I recently returned from another fantastic WinterSim conference, where the hot topic was digital twins. From vendors, to talks, to roundtable topics, everyone was talking about digital twins and how you use them as real time decision making tools.

What is a digital twin?

As this is an evolving space many of the discussions focused first on each person setting out their definition of a ‘digital twin’ (or ‘symbiotic simulation’ or even ‘system surrogate’, as some were calling them).

The general consensus was that a digital twin is a simulation that’s kept in sync with current system status. It can be used to assess past performance, predict future performance and make short term operational decisions.

However, many at the conference were frustrated by the term and felt strongly that this is just putting a new badge of what we’ve all been doing for years. I agree and disagree with that, and no, I’m not politely sitting on the fence (I don’t do fences!).

The general consensus was that a digital twin is a simulation that’s kept in sync with current system status. It can be used to assess past performance, predict future performance and make short term operational decisions.

Are digital twins new?

No, digital twins are essentially simulations and simulation has been a key analysis technique since the 1970s.

However, no one can deny the speed that the term is gaining traction. It’s exciting many industries, particularly manufacturing, with Gartner listing digital twins in their Top 10 Strategic Technology Trends for 2019. So why is this new name starting to stick? For a start, it focuses on the outcome, not the technique.

For the first time there is a simplistic way of describing what simulation delivers and the value it provides. As we all know, the best marketing focuses on the benefits to the end customer and not features. When an organization builds a simulation they don’t get a simulation, simulation is what they do. They get a mirror of their system that they can use to dissect every aspect of their performance, perform risk-free experimentation and gain insight into the future. A digital twin is the benefit they get. As a result, it’s no surprise that business leaders across the globe are now eager to embrace simulation. They want the benefit, the outcome, the digital twin.

To validate this (as a true simulation geek I have to validate everything!). I did the ultimate benchmark on the phrase digital twins – the ‘Mom’ test. Now, after 20 years she and my family understand what I do! This is why I think digital twins will stick.

The only risk is that the overuse of the term might saturate it before it has a chance to catch on.

I did the ultimate benchmark on the phrase digital twins – the ‘Mom’ test. Now, after 20 years she and my family understand what I do!

So, a digital twin is just a simulation?

No. The real-time part is the crucial difference. It has to be a simulation that is kept constantly up-to-date with current system performance and be reflective of current process. It needs to be adaptive not static. The ability to be constantly kept in sync means it’s always representative of the live system. That is what allows it to be used as an operational decision-making tool. For example, being used to decide tomorrow’s optimum production schedule, or ensuring patients are in the right bed to minimize waiting times and outliers. In both of these examples, the ability to run forward allows today’s decision to be made more effectively.

A digital twin turns a simulation from a predictive to a prescriptive analysis tool. Dare I even add another buzzword? It creates a ‘simulation AI’ tool.

Should I build a digital twin or a simulation?

In general, we don’t see many of our users building digital twins from the get-go. With all simulations it’s about first understanding the problem you’re trying to solve and doing the correct problem formulation to understand what you will need to be able to answer that question. Mostly that initial stage is about figuring out what you don’t need to model. What is the simplest abstraction of your process that you can build to answer your question?

We challenge ourselves before adding detail to our simulations – ‘will this really help answer our problem’? If the answer is no, we drop it. Your first simulations should be rapid to build and roughly 70% accurate. Often that’s enough to answer the question and other times it acts as a guide to tell you where you really need to dig deeper and add detail. We see users iterate this process by building small, simple conceptual simulations to answer critical questions rapidly. A day is about as long you need to spend on each of these ‘learning loops’. It’s all about speed to answer.

How do I turn my simulation into a digital twin?

If you use a technically purist view of digital twins definition (purist is key here, stick with me and you’ll see why) then creating a digital twin is a different process. You are building a tool, not doing analysis.

