Unleashing the Power of Digital Twins: A Comprehensive Guide
Digital twins are transforming industries by providing a comprehensive digital replica of physical environments. This article will delve deeper into the concept of digital twins and guide you through setting up and utilizing a digital twin.
Digital Twin is a buzzword that has been making waves in the tech industry. But what exactly is it? There have been multiple definitions all over the place so far. Simply put, a Digital Twin is a digital replica of physical environments that allows modeling of relationships and interactions between things, places, business processes, and people. An example is a smart building where sensors are part of physical assets like entry gates, elevators, lights, etcetera. Measuring the components within all those assets will contribute to the digital copy of the physical building.
But why do we need Digital Twins? Well, Digital Twins can provide the ability to create reusable, highly scalable detailed digital models of comprehensive environments that fuse data across the physical and digital world to track both past and present events, simulate possibilities, and help predict future events for those environments. Assume the activity of the entry gate at a smart building is measured, analyses about past can be made and the current state can be monitored. Its past behavior can even be used to predict the future state with the help of Machine Learning algorithms. Things like the occupation of the office can be predicted or a potential malfunctioning of your gates which can prevent a downtime.
But where do you start when you want to use a Digital Twin to predict future states, simulate scenarios or even automate decisions based on a digital replica of your physical environment? These six steps or stages are essential to get the maximum value out of your Digital Twin. One important note: before following the process below, the use case for Digital Twin should be clear! Suppose you want to make a digital replica of a process. There must be a reason you want to do this, like for example process optimization to reduce energy consumption or train employees when something in the process goes wrong.
Things / process
As said, a Digital Twin is a digital replica of a physical environment. The first step is to start describing all the components in your Digital Twin. About which assets or processes are we talking about? Take the example smart buildings and the entry gates again. Questions that you should answer before even start building your Digital Twin are:
- What is an entry gate? What is the definition?
- What data of the entry gate are we measuring? Entry logs or activity status? Identification? Datetimes?
- What is the unit of measure of each component of the entry gate?
- What is the frequency of measuring? Realtime? Batch?
Just start to document the metadata of your asset or process, it is like a definition framework. The reason this is so important is that there needs to be a clear understanding for your organization about Digital Twin. Important choices like what kind of sensors or the solution design are based on what is defined in this step.
And there is something else that should be clear upfront. That is to define a hierarchy model. In a hierarchical model, each level of the hierarchy represents a different level of abstraction or detail. Usually, a product is labeled in a factory. The machine which does this has different components that can be measured: the lowest level of detail. The component is part of the machine, which is part of the production process. The production process is part of a supply chain which also includes a storage process, a distribution process, etcetera. The higher the described level, the more the Digital Twin is future proof for Digital Twin orchestration: a network of several Digital Twins.
The defined hierarchy model will also have an impact on the elements that you want to measure in your Digital Twin. Assets should be able to link to other assets, so identification and relations needs to be included in the data that will be captured.
Connect & Collect
Time to connect and collect! In a lot of cases an asset is already generating data. In that case it is only a matter of checking if the right things are already measured. Use the definition framework made in the previous step to make necessary changes to the current setup. It can also be the case that an asset is not equipped with a sensor yet. Use the definitions in the previous step again to purchase the right sensor or new assets to execute the use case.
When the assets are generating data, this data needs to be collected. Typically, Internet of Things (IoT) hubs are used for this. These kinds of solutions can collect data from sensors directly or on gateway level if they are connected to the hub. From there, connected assets can be monitored and the network of connected assets can be easily extended with new ones.
Model & Store
This stage consists mostly of data engineering, but looking through the eyes of a data architect is crucial here. What is the data model behind your Digital Twin and where is it stored? I keep saying it: the first step of Things / Process is so important. For the solution design, the definition framework is leading as well. Aspects to think about are choices for the way the data is modelled. Relational? Data vault? Graph network? I am not into more detail but doing some “name dropping” here. In any case, there needs to be a choice made here.
