What is this? This is The Dots, our newsletter about exciting things we find in the world of innovation. We imagine innovation as connecting the dots; putting together a jigsaw. Our puzzle pieces are the pieces of interesting information we absorb in the world, our partners and their products. This newsletter is about these dots we find and connect.
Two weeks ago we asked you all to vote on what you thought was and was not a digital twin and, well, all I can really say is that you’re either an indecisive bunch or the definition of digital twin is (in most cases) quite woolly! Let’s look at the results in detail and see what questions arise.
A digital restaurant twin
A restaurant uses a system for managing tables. The system has a 2D map of the restaurant with all the covers marked on it. It shows how many people are at each table, when they came in and predicts, based on the status of their order, what time the table will be vacated. Front of house staff keep table occupancy information up to date. Order statuses are pulled from the kitchen management system.
Reflections
Starting out with a clear 50/50 split. A Brexit of digital twin opinions. There is a fairly common belief in the world of digital twins - that digital twins need to have a visual element - a graphical representation either in two or three dimensions. I’d hazard that those voting ‘no’ on this either think that the third dimension is an important one or believe that there has to be some direct link between the physical state and the digital state.
NASA talks about digital twins as a physical object or process and a representative digital object or process and a way of communicating between the two to keep them in sync. In our restaurant example a human being is the communication link. If a customer leaves the restaurant then their server marks the table as clear in the system. If someone marks a table as clear in the system before the table is empty then the server has to go and chivvy the customers along to clear them out.
This raises our first interesting question - in NASA’s definition of a digital twin does the communication link have to be digital itself? How much ‘human integration’ are we allowed? Where do we draw that line?
A digital bridge twin
A construction company algorithmically designs a pedestrian bridge using 3D software. They use software to run tests on thousands of bridge designs, finding one that it is safe and sustainable and that pedestrians can walk over easily without bumping into each other, even at peak times. They 3D print the best bridge exactly as it is in the model using an industrial machine that can 3D print using steel.
Reflections
This example turns the digital twin timeline around. We expect a digital twin to be built to model the physical object or process. Not the other way around. Some folks describe this as a ‘digital twin prototype’. This example is based on an actual bridge 3D printed by Arup and designed and modelled by a raft of folks with support from the Turing Institute. In their example they add a lot of sensors to the real life bridge that they use to understand whether users of the bridge are doing what they expected them to do. I imagine if I added the word sensors to the copy above then that 40/60 split would tip the other way!
A digital social housing company twin
A social housing company manages hundreds of properties across a city. They are responsible for maintaining the properties, ensuring that they are heated, have hot water and electricity and do not have damp or mould issues. The social housing company uses a housing management system that records information about which properties are currently occupied and which need tenants, when boilers were last serviced, the status of repairs for the property and more. They display this information on a map of the city.
Reflections
I think the results of this poll highlight an expectation around what constitutes the tangible aspect of a digital twin. This system is clearly replicating the physical world but in a way that we feel is conceptual. There is the concept of a tenant, the concept of a boiler that may need repairs but they’re not direct reflections of the tangible world. We have data that tells us that a specific boiler is due for a service but we do not have data that tells us how that boiler is currently functioning or what it’s doing right now. We don’t even have historically accurate data about it’s function from yesterday or a month ago.
When a boiler malfunctions a tenant might phone the association’s boiler help line who then flag the boiler for repair in the system bringing it up to date but something about this process of communication just doesn’t feel right - the tenant might not phone, they might be out or on holiday and the digital representation and its physical twin might be out of sync for a considerable period of time.
A digital mould prevention twin
The social housing company is worried about the health of tenants and wants to preemptively tackle a damp and mould crisis that is looming because of rising gas prices. The company installs environmental sensors in their tenants’ rooms that monitor temperature, humidity, air quality and air movement. They also install a device that can be used to activate a tenant’s heating system remotely. The company’s city map is updated with live sensor data. Agents within the organisation can now remotely trigger the heating in a tenant’s house if they think that there is a significant risk of mould growing in the house.
Reflections
Adding connected sensors to a system gives us much more timely feedback and removes the (considerable) time lag that requiring a human to pick up the phone introduces. This clearly reassures some of you that this is a digital twin but I imagine that this is still too conceptual for a good proportion of you. Personally I quite like this one because it allows agents at the housing association to make decisions and trigger action remotely.
“Digital first transformation” was all the rage a few years ago. This caused some mirth at Waterstons because technology led change has always been anathema here. We never want to implement a technological solution for the sake of it - there always has to be a business reason to undertake any bit of work that we do. This is a point of both principle and pride to us.
