Saving Time and Energy with Process-focused AI Simulations

Aniqah Majid spoke to University of Birmingham spinout Evophase about its location-specific wind turbines and how AI could change the way engineers design process equipment

EVOPHASE has come a long way from its humble beginnings above a Birmingham chicken shop. Last July, the University of Birmingham-based startup used its novel AI simulation technology to develop a wind turbine – the Birmingham Blade – specific to the city’s low urban wind speeds, alongside local steel fabricators KwikFab. I spoke to the small team hoping to become “the AI for engineering company”.

Aniqah Majid (AM): How did Evophase come about?

Leonard Nicusan (LN), chief technology officer: Jack [Sykes, COO], Dominik [Werner, CEO] and I are three PhD best mates who lived together in our first year PhD, funnily enough above a chicken shop of all places, and we were just incredibly keen and excited about developing AI simulation technology that could change the industrial landscape.

Even today, you can see the code that we were uploading online with the time stamps at about 3am. That’s mainly when we developed these algorithms.
Together with Kit [Windows-Yule], our CSO, we initially used these technologies in an academic context. And then after successfully pestering him for two years to start a company, Kit finally gave in and here we are.

AM: The novel AI technology you have built is called HARPPP (Highly-Autonomous Rapid Prototyping for Particulate Processes). Can you explain the thinking behind it?

Kit Windows-Yule, (KWY): The basic concept is survival of the fittest, but for simulations instead of animals. The software that we have, it can launch tens, or hundreds, or thousands of slightly different versions of a simulation of the same system, where each one has a slightly tweaked geometry or slightly tweaked operating conditions.

The ones that do not perform very well are killed off and the ones that perform very well indeed, we breed another generation of these simulations whose designs are based on the best of the last generation. And you go through this iterative process and generation by generation, the designs get better.

LN: In terms of the tech itself, we have a stack of technologies of artificial intelligence algorithms. Everything from doing the maths for us to evolving geometries and calibrations, we are making sure these simulations capture reality.

After developing amazingly accurate simulations of real industrial equipment, the next step was to use AI to go beyond what the human can tweak and change in terms of designs and conditions and allow us to find completely novel patentable designs. So, in a way, the HARPPP technology, specifically the one that optimises geometries, is a patent-making machine.

We have been able to apply them to anything from reactors for pharmaceutical applications to grinders for chocolate manufacturing, mixers for batteries and so on. So, the amount of applicability there is can be transformative across entire industrial sectors.

AM: How did you use the technology to develop wind turbines?

KWY: Vertical axis wind turbines historically only work for relatively high wind speeds, so you need to hit a pretty high wind speed before the blades even start moving for most conventional designs. It means that they simply don’t work very well in places like Birmingham, where you just don’t have the right kind of wind speeds.

The idea was to try and optimise designs for a specific region. So rather than having a good wind turbine that can be put wherever it can, we developed a specific one for Birmingham and then another specific one for Edinburgh. Obviously, historically that would be completely impossible if it’s costing hundreds of thousands of pounds and an entire R&D team taking two years to make one design.

LN: I have worked on a three-year grant that was worth £300,000 with a company whose entire raison d’etre is to make one piece of equipment, a static mixer, and it took them three years and £300,000 to create one single design.

KWY: But if you can try tens of thousands of designs in a couple of weeks, then you can optimise the design for each individual locality where you might want to have one of these wind turbines.

We think this could solve a huge problem for the UK because if you go with conventional wind turbines, one of the big problems is that most energy use in the UK is in the south, in London, and the best place to create wind energy is in Scotland. And then you’ve got to transfer that energy down and you lose, and waste, a phenomenal amount of electricity.

The idea that we had is to create local energy produced and used locally instead of wasting all this energy in transit. We believe that every city in the UK should have its own little wind turbine design, and they should be on the roof of every industry in that region.

LN: We started discussing the evolutionary algorithms we developed with the company KwikFab, a family-run manufacturer in Birmingham, who were keen to let us try to optimise these wind turbine designs. We talked about the design possibilities, what they wanted, and what freedoms we wanted to give to the AI.

We then discussed with the manufacturing team at KwikFab what kind of constraints they have, including whether we would be able to let the AI develop these angles and so on – you must make the AI representative of the manufacturing capabilities of the team you work with.

Then it was a matter of creating a simulation, validating it, making sure that the computational fluid dynamics (CFD) actually represent reality. And in this case external aerodynamics are one of the most difficult forms of CFD.

We have developed this simulation and then we let our AI run wild with it. It developed designs that we found to be seven times better converting wind and wind kinetic energy into angular rotation energy than the currently available urban designs.

We started the project in June, developed the simulation in August, ran it in September, and handed it over to KwikFab in October. And it was on campus in November.

AM: What will a commercial wind turbine look like?

Paul Jarvis, MD of KwikFab: We are still figuring this out. This prototype turbine is built for the speeds we get around Birmingham, but we have not got that far to see how varied the construction will be for other wind speeds. We want to build a few of them for operation. This is just a display build. We will be developing some working models that are a bit bigger and get them up on the roof and then testing them for six months or more and see what comes of it.

We would start off with just the same design, then probably modify them as we’re going along. Testing the modifications before we put them back up on the roof and try again. But this one does exactly what the lads said it would do and it has not stopped turning all day.

AM: What can we expect from Evophase in the near future?

Dominic Werner, CEO: We are planning to publish our software as a service platform where we are going to shift away from a consultancy-based model towards an online platform where users and companies can do exactly those kinds of simulations and optimisations that we have with KwikFab, without us needing to help them. That makes it cheaper for them, cheaper for us, and more efficient.

KWY: We are also currently working with Johnson Matthey on optimising some of the classical pieces of chemical engineering equipment and obviously partnering with a business like that, the reach they bring and the sustainability focus, is absolutely fantastic.

Aquapak Polymers are developing biodegradable and recyclable plastics, and we’re working with them not only to make their processes more efficient, but to develop more bio-derived polymers. So, we are working on a number of projects spanning at least five different sectors and trying to bring new efficiencies to all of them.