Enhancing Operational Efficiency with Hybrid Digital Twins: An Interview with Vitor Lopes

Operators reuse engineering design in operations and engineers gain field insights learned from data for the next design cycle. Used across industries ranging from virtual sensors to unmodeled physics, hybrid digital twins help companies solve many challenges. In this interview with Vitor Lopes, senior product sales manager at Ansys, we will get a deeper understanding of hybrid digital twins and the typical use cases for this technology in the energy industry.

Question: Can you give us an overview of asset management in net-zero and digital transformation contexts?

Vitor Lopes: When we think about asset management, we're typically thinking about performance, asset reliability, and safety. We're thinking about how to operate a given asset in a more efficient way. In the energy industry, when we are faced with improving efficiency, we typically like to gather offset data, analyze, and make decisions. The more quality data we have, the better we can make decisions, and that minimizes risks and optimizes the chance of succeeding in what we're trying to do.

But sometimes we don't have all the data that we'd like available to us — or sometimes we do, but the risks are far greater than the quality of the available data. Safety margins are often engineered into the process from the outset, but they are made wider when there is a lack of data or too much uncertainty. This needlessly limits the performance that we can get out of a given asset.

Recently, the energy industry has faced many challenges and economic pressures. There has been a big push to improve efficiency on existing assets. Additionally, most of the processes in this industry are thermal intensive and performance gains can sometimes represent reduction in carbon emissions. That's where digital transformation is coming heavily into play, whether it's to achieve operational efficiency or even sustainability targets through improved asset management experience.

Digital transformation enables data availability and the ability to derive more insights out of the original dataset. We can do this more efficiently than ever because of recent advancements in digital technologies that most companies are adopting to improve efficiency while keeping costs low. This movement is also enabling companies to get a head start on their net-zero targets by improving the performance of existing assets while developing/replacing the next set of assets that will be even more sustainable and efficient out of the box. Digital twins are certainly at the forefront of these initiatives.

Q: How can these initiatives demystify digital twins?

VL: The concept of digital twins is that you have an asset or process in the field, you have models in a virtual world that represent critical elements of these assets, and you have ways to keep those two synchronized. Data flows from an asset running in the field to the digital models. The models then make predictions from that measured data to generate insights that you wouldn’t be able to measure on the field. Finally, operators receive this information back from the models in time to make adjustments. With that additional data available, you can make decisions more efficiently. This concept has been adopted in the asset management industry for years, even before the term “digital twin” was coined.

What makes it different than before? Advancements in technology are enabling us to synchronize the two worlds at higher frequencies and fidelities than ever before. This magnifies our ability to make better decisions. When we think of a digital twin in those terms, there is typically a balance in creating models that have the right level of accuracy and speed. That is typically a trade-off. There is also an element of flexibility. How are you going to bring intelligence to those models? Do you have sensor data? Do you have physics models? Are they 1D or 3D? Next, there is an element of adaptability. Even if the models respond very well to field testing and can give accurate predictions on day one, how would they respond 10 years from now? Is the digital twin able to follow the aging process and adapt?

Scalability is also important to consider. Even if you have found solutions to the other challenges, how can you do it in a scalable way so that it's easy to replicate to a multitude of assets? Also considered here is the overarching challenge of interoperability and security. You want components to be able to communicate with each other and be secure.

Traditionally, two different approaches have been employed to generate the models behind the digital twin: data-based and physics-based. Ansys and many of our customers understand that there are benefits and challenges to both types of approaches in isolation. To gain the most accuracy, speed, flexibility, adaptability and scalability to surpass the traditional challenges, Ansys merged both approaches into a hybrid solution, which enables an interested party to start off heavier on the approach of higher maturity and readiness, collect initial value, and evolve from that by bringing more elements of the second approach to gain even further insights and benefits.

Q: Can you tell us about Ansys' hybrid digital twins for online adaptive virtual sensors?

VL: Hybrid digital twins not only generate intelligence from both physics and data models, but they also combine their potential. If you are approaching your digital journey from a data standpoint, you may learn over time that these models tend to reach limitations around what you've historically been measuring. For scenarios where a good amount of quality and relevant historical data is available, the data-based approach can be a good starting point. That said, the moment you need to predict beyond that, it's a little bit more complicated — not just in terms of the operating ranges, but also in terms of variables for which you never collected data. Here, there are challenges to extrapolate and explain new phenomena.

