EvoPhase: AI-led 'evolutionary design'

The team behind EvoPhase has a fundamental understanding of particle and fluid dynamics, and is one of the first research groups in the UK to explore the intersection of AI and particle technologies or complex fluid dynamics.

They originally set their sights on optimising the design of industrial equipment, including mixers, dryers, roasters, and blenders, for processing particulate materials. Despite their extensive use, the characteristics of granular materials remain poorly understood, and their behaviour is so complex that mathematical modelling is a significant challenge.

For this work, EvoPhase uses a numerical method called DEM (discrete element method) to simulate the movement of millions of small particles, and coupled it with evolutionary AI algorithms, that work like natural selection to ‘evolve’ entirely novel designs by selecting the best combination of parameters (such as power, throughput and mixing rate) rather than trading these targets off against each other.

Beyond granular materials

The team’s expertise extends beyond granular materials. Every single team member has a deep understanding of complex fluids and one of the other things EvoPhase does is to model the properties of fluids and optimise formulations, using a similar evolutionary approach.

The team recently completed a project for a Fast Moving Consumer Goods (FMCG) company that involved optimising both the fluid and the equipment used to process it.

Here the researchers were using numbers that are almost impossible to measure in real life, and they used their technology to optimise the parameters of a simulation, to provide the company with a really accurate model of this horrifically complicated fluid they work with.

The project had relatively few constraints in terms of what could optimised, or what could be changed within the system, and the final design delivered a tenfold improvement in the mixing rate.

Evolving a wind turbine

The team recently set their own challenge to test the power of their evolutionary algorithms – and designed a wind turbine that would work in the extremely low wind speeds in Birmingham, the UK’s second city.

The AI was able to generate, test, and refine over 2,000 wind turbine designs in just a few weeks, achieving what would have taken years and millions of pounds through conventional methods.

The result was an entirely novel – and bespoke – design for a set of curved blades that spin around a central point, and the simulations showed this design will be up to seven times more efficient than existing designs used in the Birmingham area.