Literally billions of euros in are invested digital twins. Market analysts have reasons to believe that digital twins are really a good investment. Why do they believe that? What is the business case behind all these investments? As a company that provides the technology behind digital twins, VORtech often discusses such questions with our clients. This gives us a good insight in the arguments that are typically made.
In previous articles I wrote about digital twins from a technical viewpoint and I discussed the conceptual confusion around it. In this article, I will look at digital twins from the business perspective. Where is the value in digital twins? What are the costs? The sort of questions that your CEO and CFO will ask if you consider joining the hype.
Why would anyone want a digital twin?
There is not a single answer to the question why anyone would want a digital twin. For owners of a production facility that use digital twins the reason is different than for suppliers of devices that develop digital twins. In other words, it depends on your place in the value chain. In this article, I will only discuss two of these two actors: the users and suppliers of digital twins. I will not discuss the other actors in the value chain, like researchers/developers, it-companies, and cloud platforms. These derive value mainly from the services they deliver and less from the digital twins as such.
The reason for having a digital twin is also different across sectors: digital twins in health care serve other purposes than those in manufacturing. At his moment, the sectors that are leading in the adoption of digital twins are manufacturing and energy and utilities. Together, they account for over 50% of the investments in digital twins. So, I will concentrate on those.
To complicate things further, not all digital twins are created equally and thus their reason for existence varies. Some are more sophisticated than others, from simple data historians with a nice graphical interface to AI-powered autonomous control systems. Some apply to products, some to processes and some to entire systems. In this article, I will start with the most basic form of digital twin and move on from that.
Why are digital twins useful for their users?
Let’s first look at the benefit of digital twins for users in manufacturing companies, energy suppliers and utilities, going from the most basic form of digital twins to the most elaborate.
Monitoring and controlling assets
On the most basic level, the digital twin is no more than a data historian with an intuitive user interface. In this case, the need for a digital twin is not too different from the need for a database with a dashboard. Or a SCADA system. It’s just that you want to monitor your physical assets and also collect data about them for future use. Such future use could be simply to know what happened at a certain moment. Or it could be the analysis of the historical data to find optimizations. Again, these types of digital twins are little more than what we already have had for decades and the question about their value is no different from before.
Even so, this basic type of digital twins has taken on a surprising new relevance in the days of Corona. As people are banned from the production facilities to keep them safe from Covid19, a digital copy of the factory that can be operated from home became a huge advantage overnight. Perhaps, this validated the entire digital twin hype in a single stroke.
Running what-if scenario’s
Most people would not call a data historian a digital twin, even if it has a nice 3D graphics interface. The added value of digital twins really comes from the functionality to simulate and run what-if scenarios. This means that you can experiment with operational settings of an asset or process without the risk to damage anything. It’s like a computer game where you can just restart after being shot down (or have any other fatal event happen to you). If the simulation shows that your machines will be harmed, you just try something else. Right until you’ve found the sweet spot where your operation is optimal and nothing gets broken.
A nice example is 4DCOOL (full disclosure: I’ve worked on it myself). It’s a digital twin of the indoor climate of data centers. Data centers typically spend half of their operational costs on cooling. And this cooling is always kept at a somewhat conservative set point because overheating of a server is to be avoided at, almost, all costs.
4DCOOL makes it possible to find a much cheaper setting of the cooling without endangering any of the servers in the room. By combining temperature data from sensors in the server room with a flow simulation model, 4DCOOL builds a model of the temperature distribution in the room. As it uses actual measurements the model closely corresponds to the real thing at any moment.
Operators can experiment with the settings of the cooling to see in the model what the effect is likely to be. In doing so, they also gain an understanding of the way the setting influences the temperature throughout the computer room. This makes the operators more confident in their actions and helps to think of structural changes that might be helpful. Even for a small data center, 4DCOOL has delivered more than 100k in savings on an annual basis.
Not all sectors will have such large opportunities for optimization from the ability to run what-if scenarios. But for anyone considering the use of digital twins, it may be worthwhile to consider how far they might come if they can freely experiment with their assets.
