered solutions to accelerate technology development in our chosen domains. To do this, we need to create expressive new digital design representations to work in previously inaccessible design spaces. Typically, this would move away from simpler, direct representations, into indirect representations that are more freeform in the possible designs they encode but require significant advances in techniques to efficiently explore these more unbounded spaces. The designs must also be printable, so we can choose to fabricate and experimentally verify any design.
Benefits of an AI for design system include the ability to explore a huge range of previously-unimaginable designs in a principled manner, quickly, and cheaply. The end goal is a series of accurate models, and a series of novel industrial components generated by employing computational design techniques on those models.
Dr David Howard
- David is a Senior Research Scientist in the Cyber Physical Systems program at CSIRO, Australia’s national science body. He leads multiple projects at the intersection of robotics, evolutionary machine learning, and the computational design of novel physical objects, previously leading a portfolio in the AIM Future Science Platform and currently leading the AI4Design portfolio. His interests include nature-inspired algorithms, learning, autonomy, soft robotics, the reality gap, and evolution of form. His work has been featured in local and national media.
- Primary Emaildavid.email@example.com