In the next five years, the level of enterprise digitalization is expected to double, with predictive maintenance leading the investments race. Inspired by the improvements arising from applying IoT-driven predictive maintenance solutions, 55% of enterprises are piloting predictive maintenance initiatives. However, only 4% of maintenance professionals consider digital twin technology to be a key predictive maintenance enabler.

Digital twin technology implies creating a virtual representation of a physical asset or a system, e.g., an industrial machine, a production line or even an entire factory, to model its state and simulate its performance. Digital twins are continuously learning systems, powered by machine learning algorithms, which makes them adaptive to the changes in the state and configuration of a physical twin.

In an industrial setting, digital twins are used to improve product design, monitor equipment health to identify potential degradation, simulate manufacturing operations, and more.

To understand how digital twins enable predictive maintenance, let’s consider a simplified example of predictively maintaining a centrifugal pump.

Step 1. Creating a pump’s digital twin requires:

Building its accurate 3D model; and powering the model with IoT data.

To build a 3D model, modeling experts collaborate with mechanical, electrical and process engineers to describe and virtually present physical properties of the pump and its components (e.g., an impeller type, the number of suctions, etc.). Then, the 3D model is powered with IoT data fetched from sensors attached to the pump. This data includes records about a pump’s performance, condition and environment (e.g., temperature, voltage, inlet pressure, etc.).

To improve the model’s functionality, the digital twin software is integrated with enterprise and shop floor management systems. Fetching contextual data (e.g., regulatory, financial, operational data) from, say, ERP, the digital twin could predict how a pump will function under varied external conditions.

Step 2. Putting the digital twin into action

The digital twin-based predictive maintenance software takes in real-time sensor records about the health and working conditions of a pump and analyzes it against historical data about the pump’s failure modes and their criticality, and contextual data fetched from enterprise and shop floor management systems (e.g., pump’s maintenance data).

A neural network detects abnormal patterns in the incoming sensor data and reflects the patterns in predictive models, which are then used to predict failures. This way, if a pump’s current configuration is likely to lead to a failure, the digital twin software localizes the issue, assesses its criticality, notifies technicians, and recommends a mitigating action.

Along with the prediction of failures, digital twin technology provides:

The ability to calculate maintenance-related KPIs. Combining historical data about failures, risk factors, machine configuration and operating scenarios, a digital twin can calculate maintenance-related KPIs: RUL, EoL, MTBF, and more.

The ability to forecast the behavior of machines under different circumstances.Being an accurate real-time model for an object’s condition and performance, a digital twin is used to run simulations and predict how an object will "behave" under certain factors, e.g., runtime, exposure to severe operating conditions, etc.

The ability to simulate different maintenance scenarios. Technicians use digital twins to test maintenance scenarios or particular fixes and see how they work for a piece of equipment before applying them to the physical twin. 

The challenges

Although digital twin-enabled predictive maintenance offers many benefits, its deployment may pose the following challenges:

An accurate model should precisely reflect the physical twin’s properties.A digital twin should precisely reflect all the properties of a physical twin, including mechanical (suction pressure, design temperature, etc.) and electrical (capacitance, conductivity, etc.) ones. It requires input from facility managers, process engineers, electrical engineers, equipment vendors, and other parties, which adds complexity to the deployment.

Detailed blueprints of a machine's failures are required.To predict failures, a digital twin should be fed with data about equipment failure modes. This data should be gathered for an extended period of time(say, a year) to observe a machine throughout its degradation process.

A digital twin requires remodeling with any change in equipment’s configuration or element state.Any modification affecting equipment performance requires a change to its model and underlying algorithms. Such modifications – at a machine level (replacing original parts with made-to-order ones) or at a factory level (changes to the operational policy) - are not always reflected in factory specifications, thus, cannot be precisely simulated, which escalates the risk of errors.

Although deploying a digital twin-based predictive maintenance is time-consuming and labor-intensive, the technology offers the ability to timely recognize disruptions in asset performance, forecast potential problems and simulate various maintenance scenarios. It helps enterprises eliminate machine downtime, reduce equipment maintenance costs, improve equipment reliability and extend its lifespan.

Boris Shiklo, CTO at ScienceSoft, is responsible for the company’s long-term technological vision and innovation strategies. Shiklo has a solid background in IT consulting, software development, project management and strategic planning.