Training your operators
The simulation aspect of digital twins is also an excellent tool to train new operators. This, in itself, is a significant advantage as there is a certain worry in several sectors that experienced operators will soon retire and new ones with the right qualifications are not readily available. Having a good training facility for new operators can therefore prove to be very important.
Introduce artificial intelligence in your systems
One step further beyond mere simulation are digital twins that have some autonomous intelligence built in. With the data that is collected from the sensors, machine learning algorithms can be trained. Such algorithms could detect abnormal behavior quickly and learn to categorize the abnormal behavior and suggest corrective action. The value here is that problems may be detected earlier, and better actions can be initiated.
A particular form of functionality where algorithms learn to understand the behavior or assets through the digital twin is found in predictive maintenance. Here, the abnormal behavior that the algorithm picks up is the degradation of the asset and the corrective action that it suggests is the advice on when to do maintenance. The business case for predictive maintenance is well established and as digital twins are a conceptually nice way to implement it, this business case carries over to digital twins as well.
Over time, the algorithm might acquire the same experience as the operators as it learns to deal with all kinds of events. This, in turn, would allow using less experienced operators. The advantage is obviously in the fact that a less experienced operator can be cheaper and easier to find than experienced operators. This also ties into the point that was raised above: as older operators retire, they take with them a huge load of knowledge and experience that is hard to replace. It would be nice to have their experience captured in the digital twin.
Algorithms that are trained through machine learning can also be used in control, where they basically take the dull stuff away from operators and help to avoid human errors. Some operators will probably always be needed to handle the really rare or unexpected events. And surely technology is not yet at a point where it is fail-safe enough to be left to its own devices. But it is worth considering what the savings would be if less operators would be needed.
So, summarizing, the upside of digital twins for manufacturing companies and utilities would be in a more efficient operation, in working remotely and with less (and less experienced) operators and in predictive maintenance.
Why are digital twins useful for suppliers?
Suppliers have their own reasons to develop digital twins. On the one hand, they have an interest in digital twins for their own production but on the other hand it also gives them competitive advantages in the marketplace.
Improving development and production of devices
First about the advantages for the suppliers’ own operation. Digital twins are used more and more in the design and build phase of a device. During design, it comes naturally with the use of digital design tools like CAD tools. A digital twin not only captures all the design information but, if properly set up, can also collect the history of each instance that is produced. This is helpful because it gives all the people that are involved in the design and production a shared set of information which avoids the typical problems that arise if every stage uses its own information systems. In particular, the digital twin would be useful in quality management as the entire history of any particular product is available and any problem that is detected can be traced back to its origin.
A competitive advantage
The second advantage of digital twins is in the marketplace and that is probably the most important one. Suppliers can augment their device or machine with additional intelligence from pre-trained algorithms which would make the device more reliable or more efficient for the user. And it offers training opportunities for customers to learn and work with the device.
Obviously, the supplier that has the best algorithms and interface would have a serious competitive advantage as long as the product itself is also good. At this moment, having a digital twin to go with a device or a machine will allow the supplier to charge more. But soon enough, competitors will come along and supplying a digital twin will become a necessity to remain competitive at all.
Data for additional services
A third reason for suppliers to work on digital twins is that, if properly organized, it allows them to collect data from their products as they are used operationally. This allows the supplier to sell extra services to the customer in the form of advice on better settings or proper management. And it allows them to collect data that is essential for product innovation. When this operational data is combined with the data collected during the production of the device it will also make it possible to connect later-stage problems to early-stage anomalies in the production.
Note that not all customers will be comfortable with giving their suppliers real time insight into their operations. But forward-looking companies will see the benefits and sooner or later many asset owners will require these kinds of services from their suppliers.
Summarizing, the benefits of digital twins for suppliers are in better control of their own development and production process, in more margin on their products and in extra services that can be supplied.
What does a digital twin cost?
The costs of digital twins, again, depend on the context. If you are using a digital win only to store information, the cost is not really different from that of an online database. It’s about defining your data model, connecting to the data streams and building a nice user interface to work with the information. So, nothing here that is really different apart from the fancy stuff in the user interface.