On the other hand, if you are approaching your digital twin journey from a physics-based standpoint, physics models generate a lot of understanding as far as explainability goes. Using physics models to train these digital twins gives you much more control over the predicting envelope. You can make sure to have a model that is trained within the boundary conditions that you expect to use for the asset in the field. Not only that, but if you're talking about 3D physics models specifically, you can define localized measurements anywhere in your digital asset, which is very powerful. This is essentially the concept of virtual sensors.

However, most of the time the physics models are not perfect. There could be missing physics, or the assets may have aged. Bringing the measured data from the field to fine-tune these models becomes very relevant. This is the main concept of hybrid digital twins.

In this context, what Ansys offers is a platform that can take you on this journey. You can build and validate models from your own environment, whether they’re 3D or 1D. You can take it one step further and create deployable units that will connect to assets and calibrate them. You can even set up techniques to be able to adjust these models online.

Finally, there is the ability to scale this. Whether it involves a container or a web app, there are several different ways to take out a model that was built within the Ansys Digital Twin platform and bring it externally so you can place it in your environment and run it with your live sensor data. You can use the models to feed what you can measure and predict what you cannot measure.

Q: What are the typical use cases for hybrid digital twins?

VL: In our industry, we have a lot of different players. In different parts of the business, we may have operators; service companies; engineering, procurement, and construction (EPC) experts; or original equipment manufacturers (OEMs). Depending on how you are using digital twins to bring value to your company, you may have different manifestations of these use cases. For instance, manufacturers stand to build digital twins using the design information that they already have from their systems. They can probably sell operators an added service or premium on top of their existing products by offering a digital twin. Alternatively, the operators may develop in-house twins to optimize runtimes, uptimes, and yields and predict maintenance, failure rate, and future performance.

In the most common use cases for hybrid digital twins, you are typically trying to change or kick off a new process. You can use these digital twins offline as part of virtual commissioning. You can try to understand what-ifs and learn how to define set points. Then you can take it online to receive virtual sensing data. You can start using it for monitoring, then expand through automation and optimization. Finally, you can also take these virtual sensors and use them for predictive maintenance. Even if you have to service an asset on the field, being able to know service schedules ahead of time often results in more cost savings for companies, as they can avoid unplanned downtimes.

Q: What is unique about the solution?

VL: There are three main aspects. First, we can take results from 3D models and use them to train reduced-order models (ROMs) for a variety of applications and use cases. In the past, the 3D models had limited application for online asset management given the extensive runtimes. However, we're able to take these models and make them run much faster while maintaining their accuracy and fidelity. Second is the ability to use measured data and machine learning technologies to calibrate or model the residual physics, the concept of hybrid twins is that they combine the best of both approaches. Third is having the ability to containerize these models for scalable deployments. You can work outside the Ansys environment and connect those containers or platform-agnostic products to your own environment. These use representational state transfer (REST) APIs to connect to your Internet of Things (IoT) or edge devices for quick deployments.

Ansys has been around for 50 years, and our digital twin platform has been around for at least a decade. We can not only help you with the knowledge that we've acquired in different industries, but also across industries. There is plenty of expertise that can be cross utilized between fields, and our technical experts can work with you to build that first solution.

Q: Is it viable to spend effort on developing digital twins for legacy assets with limited data using physics-based models?

VL: It will depend on how little information you have. Assuming you have physics-based models and limited data, one obstacle you may have is validating these models. This may pose a challenge in trusting the results to be able to drive decision-making. But with that level of understanding, if you can quantify that level of uncertainty and decide what level of risk you are willing to take, you can trust that the initial digital twin will evolve with time.

As you collect more and more data through hybrid techniques, you can adjust and validate these physics models — or you may even realize they were good from the beginning. This quantification will reduce your engineering barriers because you can trust your digital twin.

Q: What if we don't have 3D physics models?

VL: We have many customers today that are using in-house codes,1D models, or performance curves from manufacturers. Oftentimes, this is a good first step. But what if that's all they have? Is it prohibitive to start? Not really. What’s important to understand is the level of uncertainty those models will bring. If you find you have inaccurate 1D models and your in-house code is taking too long to run, we have solutions.

You can use ROM techniques to speed up models and hybrid analytics to fine-tune them. In conjunction with what you already have and our platform, we can put you on the path to generate a meaningful twin in which you can understand the level of decisions and develop a plan to work in parallel on creating 3D models. Not every single application will require a 3D model, but even in the ones that do, you can start with 1D and work your way to 3D.

If you are interested in Ansys Digital Twin solutions, please contact us to get started with a free trial.