The Automotive Bot Challenge (ABC) is one of our current projects. Our goal is to create the first electric vehicle architecture with the aid of artificial intelligence. We are currently working on power transmissions, and looking for partnerships to push this project forwards.
Europe has set a goal: a 55% cut on CO2 emissions by the year 2030. As a consequence, a great number of car manufacturers is working towards making their fleets 100% electric before this deadline.
The question is – how can this transition be financially viable for both manufacturers and consumers? A 30% price gap is currently observed between lower-end electric vehicles and fuel ones, the former being the priciest.
Electric vehicles currently face 2 main challenges – an environmental one, and a financial one.
The goal of this project is to reduce the mass of an electric vehicle by 20%, and to improve its autonomy by 30%. In order to reach said goal, we are developing bots that help engineers come up with new designs. The kind of work engineers currently do is kin to craftsmanship. Engineers and technicians design each component of a vehicle with a 3D CAD.
The artificial intelligence of Dessia’s platform generates a great number of solutions for a given engineering problem. The software our platform creates to generate said solutions is a “bot”. When we think about it, the traditional way for an engineer to design a battery is to keep in mind a solution that they have found has worked in the past. Dessia’s artificial intelligence generates all possible solutions, in order to find and keep the best one.
We aim at creating bots for a variety of components – electrical engine, battery, power electronics, tie rods, car platforms. The bots will work in coordination searching for disruptive electric vehicles architectures.
Dessia’s initial scope was electric speed reducers. The goal was to create our first bot, which would have generated all possible reducers. These components connect the electric engine to the wheels, which means they have two main jobs – shifting the axle rotation of the electric engine off the wheels axle, and suggesting a reduction ratio between the wheels and the electric engine.
A number of reducer architectures can be designed starting from architectures of simple gears and epicyclic gear trains. As of now, we have obtained our first results based on architectures with 2 or 3 driveshafts.
As you can observe in the picture below, the connection of the electric engine to the wheels lies on one or two gear ratios.
Our approach optimizes the size of said reducer by taking into account the gear ratio between the electric engine and the wheels, as well as the design area (the volume we want to fit our reducer into). Eventually, and with the aid of a genetic algorithm, we obtain a 3D geometry (see below) with the following information:
Example of 3D solutions
We obtain nearly 900 different options, including those with 2 or 3 shafts.
The question remains – how can we help engineers make their choice? “Clustering” groups solutions together into families. The clusters we build rely on a norm requiring maximal pressure on the cogs and on the total mass.
We end up with 5 clusters: 3 of them are 2-shaft solutions (red, orange, and green cluster), and the other 2 are 3-shaft solutions (blue and rose cluster). We can observe a 10% decrease in maximum pressure between the best 2-shaft solution and the best 3-shaft solution. A slight decrease of the total mass in the 3-shaft architectures is the result of the bot generating an extra wheel.
The analysis of 3-shaft systems can go two different ways: in the blue cluster we find the high contact ratio solutions (average count of cog contact or teeth meshings), and in the pink one we find the solutions with lower contact ratio. The separation of the two clusters is due to two reasons:
Two 3-shafts architectures close to optimum 2
As for the 2-shaft solutions, they come with 3 clusters. The bot’s solutions only feature two wheels making contact (no countershaft, which would make it easier to change the settings). The difference among the three clusters lies in the average number of teeth of each architecture. We chose to prioritize 2 optimums: the first one has the best contact pressure which makes it last longer (blue cluster), whereas the other optimum is more relevant in terms of mass. Please note that clustering makes the analysis of the feasible solutions faster – in our case, we have excluded the red and rose clusters for feasibility reasons.
An improvement of our existing architectures will follow this first study, and will take into account: carters, shaft design, and epicyclic gear train. We took a fairly simple approach to the shafts we have generated throughout this study – the diameter of the cylinder has been stable all throughout the shaft line. We will soon publish results with a more advanced definition of these shaft lines, taking into account their interruptions, flutings, and process feasibility rules.
Using a bot for the exhaustive generation of reducers brings several perks:
Dessia is currently working on integrating finer modeling elements and on new technological bricks, so as to broaden the possibility space.
Our vision? To change the game through full-scale digitalization of engineering, to provide engineers with helpful data in their decision-making process.
Learn more about this project in our YouTube video!