About a decade ago, MIT researchers discovered a technique that speeds physics modeling by 1000X. They spun this out into a new company, called Akselos, which has been helping enterprises to weave the tech into various kinds of digital twins used to improve shipping, refining, and wind power generation.

A digital twin is a virtual representation of an object or system that spans its lifecycle, is updated from real-time data, and uses simulation, machine learning, and reasoning to help decision-making. Connected sensors on the physical asset collect data that can be mapped onto the virtual model.

The specific innovation improves the performance of finite element analysis (FEA) algorithms which underpin most types of physics simulations. Akselos experience over the last decade can help executives explore the implications of the million-fold improvements in physics simulation that Nvidia is now demonstrating thanks to improvement in hardware, scalability, and new algorithms.

VentureBeat caught up with Akselos CEO Thomas Leurent to explain what these broader improvements might mean for the industry as a whole. At a high level, faster simulation makes it easier to compare design tradeoffs leading to more efficient products, reduced cost, enhanced performance, and better AI algorithms. Practical benefits have included trimming one-third the weight of wind towers and improving the safety of oil vessels.

The role of simulation in digital transformation

Digital twins are more like a design pattern than a tech. Enterprises assemble the various pieces into a solution, just like with building a data pipeline. Various PLM, construction software, and industry-specific vendors are building out portfolios to support a wider range of digital twins capabilities including physical simulation. A faster simulation engine enables companies to explore new ways of infusing simulation across the ideation, design, procurement, phases it design better products, and drive digital transformation.

Akselos is a best-of-breed simulation platform designed to improve finite element analysis, a crucial component of many types of physical stimulation. Akselos figured out how to speed the core algorithms about 1000-times about a decade ago. All of the other PLC and CAD vendors are exploring ways to do something similar.

But how exactly does a 1000-fold speedup in simulation translate into business value, since simulation is but one part of a larger business and technical process? Other companies are likely to take advantage of Akselos’ experiences as they build out their simulation infrastructure using some combination of faster hardware, better algorithms, or both.  The GPUs are already 1000-time faster than they were when this research started, and when combined with even marginal algorithm improvements companies are going to look for ways to creatively “waste” simulation cycles to see gains in other ways.

Akselos customers have discovered several ways to translate faster simulations into business value. For instance, Shell oil discovered a faster design process for a specialized multi-billion-dollar oil tanker, which reduced the number of weak points at the same time. Other customers reduced the material in a wind turbine by 30%.

Other companies are likely to see similar kinds of gains as they rethink the way that faster simulation can be applied to their engineering and deployment handoffs for other physical things such as factories, cars, medical devices, and more.

VentureBeat: What’s your overall take on some of the ways that improvements in modeling and simulation techniques could improve the use of digital twins?

Thomas Leurent: Digital twins for industrial assets can only benefit from using the mechanical engineering simulation tools that were used to design them in the first place — and those are all based on finite element analysis (FEA). The most stringent standards for operations also rely on FEA in order to operate critical assets such as refineries, ships, oil rigs, etc. But FEA is too slow to be used for digital twins in the operational phase. Therefore, a once-in-a-generation upgrade was needed to boost the core algorithms, to enable FEA to support near real-time, parametric, and connectivity-enabled use cases.

VentureBeat: What is the big deal with reduced basis finite element analysis – what is it so much faster than traditional modeling techniques?

Leurent: FEA is actually a very old and inefficient algorithm. It uses meshes (e.g., millions of triangles or tetrahedra) to define the geometry of a part. That’s fine. The problem is that FEA assigns degrees of freedom to each node in the mesh, and that’s actually complete overkill. FEA ends up solving problems in spaces with millions of dimensions, which is very expensive and cannot be done in real-time.

RB-FEA, Akselos’ pioneering technology, understands that and it looks for what Prof A.T. Patera at MIT calls ‘the manifold beneath’. That’s a subspace, much smaller than the original FEA space, and still way big enough to guarantee that the problem behaves in that subspace.

We call that the RB space, for a reduced basis (even that RB subspace is overkill, but it’s 1,000x less overkill than the original FEA space). We solve the problem in the RB subspace, which is 1,000x more efficient, and then we have all the maths to project back into the FEA space that engineers are used to and that standards recognize. To engineers that’s really transparent — you just get RB-FEA computations running at lightning speed when they used to be slow with FEA. In practice, all of this means that FEA is suitable to run simulations at the mechanical-part level, but it hits a wall beyond that. RB-FEA can run full accuracy simulations at the system level and down to the mechanical part level, without the need for sub-models. That’s a vastly improved workflow.

VentureBeat: Where are simulation providers seeing the biggest new uptake in 2021 of simulation technology for digital twins, specifically in what industries and what types of products, and why?

Leurent: The two industries we see generating the strongest pull include offshore wind and oil and gas. There is enormous growth in offshore wind with over 95% of the capacity yet to be built to meet IEA 2050 net-zero targets. There is significant demand for technology that can de-risk both the design and operations of offshore wind structures. Powerful engineering simulation using digital twins allows developers and operators to analyze thousands of ‘what-if’ scenarios in a safe environment.