Digital twins need to be verified and validated to be 100% accurate. The detail does matter here, as does the flexibility of the tool to analyze many problems, the speed of execution to run forward constantly and the ability to optimize across thousands of scenarios. Now you are deploying a simulation into production. It is much closer to software development than analysis, so you have to code it for 1 to n, not 1 to 10. Today you have 200 staff, but 9 months from now there could be 1,000. Every part of the simulation has to be built to extend and everything needs to be a variable.

It has to sync from live data sources and pull every piece of input data dynamically; from system sensors providing current status, to shift patterns, to product routing, to process rules and structure. Yes, even the structure of the simulation itself needs to be able to be updated, ideally with no manual intervention. At the very least it has to self-validate enough to highlight when the twin is no longer accurate and needs to be updated.

I could go on, but this should give you the picture of why when you build a technically pure (there’s that condition again…) digital twin, you’re building a simulation system, not just a simulation which takes a different kind of thinking and overhead to build.

Are they valuable? Absolutely, they are transformative. Are they the right solution for every organization? No.

How many organizations can afford to invest in a digital twin of this type? Not just from the perspective of building of it, but also the ongoing maintenance and investment in equipment needed, as well as the steady stream of requests from users for new features? Very, very few. So, if we want to be purist about it (that’s the last time I’ll dangle that phrase without explaining!) it’s hardly surprising that I didn’t hear a single case study at WinterSim that really matched this definition. On this basis I disagree with my WinterSim colleagues, very few simulations built are actually digital twins.

How many organizations can afford to invest in a digital twin of this type? Not just from the perspective of building of it, but also the ongoing maintenance and investment in equipment needed, as well as the steady stream of requests from users for new features?

Are digital twins just another buzz word then? An unobtainable goal?

No, what I’ve described is the 100% technically accurate definition. If the simulation and real life processes don’t coexist in a symbiotic relationship then for some people, it’s not a digital twin. To build that type of digital twin, a simulation system continuously fed second by second with live data streams that dynamically updates itself to stay current and answer any problem, is rarely practical and I believe for today it’s an aspiration and shouldn’t be the definition.

Let’s get practical. While Internet of Things (IoT) and other automated data collection devices have made incredible advancements in recent years, most organizations don’t yet have access to these devices. Even if the data is there, it still needs to be tamed into a manageable streams. Process mining and other AI techniques are also advancing, but again, they’re not yet accessible to all. They are a long way from being able to sense and adapt to system changes automatically, someone has to tell the digital twin that the process has changed and build the changes into the system.

What level of digital twin is achievable?

Digital twins in their purist sense are technically unachievable for most organizations today. On this point I agree with my WinterSim colleagues. For today, a lot of the simulations we all build are digital twins because that is what is useful to an organization, what is achievable for most right now.

For today, a lot of the simulations we all build are digital twins because that is what is useful to an organization, what is achievable for most right now.

More than that, I believe those simulations to be just as transformative as a ‘pure’ digital twin. They’re built for a specific purpose so the noise is automatically filtered out. They bring focus and clarity and can be built quickly and made accurate with little effort using readily available data sources. To be honest, I’m not yet convinced that even when the data and tech has caught up to readily allow for a pure digital twin that they will be of general value, or deliver the best ROI from simulation.

Let’s own digital twins.

We failed as an industry for 30 years to articulate clearly and captivatingly what we did, let’s not fail again. Rather than blindly adopt an industry buzzword, I challenge our industry to own digital twins and drive a change in the definition. We should take our expertise and educate the industry about how to obtain a digital twin that will add value to organizations right now. A practical, useful digital twin.

Let’s make it understood that a digital twin fed with accurate data is usually enough. Continually feeding it with data is an option, not a requirement. Valuable simulations can be built without IoT devices, for very little investment. They can be quickly changed or even thrown away for new ones to be built in their place that answer your evolving challenges.

How should you pitch simulation in the meantime?

So if you’re trying to sell simulation to your boss or anyone else who’s not sim aware, then I’d try pitching a digital twin, not the full technically accurate digital twin. Just try pitching “simulation is like having a digital twin of our business, we can experiment learn and understand from it to predict and optimise our performance” and you’ll see their eyes light up.

If you need any help or advice, get in touch – our team are always happy to help!