Achieving Real-Time Predictive Maintenance for Wind Farms with Hybrid Twin
Staff posted on November 14, 2019 |
Data and physics-based models in a Hybrid Twin control operational costs and improve performance.

ESI has sponsored this post.

To increase competitiveness, manufacturing companies are not only reducing product development costs, they are also investigating the impact of product design onto lifetime operational costs – an important choice criterion for their customers who need to decrease the total cost of purchasing and operating products over their entire useful life. This is particularly true for expensive, complex products such as wind turbines, oil and gas infrastructures, aircraft engines, rail equipment and manufacturing lines, which may cause skyrocketing maintenance costs.

Sadly, understanding the complex interactions between mechanical, electronic, fluid components and the embedded software that controls them is no easy task. Operating companies often find it difficult to predict the real-time real-life behavior of such advanced systems. Luckily, if the emergence of IoT and sensor technology has brought new challenges, it’s also bringing solutions to tackle them through continuous remote monitoring and control.

Capital expenditure vs cumulated operating expenditure over product lifetime. Example of wind turbine.
Capital expenditure vs cumulated operating expenditure over product lifetime. Example of wind turbine.

The WindTwin Project: Decreasing Maintenance Costs for Wind Farms

Currently, there are about 216,000 wind turbines operating around the world, with 38,000 incidents occurring per year. While preventative maintenance costs 25 percent less than reactive maintenance, predictive maintenance costs 47 percent less. Moreover, when it comes to delivering product performance, and increasing wind turbine effectiveness, it is now possible to use data analytics, processing and visualization.

Agility3, Brunel University London, Dashboard, ESI and TWI formed a consortium to explore possible solutions for improvement. With funding from the British government via its Innovate UK program, the consortium launched a 30-months co-funded R&D project named “WindTwin” back in July 2017.  The project aims to deliver a cloud-based platform that combines data-analytics and machine learning techniques (as applied on sensor-data collected from wind turbine behavior and operational conditions), with physics-based modeling to create a Hybrid Twin that can be used for preventative and predictive maintenance of wind farms.

The Hybrid Twin of a wind turbine combines a physics-based virtual prototype of each system (or sub-system), using simulation data, with actual sensor data provided by the real wind turbine whilst operating in real-time. The capability to gain physics-based predictions in real-time is down to ESI’s ability to leverage Reduced Order Modeling methodologies. With this hybrid model in hand, operators will be able to predict failure (fault modeling) and anticipate maintenance needs ─ ultimately optimizing costs associated with maintenance and downtime.

The WindTwin platform will use the Hybrid Twin to provide continuous preventative and predictive maintenance information, condition monitoring, and for understanding power setting operation scenarios.

The WindTwin platform will help energy companies visualize operational faults in real-time.( Image courtesy of Innovate UK.)
The WindTwin platform will help energy companies visualize operational faults in real-time. (Image courtesy of Innovate UK.)

According to Fadi Ben Achour, Director of Hybrid Twin Services Development at ESI, the use of the WindTwin platform will reduce maintenance and monitoring costs by 30 percent.

While the WindTwin application applies here to renewable energy, the Hybrid Twin approach is relevant to a whole range of industries, including aerospace, marine, robotics, heavy machinery and rail.

The Hybrid Twin: Aiming for Greater Predictivity Than a Digital Twin

A digital twin combines a real-life hardware system and the data it collects. It can receive information from the hardware version, enabling monitoring of system health and status, as well as allowing preventive maintenance, performance optimization and what-if scenarios based on experience drawn from past system behavior.

Effectively, the real-life version is an Internet of Things (IoT)-enabled device that sends all manner of telemetry and information to the digital version. Information such as position data, temperature, and vibration or fluid flow can be measured by sensors on the hardware, and then be represented in the digital version. Many of these digital twins run complicated calculations on large datasets (including 3D data) and don’t necessarily run in real time due to the time lag.

But here are the drawbacks: digital twins suffer from the fact that the use of sensor data alone to train machine learning software is limited to data from the past. This makes it difficult to detect new anomalies and data patterns that haven’t occurred in the system before. As a result, the faults may be misdiagnosed, or new aberrations missed by the detection process altogether. In turn, this may result in system failure, unplanned down time and increased operational costs.

What’s Missing? Physics.

Virtual Prototyping pioneer and specialist ESI adds to the purely data-driven model (Digital Twin) a highly predictive physics-based parametric model to form a Hybrid Twin.

Thanks to this, operating companies can get a global system vision (system modeling) in which each sub-system is described using the parametric Finite-Element (FE). A Hybrid Twin helps predict system behavior in situations for which no empirical data is available: ageing product or exceptional/accidental climate conditions, for example. With so much refined information available, including FE, the next step was to provide information in real-time.

ESI has tackled this issue. Thanks to recent advances in Model Order Reduction techniques, ESI runs calculations on a physics-based virtual prototype accounting for nominal product design, in real-time.

The Hybrid Twin: Combining the Best of Both Worlds

Jean-Louis Duval, Innovation Director at ESI Group, explained: “On one hand, the Digital Twin methodology builds models that are entirely based on data, mimicking the past behavior of real, hardware systems. On the other hand, OEMs develop physically realistic digital 3D models in the early design phase, but these are never used in product operations because they don’t provide information in real time.”

He added, “ESI’s Hybrid Twin concept uses an accurate Virtual Prototype under the form of a real time parametric model that we then enrich with real-life data, used to train Machine Learning for superior accuracy predictions. This way, the Hybrid Twin follows system evolution in real-time and delivers the ability to precisely predict product behavior. Our approach leverages models developed throughout the Product Lifecycle Management to help operating companies enter a wider strategy of Product Performance Lifecycle.”

To learn more about Hybrid Twin, visit

Download the paper “Virtual, Digital and Hybrid Twins: A New Paradigm in Data-Based Engineering and Engineered Data” as published in Computational Methods in Engineering (26 November 2018).

Recommended For You