Getting started with digital twins
By Branko Dijkstra, Principal Technical Consultant, MathWorks Australia
Friday, 16 April, 2021
One of the primary catalysts for engineering and design teams to be able to actualise the benefits of Industry 4.0 has been the emergence of digital twins.
After many years of excitement and discussion around Industry 4.0, manufacturing, energy and transportation, companies are increasingly realising the benefits of Industry 4.0 applications. One of the primary catalysts for engineering and design teams to be able to actualise these benefits for their companies has been the emergence of digital twins.
What is a digital twin?
A digital twin is a real-time digital representation of a real-world physical asset in operation that reflects the asset’s current condition and provides relevant historical data. Companies can leverage digital twins to analyse the real-time performance, optimise the operation, predict future behaviour or refine control of its assets, such as pumps, engines, power plants, manufacturing lines or fleets of vehicles.
Why use a digital twin?
There are numerous ways engineering teams can design and deploy digital twins to deliver value to their companies and optimise future operations.
Digital twins capture the physical asset’s history — being updated periodically to represent the real asset’s current state. Over time, these past states become the asset’s history. The type of information included in this history differs, based on how the digital twin is being used and what is captured in the current state. For example, a digital twin used for fault classification will capture a history that includes a specific pump’s operational data from its healthy and faulty state. In the future, engineers can then compare the operational data from that pump to the digital twin histories of other pumps to understand how they behaved under similar faults and predict the effect on the fleet’s efficiency.
The ability to monitor a whole fleet of assets using digital twins brings advantages to planning operational events and improving maintenance strategies.
For example, when a specific pump is nearing failure, the digital twin can assess how this will affect the efficiency of the fleet and potential costs. This informs the company when making the decision between ordering a new part and waiting for it to arrive or paying more for expedited shipping to get the part as soon as possible.
Simulating future scenarios
Companies can use digital twins to simulate future scenarios to see how factors such as weather, fleet size or different operating conditions affect performance. This approach helps manage assets and optimise operations by informing maintenance schedules or flagging expected failures in advance.
Digital twins can be leveraged by companies for a variety of applications, including anomaly detection, operations optimisation and predictive maintenance.
The digital twin models run in parallel to the real assets and flag operational behaviour that deviates from expected behaviour in real time. For example, a petroleum company may stream sensor data from offshore oil rigs that operate continuously, while the digital twin model looks for anomalies in the operational behaviour to flag potential equipment damage.
Companies can apply variables such as weather, fleet size, energy costs or performance factors to trigger hundreds or thousands of simulations to evaluate readiness or necessary adjustments to current system setpoints. This approach makes it possible to optimise system operations to mitigate risk, reduce cost or gain system efficiencies.
In industrial automation and machinery applications, companies can use digital twin models to determine remaining useful life and the most opportune time to service or replace equipment.
In a characteristic smart connected system topology as shown in Figure 1, the digital twins could be executed on the smart asset, at the edge or on the IT/OT layers depending on the required response time of the application. For example, predictive maintenance, a common Industry 4.0 application, generally requires making real-time or time-sensitive decisions — meaning the digital twin should be integrated directly with the asset or at the edge.
How does a digital twin work?
An IoT application drives what needs to be modelled as part of a digital twin. A digital twin model will include the required components, behaviours and dynamics of the IoT asset. Modelling methods generally can be grouped into two types: first principles or physics-based methods (eg, mechanical modelling) and data-driven methods (eg, deep learning). A digital twin can also be a composite of various modelled behaviours and modelling methods and is likely to be elaborated on over time as more uses are identified.
The models must be kept up to date and tuned to the assets that are in operation, which typically involves direct streaming of data from the assets into algorithms that tune the digital twin. This makes it possible to consider aspects like asset environment, age and configuration.
Once the digital twin is available and up to date, it can be used any number of ways to predict future behaviour, refine the control or optimise operation of the asset. Some examples include simulating sensors that are not present on the real asset, simulating future scenarios to inform current and future operations or using the digital twin to extract the current operational state by sending in current real inputs.
How to build a digital twin
Engineers will be increasingly asked to develop digital twins for their company given the above benefits. Here are two methods design teams must keep in mind as they prepare, build and apply their digital twin models.
A company looking to optimise maintenance schedules by estimating remaining useful life (RUL) will use a data-driven model, as the type of the data from the asset will determine which model teams will be using. Similarity models can be used if the company has complete histories from similar machines. If only failure data is available, then survival models can be used, and if failure data is not available, but the safety threshold is known, a degradation model can be used. If failure data is not available but a safety threshold is known, degradation models can be used to estimate RUL. In this RUL scenario, the degradation model is constantly updated using the data from the asset measured by different sensors — such as pressure, flow and vibration in the case of a pump.
If a company wants to simulate future scenarios and monitor how the fleet will behave under those scenarios it would use a physics-based model, which is created by connecting mechanical and hydraulic components. This model is fed with data from an asset, and its parameters are estimated and tuned with this incoming data to keep the model up to date. Engineers can then inject different types of faults and simulate the pump’s behaviour under different fault conditions.
How to apply digital twins
Design teams need to create a unique digital twin for every individual asset. This means that for each asset at a different location, teams must create a unique digital twin that has been initialised with the specific asset’s parameters. The total number of unique twins will depend on the application. If teams are modelling a system of systems, they may or may not need a twin for each system of components depending on the required level of precision. For example, if the intention is to run failure prediction and fault classification, design teams need to create different models that serve these different purposes.
Delivering value with digital twins
The flexibility and various potential benefits of digital twins make them a top priority for companies transitioning to Industry 4.0. Having an up-to-date representation of real operating assets lets engineering and design teams unlock insights in data to optimise, improve efficiencies, automate and evaluate future performance — all delivering cost savings and shorter development timelines.
The sad reality of modern digitalisation initiatives is that by enabling greater connectivity we...
When evaluating the application of AI to industrial control systems it is important to maintain a...
The differences between condition monitoring and predictive maintenance.