Electric & Hybrid Vehicle Technology - July 2021
Opinion
2021-07-21 02:28:58
▶▶What is a digital twin? Industry seems particularly interested in them and there seems to be a lot of hype around what they are. BMW and NVIDIA just announced a partnership to use digital twin technology and it seems many other OEMs such as, GM, Ford, Toyota, BYD, also have divisions working on digital twins.
At its simplest, a digital twin is defined as a virtual model of a process, product or service. The comparison between the virtual and physical world allows the virtual world to be interrogated in ways that are not possible in the physical world, with the learnings then applied to improve the physical world.
In the application of batteries, for example, this typically means a model of a battery. If a digital twin was perfect, and everything that was done to the physical battery was done to the digital twin, then in theory the behavior of the model would match the behavior of the physical battery until the end of life. However, this is impossible.
Firstly, as statistician George Box said “All models are wrong, but some are useful”, it is simply not possible to capture all the complicated physics and complex interactions of anything, let alone a battery, with 100% confidence. Secondly, no measurements are perfect. They have limited accuracy and precision, therefore the inputs into the model will have errors too.
To overcome these problems the digital twin has to be regularly adjusted by comparison with those measurements so that it doesn’t deviate too far from reality.
This process seems to me to be remarkably similar to model-based state estimation. For example, to estimate state-of-charge a reduced order model is run. Current is measured and put into the model, which predicts voltage and compares this to the measured voltage. The model is then adjusted to minimize the error between the two. The model then contains variables such as state-of-charge, which can be observed in the model but cannot be measured directly from the real device.
"A digital twin is defined as a virtual model of a process, product or service"
If they are so similar, then what is the difference, or is there any difference at all? There does seem to be an increase in complexity when most people talk about or refer to a digital twin. State estimation is normally limited to one or two states of interest, and any other physics not needed is ignored. Therefore the goal is the create the simplest model possible that describes only the physics necessary to achieve the objective. Whereas digital twins tend to try to reproduce as much of the behavior as possible, to be as accurate a reflection of reality as possible.
The assumption that we know what needs to be included does not need to be made, instead we may find new states are important that had not been considered before. Therefore a digital twin is the opposite of model order reduction, instead it is the attempt to create the perfect model. However, those seduced by its attraction should always remember that all models are still wrong. What is important is understanding how wrong or, to put another way, how useful.
Dr Gregory Offer is a Reader in mechanical engineering at Imperial College London. His research focuses on battery, fuel cell and supercapacitor technologies, mainly in transport
©MAB - Aviation & Auto. View All Articles.
Opinion
https://ehv.mydigitalpublication.co.uk/articles/opinion?article_id=4077897&i=715185