I refer back to this ideal often and when I think about digital twins the question on the tip of my tongue is always ‘why?’. What is the driving need that means that a digital reflection of a physical object or process is the answer? If you take this approach then you can quite quickly decide on the required granularity, fidelity and timeliness of your digital twin.
Is this just a low fidelity, low granularity digital twin?
A digital sustainable building twin
A university research centre has a fancy new building. The building is brimming with both environmental sensors, like air quality, circulation monitors, temperature and humidity, and physical sensors that determine things like whether individual windows are open, lights are on and whether rooms have people in them. The university has an exact 3D replica of this building built using Unreal Engine. Sensors in the real building feed sensor read outs in the virtual building. When lights are turned on and off in the real world they are correspondingly flipped in the virtual world and vice versa.
Reflections
Well, no surprises here. This is the classic definition of a digital twin and no doubt our great friend and partner Miles at the Institute for Safe Autonomy will be pleased to hear that his building has passed the digital twin test according to our avid and smart readership. Miles your new “official digital twin” plaque will be winging its way towards you shortly.
A digital machine twin
A coat hanger manufacturer is building a new factory line. They have bought a complex and expensive set of robotic arms that can fold metal quickly and efficiently. They want to parallelise as much of the process as possible but the arms keep getting in each others’ way as they work bringing the line to a halt. They build a 3D replica of their line with physically modelled robot arms that they can use to efficiently try different set ups without risking damaging the arms.
Reflections
I would have thought this one would be higher. This is the archetypal technologically advanced digital twin example. It’s what big businesses like Siemens and Nvidia are driving people towards - digital representations of actual physical objects modelled with physics. High fidelity, high granularity and extremely expensive both in development and real time computation.
This is an extremely interesting space to be in. The technology used for this kind of modelling in realtime has really been driven by the development of video games over the past thirty years. Games were really the first place to demand the real time physics engines that we now see used in films, architecture, manufacturing and more. We’ll definitely be digging more into this space in the future.
A digital factory process twin
A lawnmower manufacturer wants to make their complex manufacturing process more efficient. The do a full time and motion study and then create a spreadsheet that models the process in some detail. It’s extremely complicated but it effectively maps the branching manufacturing process along with all sub assemblies along with all the process and wait times and cells that can be modified to represent wastage percentages and error rates. The spreadsheet lets them see where bottlenecks would occur if they made individual processes quicker. They tweak parallel parts of the process in the spreadsheet to see whether they could adopt a Just In Time approach and then modify the line to reflect those changes.
Reflections
Ah spreadsheets. Most of the world runs on complex spreadsheets filled with macros and clever formulae. In my experience now matter how clever and smart a business looks from the outside there is always a spreadsheet holding it all together underneath. Personally, I love a spreadsheet that models a real world process, despite how risky they are and how much of a pain in the backside they are to maintain. As long as you are the person who created them in the first place they’re easy to modify and quick to debug and often deliver a perfect solution.
Of course if you’re not the original author then they’re an absolute nightmare and you will inevitably need the services of one of our most in-demand teams to sort it all out.
I actually really like the concept of a spreadsheet as a digital twin. I’ll try to explain why in the conclusions that follow.
Conclusions
At this point, I’m beginning to think that any digital representation that allows you to make better business decisions could be described as a digital twin but I could be wrong.
It’s very easy to see how pursuing ‘digital twins’ could take you down a very technical route. The technologies behind the most prominent examples of digital twins is fancy, exciting and attractive. You could easily spend multiple millions of pounds instrumenting a factory line, building it in a complex modelling package and making sure that all your processes are represented perfectly only to find that you can’t make decisions any better than you could before.
Alternatively you could end up describing any spreadsheet that vaguely represents the real world as a digital twin losing the benefits that a ‘better’ digital twin might be able to bring you - benefits around simulation or timeliness or flexibility.
I think our next explorations need to be around granularity, fidelity, timeliness and simulation. I’m also wondering how much we can learn from our friends in the data team who spend a lot of time thinking about these exact things from a reporting perspective.
There’s clear a lot more for us to think about. Let us know in the comments if you think we’re missing anything.
And in case you missed it…
The remarkable explainer channel 3blue1brown has done a great job of explaining how attention works in transformers. Transformers are the building blocks of modern AI. They are the root of ChatGPT, MidJourney and all of DeepMind’s clever creations. This is a great, clear explanation that, even if you don’t follow all the maths, does a wonderful job of illuminating the subject.