Is your organization using digital twins today? The answer depends on how you define “digital twin.” At the most basic level, it is a virtual representation of a physical part, product, plant, or process – which can be a relatively low hurdle to clear for organizations with sophisticated digital capabilities, making it easier to answer “yes.” After all, at any given moment your organization may have many models in place that fit this loose definition.

In practice, though, what passes for a digital twin in many industrial organizations doesn’t come close to approaching the full power and potential of the idea. This gap between promise and reality is especially pronounced when viewed through the lens of a specific capability or organizational requirement such as predictive maintenance. In predictive maintenance, the desired outcome is beyond clear: Whether managing fleets of vehicles more effectively or keeping factories running at full capacity with machines that require minimal downtime for repairs, predictive maintenance strategies promise to identify and help address maintenance issues in real time before they grow into even bigger problems.  Just as important, it helps maintenance leaders more accurately anticipate the resources required for future maintenance of physical assets. How likely is it that a physical asset will fail before its next scheduled maintenance? What can be done today to mitigate that risk? Digital twins give leaders the tools they need to answer these types of questions today, which has a direct impact on the cost and time required for maintenance today and in the future.

In this context, a digital twin could serve as the primary engine of predictive maintenance, drawing from a host of constantly updated sensor-reported data (as well as data from other sources) to create accurate, real-time models of real physical assets. By tracking changes in these models in real time using real-world data, digital twins introduce the ability to dramatically improve immediate maintenance capabilities in the moment, as well as to anticipate future maintenance needs more accurately to inform resource and budget allocations. Yes, they can set maintenance activities in motion at the precise moment a potential problem is identified – a powerful capability on its own. But along the way, they can also identify the root causes driving maintenance needs that are likely to emerge in the future, using models to test out future scenarios. With this broad range of capabilities, digital twins can have a measurable impact on both planned and unplanned maintenance – leading to significant savings. It’s no wonder so many leaders are so keen to have these capabilities in place.

On the path to maturity: From digital shadows to digital twins

If your models don’t enable that type of insight and response today, they may be better defined as “digital shadows” rather than digital twins. Digital shadows tend to use less data overall, much of it backward-looking, and as a result tend to describe what just happened or what is happening right now – but not pointing the way to what is likely to happen in the future. While this can be useful in some ways, it’s not nearly enough to enable real predictive maintenance.

Running digital shadows rather than digital twins isn’t a failure.  It’s a key step on the path to maturity, and an opportunity for organizations to augment their existing strengths and capabilities to develop fully operational digital twins.  It helps to start with a clear understanding of what a mature, full-featured digital twin looks like for your industrial organization.

Among many digital twin maturity models, the one developed by Atkins Realis, the international engineering, procurement and construction firm, is notable for its clarity. As shown in the graphic [at right, below, wherever it appears in the blog], this maturity model suggests that while the most basic digital twins simply capture and reflect the current reality, those that are most sophisticated could conceivably operate autonomously.

For most maintenance organizations, the fourth level of the model is a more attainable goal today. At this level, a mature digital twin is its ability to enable a constant bidirectional flow of data between physical and virtual models, whether for a single asset, a fleet, or other large portfolio of industrial assets. To understand what this looks like in practice, imagine you’re responsible for engine maintenance on a fleet of aircraft. Starting with engine performance data, you create models that show the full range of types of engine failures and their causes. Along the way, the models help identify that engines operating in dusty conditions are at frequent risk of failure – all based on historical data. You develop a finely tuned understanding of time to failure based on the duration of exposure to these conditions. The models are sophisticated, trustworthy, accurate – and ready to be augmented with live data from engines in use in aircraft across the fleet.

You outfit the aircraft engines with sensors set to detect a wide range of conditions, including exposure to particles, and the data is constantly, automatically fed into the models. This is the moment at which a digital shadow becomes a digital twin, pumping bidirectional data between aircraft engines and their digital twins to create active risk profiles of individual engines. With these insights, a mature digital twin can flag potential risks before they emerge, notifying human operators who can decide whether and how to intervene as part of a larger predictive maintenance strategy.

Building the right foundation of data

In predictive maintenance, mature digital twins rely on much more than just sensor-generated data. It’s often said that there’s virtually no limit to the data that can be measured, managed, and analyzed today – but because there are limits to the amount of data that any given organization can manage given cost and technology constraints, it is important to distinguish between peripheral, “nice-to-have” data and data that is essential to the functioning of the digital twin. These are the five core types of data that every mature digital twin should have to enable predictive maintenance.

• Digital model: Geometry data of a physical object.
• Static parametric data that represents specific asset or component characteristics.
• IoT data: Real-time, high-fidelity sensor data generated by internet-connected machines. This is sometimes referred to as telemetry data.
• Service data: Inspection, repair, and service records for the asset over its lifecycle.
• Simulation data: AI and/or physics-based that can use real-time readings and synthetic data to produce results under different performance scenarios.
• Physical entity control: Resulting data that supports human-assisted, augmented, or autonomous actions.

Depending on the organization, its strategy, and its predictive maintenance needs, it is possible to build out these layers of data in steps rather than all at once.  For those with digital shadow capabilities, several of these data types may already be readily available.

How to get from here to there

Having established the distinguishing characteristics of a mature digital twin, the obvious next question is: How do we get from here to there as quickly as possible? In the next article in this series on digital twins, we’ll take a closer look at the practical steps required to move up the maturity curve, using a real-world example enabled and executed by SAS. Even better: It’s about motorcycles!