But, as I said before, that is not what most people would call a digital twin. A real digital twin, with a model and all sorts of algorithms built in, is more expensive. For owners of assets, the costs usually translate in a fee for use and costs for integration in the IT environment. The suppliers bear most of the investment and risk.
The costs for asset owners
Owners of assets would normally get access to digital twins of their machines from their supplier, along with the machine itself, and then have it integrated in their it-environment by an integrator. If the digital twin is about the production process or the logistical process, an asset owner will usually hire a specialized it-company.
A digital twin will typically be part of an IoT framework in the company. These days, there is a lot to choose from in terms of IoT frameworks, from proprietary frameworks like MindSphere and Maximo to open source stacks like ThingsBoard and Kaa. All of them typically adhere to some external standards for data collection, like MQTT for instance, and they all have some hooks to incorporate models and complex algorithms.
Implementing such an IoT framework in the production facility is usually done by a specialized service provider who then also takes care of integrating the models and algorithms that are the essence of the digital twin functionality. More often than not, the digital twin will run remotely on the server of the supplier but in some cases, it may also run on the local infrastructure.
The development costs for suppliers
Suppliers will usually be the developers of a digital twin for their product. For them, the costs depend on the level of complexity that is needed. There are many environments like MATLAB Simulink or Modelica that let you build simulation models relatively easily. Sometimes, the model can be entirely data-driven, and no explicit modelling is involved. In that case, the entire toolset from machine learning is available to train a model. Then you are talking of things like scikit-learn or Tensorflow, to name only a few of the many that are out there.
If the behavior of your asset is relatively simple, then building the simulation model is not necessarily that much work. Say from a few weeks to a few months.
But if the behavior is more complex, like in process industry plants or in biochemistry, things may get hard. For cases like these, there are multi-physics tools like Comsol that do a pretty good job for explicit modelling. But then you will probably need simulation engineers to do the work and it will probably cost more. In general, you would be looking at an effort of a few person months. Or even a few years if some basic research is involved to understand what is going on. In that case, however, the effort is probably not only aimed at building the digital twin as such, but rather at enlarging your understanding of the processes that are relevant for you. The digital twin is then only a form in which this knowledge is instantiated.
The final level of complexity in a digital twin is that of the intelligence. This refers to digital twins that operate for themselves at some point. Think for instance about a digital twin of a machine that orders new parts for its physical twin when it notices that maintenance is required. Any generic estimate for this kind of functionality is useless. Something simple can be made in a few days but if you get really going in this direction, the sky is the limit.
Maintenance of the digital twin
As every asset manager knows (and a digital twin is essentially also an asset, just like its physical twin), building an asset is only part of the game. Once it is there you have to maintain it. That is also true for digital twins. Usually, the maintenance will be done by the suppliers.
The good news is that maintenance of a digital twin is probably far cheaper than that of a physical twin. I say ‘probably’ because, as far as I know, there is still very little experience in this field. But I suppose that as long as the structure or the processes of the physical twin do not change, there is hardly any need to do maintenance on the digital twin. It is just one more application running on your servers.
The most that you may have to do is some sanity checks on the functioning and data of the digital twin and the usual updates that go for all applications. For a well-constructed digital twin, therefore, the maintenance cost is small, even though you should still count it because it will be recurring costs for many years to come. Once again: this only holds if your physical asset is relatively stable. If not, then the costs for adapting your digital twin over and over again can still be very significant.
Now what?
The benefits and costs described above are obviously pretty generic. Filling in the numbers should be done by each company for itself as this is all very dependent on the actual context. When it comes to the effort to develop a digital twin, you can contact specialized companies like VORtech. They will gladly help you to choose the right technology and approach and to estimate the investment that is needed.
From the many enthusiastic articles about digital twins, you might assume that the business case is always there. That is probably right. But after all, it is better to make sure for yourself as the introduction of digital twins is quite a step in the development of your business. That should not be taken lightly.