Simply put, a Digital Twin technology is a digital replica of a physical asset, projecting physical objects into the digital world. There have been numerous Digital Twin concepts since the idea was introduced in the early 2000s.
Given the huge amount of information produced by the Industrial Internet of Things (IIoT), engineers often faced with an overwhelming amount of data that is difficult to interpret. Digital Twins have emerged as a key element needed to unlock the power of IIoT by enabling easier an interpretation and empowering data-driven decision making.
Digital Twins in IIoT allow engineers to monitor real-time analytics and test predictive maintenance to optimize asset performance, and also provide data that can be used to improve the physical product design throughout the product’s life cycle.
A standard approach for today’s Digital Twin is to use sensor data to calibrate statistical models or machine learning-based models that attempt to represent the behavior of the asset. The advantage of these types of statistical models is that they are quick to evaluate and can be used in real-time. However, the disadvantage is that they aren’t physics-based: they model assets empirically, which doesn’t account for actual physics, so when working conditions change, the applicability of the model becomes quite uncertain.
Finite element analysis (FEA) provides an alternative route to Digital Twins, via first principles of physics. An FEA physics-based model enables us to understand complex processes and predict future events. However, FEA is simply not fast enough to fulfill the needs of a High-fidelity Digital Twins in most operational contexts. Conventional FEA can only deal with small-scale models, and cannot handle complex defect regions affected on the global model. Therefore, FEA simply lacks the speed needed to make predictions and real-time decisions representing a huge barrier to overcome in order to reach the “perfect” Digital Twin.
Akselos resolves this lack of speed via Reduce Basis FEA (RB-FEA) by being 1000x faster than traditional FEA, hence enabling detailed physics-based Digital Twins without sacrificing accuracy. This technology has extensive research in academia conducted over the past 15 years by research groups including MIT and École polytechnique fédérale de Lausanne (EPFL). RB-FEA enables a physics-based Digital Twin.
Akselos’s RB-FEA Digital Twins can also complement well the statistical and machine learning-based concepts that are often used, as described in the two points below:
The Digital Twin can be used to generate input data for training a machine learning model, because we often don’t have enough “real world” data to train properly. This is especially true of failure modes: Systems fail rarely so we don’t have much real-world failure data to train on, hence it is helpful to generate artificial failure data from a Digital Twin and use that to train a machine learning model.
Interpretation. A machine learning model can detect anomalies, but to understand what caused the anomaly, and to prescribe the appropriate remedial action, a physics-based model is necessary.
Akselos’s technology enables sensor-calibrated physics-based models that operate in real-time. This goes beyond the conventional concepts of Digital Twins, and as such Akselos Digital Twins that are coupled with sensor data are given a distinct name: the Digital Guardian.
Thomas Leurent, CEO of Akselos, says: “We use the first principle of physics to understand everything that structure is going through throughout its lifecycle. As we can compute this information extremely quickly, we have a window into the future of assets and can accurately predict how long the asset has before any risk of failure. We call it a Digital Guardian.”
The key point for this acceleration is a divide-and-conquer approach in which data is pre-computed for components and then re-used “in a flash” whenever a new analytic is performed. The components are engaged with parameters (geometry, density, stiffness), corresponding with working conditions given by sensor data, that the Digital Guardian can be quickly and easily modified and re-analyzed in real-time.
Thach Luu, Senior Engineer at Akselos, comments: “Making sure that we save time while being 100% accurate is every engineer’s dream. At Akselos, we have done exactly that, with the Digital Guardian. We have engineered a program which saves time and ensures 100% accuracy.”
In the market today, Akselos provides predictive Digital Guardians for critical oil and gas assets such as FPSOs, semi-submersibles, onshore machinery and refineries to deliver real-time, condition-based monitoring and predictive analytics. RB-FEA, when coupled with sensors, machine learning and data analytics, creates a sustainable operating model for any kind of assets during its life cycle.
By David Knezevic, Phd, CTO
Have you also read:
|How Can Digital Transformation Power the Move to Zero Unplanned Downtime?|