Regarding storage it is important to consider if only real-time data is needed. If the use case for example only needs monitoring of the asset state or only a rule-based model to make decisions, it is not needed to set up a long-term storage for your Digital Twin. However, usually the use case involves simulation and optimization (at least as a sort of “next step” or end goal) for which this long-term storage is needed. “Name dropping” alarm! Graph database? Data Lake? Data Warehouse? Data Lakehouse??
An important and not yet mentioned aspect here is data quality. In the world of sensor data quality is always a theme. I have seen cases where sensors were placed upside down (“why do we get these negative values?”) or where the exact same sensors in the same asset were reporting very differently due to measurement error. Make sure yoy take these cases into account when data is stored.
(Geo-)visualize
Time to visualize your Digital Twin! Until this step, the task was to get the fundamentals right on which we can build visualizations or even simulations and optimizations that I will be describe in the next steps.
It is important to keep in mind here is your use case. In the visualizations stage typically descriptive and diagnostic analyses will be made. In many cases this will be the end, just reporting the state and its past behavior is often sufficient. Think about use cases for alerting or a rule-based model. For example, an Asset Manager at a port wants to know if the force on a quay wall reaches an alarming threshold value. At this value, the manager needs to take immediate action to avoid damage to the assets.
Think about the level of visualization you want to reach. In some case a simple dashboard without any spatial component can also be enough, but I also have seen cases in which a full 3D animation was made in which past behavior of assets was displayed. A municipality displaying to its residents what a new bridge’s effect is on their environment is such an example. The important thing is to keep in mind what is the best for the end users. Make sure they are part of the process of building your visualization.
Simulate
The simulate part builds on the visualization that was chosen. Simulation comes into play if you want to predict what will happen or when you want to run certain scenarios that can happen. The fundament is learning from the past and applying these learnings to predict the future. Also, Machine Learning can come into play here.
Typical examples of simulation are Virtual Reality (VR) cases in which a user can behave and make choices that trigger new events in a virtual world. But what about other solutions? For me, a prediction based on data of a Digital Twin is also part of this simulation category. I agree, prediction and simulation are two different things, but they both tell something about the future. Back to the example of the Asset Manager. There might be a situation where weather conditions, type of ship, the number of operational cranes are known for the berth visits of the next day. If a Machine Model is trained on historical situations, the Asset Manager will know a day in advance if alerting values for the force on the quay wall will be reached and he can take action on forehand.
Optimize
The optimize step is the last part of fully utilizing your Digital Twin. As I said before, sometimes a use case only needs a descriptive or diagnostic solution (or analysis) and it is sufficient to “stop” at the (geo-)visualize part. Optimization is really on the more complex side of advanced analytics, and it answers the question: “how can we make it happen”?
In optimization there is always a future state that you want to reach, which will determine your actions or behavior to reach that state. For example, in a distribution 10.000 bottles of water need to be delivered to a festival as soon as possible. Delivering all 10.000 bottles of water at the festival is the future state in this case. Suppose you used your Digital Twin for simulation before and there are several parameters that you can influence (type of transport, which warehouse, etcetera). The future state can be used to set the optimal setting of your parameters of influence.
You can even take it a step further and automate choices or processes. In case the desired future state is known during a process and the optimal setting of input parameters can be set, no human interaction is needed. The necessary condition here is that your system is designed to operate autonomously.
Wrap up
In this article I started to describe what a Digital Twin is exactly and some of the benefits of it. Next, a 6-step approach to unleash the power of your Digital Twin was described:
- Things / Process
- Connect & Collect
- Model & Store
- (Geo-)visualize
- Simulate
- Optimize
It is important to note that data is core for the whole approach. Without sufficient or accurate data Digital Twins are useless. To learn more, stay tuned for part two later this year in which I will dive deeper into Artificial Intelligence concepts within a Digital Twin.
