digital twins

Predictive maintenance using a digital twin

Predictive maintenance helps you determine the remaining useful life of assets. Creating digital twins helps you create up to date representation of your assets.

Every day we rely on a wide range of machines, but every machine eventually breaks down unless it’s being maintained. Predictive maintenance lets you estimate when machine failure will occur. This way, you can plan maintenance in advance, better manage inventory, eliminate unplanned downtime, and maximize equipment lifetime.

One big challenge with predominantly used preventive maintenance is determining when to do maintenance. By scheduling maintenance very early, you’re wasting machine life that is still usable, and this adds to your costs. The question is – Can you predict when machine failure will occur ? The good news is that predictive maintenance lets you estimate time to failure. It also pinpoints problems in your complex machinery and helps you identify what parts need to be fixed.

Assume we are at a well site where we operate multiple pumps to extract oil and gas from the ground. Here, let’s take our asset as the pump. However, it could be a valve on a pump, or the well site with multiple pumps too.

We want to prevent failures of various parts of valves, seals, plungers etc. by predicting them in advance; identify faults that develop in this system and get insights into what parts may need repair or replacement.

Digital Twins come to use in these scenarios.

Digital Twin

A digital twin is an up-to-date representation, a model, of an actual physical asset in operation.

A digital twin can be a model of:

  • A component
  • A system of components
  • A system of systems

Examples include pumps, engines, power plants, manufacturing lines, and a fleet of vehicles.

In Industry 4.0 applications, models can determine remaining useful life (RUL) to inform operations on the most opportune time to service or replace equipment. Digital twins can also be used for other purposes such as virtual commissioning or to influence next-generation designs.

Digital twin : Modeling methods

A digital twin model will include the required components, behaviors, and dynamics of the  asset. Modeling methods generally can be grouped into two types: data-driven methods and first principles or physics-based methods.

Data-Driven Modeling

Data-driven modeling techniques are especially useful when you do not have sufficient information about your system. In this case, you can ensure model accuracy by choosing a modeling technique such as machine learning, deep learning, neural networks, and system identification that is right for your experimental or historical data.

For e.g., if you are estimating RUL, then you can use a data-driven model that uses historical data to predict the time until failure.

If you have complete histories from similar machines, then you can use similarity models. If you have data only from time of failure, then you can use survival models. If failure data is not available but you know of a safety threshold, you can use degradation models to estimate RUL.

Physics-based Modeling

Physics-based modeling involves designing the system from first principles. Models can include mechanical, hydraulic, and electrical components. Now let’s say you want to simulate future scenarios and monitor how the fleet will behave under those scenarios. Then you can use a physics-based model.

Consider a physical model of a pump that is created by connecting mechanical and hydraulic components. This model is fed with live data from the pump, and its parameters are estimated and tuned with this incoming data to keep the model up to date. Using this model, you can inject different types of faults and simulate the pump’s behavior under different fault conditions.

How many models should you create?

For every individual asset, you need to create a unique digital twin. The total number of unique twins you need will depend on your application. If you are modeling a system of systems, you may or may not need a twin for each system of components depending on your required level of precision.

In summary

  • Predictive maintenance helps you determine the remaining useful life of assets
  • Creating digital twins helps you create up to date representation of your assets
  • Data-driven modeling and physics-based modeling helps you model the digital twin
  • For every individual asset, you need to create a unique digital twin


R Vijayalayan  manages the automotive industry and control design vertical application engineering teams at MathWorks India.

Leave a Reply

Your email address will not be published. Required fields are marked *