In design, we have shown that we can enable up to 30% capex savings on the foundation through advanced optimization with our partner Lamprell, and there is more potential. In operations, we are the only technology provider that is able to analyze the structural health down to the square cm level. This operational digital twin is an absolute breakthrough for operators, as it provides actionable intelligence on how often they should be inspecting which parts of the structure.

The energy transition is making large oil and gas major re-evaluate major investment decisions and is driving a push towards finding ways to get more out of existing assets. That requires staking structural digital twins in an operational environment. The highly detailed models that advanced engineering simulation brings, enable a safe and efficient way to understand asset/equipment behavior and to extend its life.

VentureBeat: What are some of the kinds of use cases where you have seen significant benefits compared to traditional modeling and simulation approaches?

Leurent: We’ve compressed what used to be a six-month workflow for analyzing Shell’s floating production storage and offloading (FPSO) tanker boats, into less than 48h, while increasing the accuracy by 10x.

Other examples of use cases include the self-assessment of structural damage in flight by a drone or aircraft. And then of course offshore wind. This technology will help reduce the cost of offshore wind tremendously. Particularly floating offshore wind, which constitutes one of the largest sources of renewable energy on Earth, once unlocked.

VentureBeat: Could you walk us through how these kinds of benefits show up in practice – for example, how does a 1000X modeling performance advance translate into practical benefits, like reducing the amount of material in a wind turbine platform and its overall cost?

Leurent: RB-FEA has resulted in some of the very largest (and most complex) assets on the planet, like Shell’s Bonga floating production, storage, and offloading vessel, having a digital twin that is based on the physics (accounting for variables like hull fatigue, tank loading, waves) and compatible with standards. This earned an award for the best paper award at the Offshore Technology Conference 2021. And Akselos’ product line supports the protection of $7bn (per year) of oil equivalent production.

A digital twin with RB-FEA 30% reduction in inspection cost on an FPSO, but more importantly, look in the right place on a huge asset and detect defects early in order to avoid major problems. On the Bonga FPSO the benefit of increased accuracy has led to 15,000 top-tier fatigue locations being reduced to 230 true fatigue hotspots in the most critical locations. That is of enormous value to the operator, as they now have actionable information to drive inspection and maintenance activities where it matters most.

The benefits in offshore wind have equal, if not more, potential. For example, on the design side, we have worked with Lamprell to reduce the amount of steel in offshore wind foundations by up to 30%. This not only has direct benefits through lower material cost, but there are also very significant knock-on effects when you consider the amount of welding needed to put the foundation together as well as transportation.

When an optimized design is brought to life within operations, and crucially for wind farms, the impact is a 1000x speed up. It means an operator can make informed decisions on when to execute maintenance, and how to adjust the operating window of the turbine to avoid foundation failure if the next maintenance opportunity is some time away.

The benefits are further compounded for floating offshore wind, where the foundation and turbine have more dynamic loading. Those types of gains will be critical to lowering the Levelized Cost of Energy (LCOE), the driving scale in the floating wind. For the world to meet the IEA roadmap, those kinds of gains are an absolute necessity.

VentureBeat: How do you expect the use and capabilities of better simulation techniques like RB-FEA and related approaches to evolve in the near future, particularly as it relates to improving digital twin-related workflows?

Leurent: Understanding in real time the structural integrity of an asset is a game-changer for:

• Optimal operations
• Life extension of assets
• In-operation design (designing the next generation of assets based on the data generated from the digital twin)

Today, Akselos digital twins are deployed on assets worth billions of dollars, globally. This crosses complex (and in most cases aging), legacy oil and gas assets to cutting-edge demonstrator prototypes in the floating wind.

We’re working on making the software ever more real-time, in some cases, our physics-based digital twins interpret new data every second. That speed also enables combining AI/ML with physics-based simulations, a game-changer with vast potential. That’s what won us the AIAA best paper award 2020 for multi-disciplinary design optimization. Here RB-FEA gives a much richer, cheaper, and more accurate dataset.

The team is also working on capturing more and more physics (multi-physics and nonlinear for example). And we’re working on very powerful features of RB-FEA for optimal design, including the possibility to re-engineer an entire wind turbine system based on materials upgrades, or new design ideas within weeks.

VentureBeat: What are your main takeaways for other companies that may be exploring ways to take advantage of simulation improvements thanks to industry trends in general? 

Leurent: Probably the single most important thing is to push the imagination of what is possible. In an increasingly sensorized and robotized world, simulation technology is becoming an increasingly powerful tool to generate competitive advantage. For example, we could start to run and optimize windfarms on a turbine-by-turbine basis. Data from inspection drones and sensors on the turbines could help make health assessments of each turbine and enable operators to make informed decisions on how hard they should run each turbine depending on the power price (no point to run a turbine at high speed if that costs more ‘life consumption’ than the revenue it generates).

In downstream oil and gas, we are doing near real-time analysis to help our customers shave time off the critical path and increase uptime, and without simulation technology, this wouldn’t be possible.

Ultimately, if you are an asset owner I think it will be key to consider how different data sources and tools can be combined with simulation technology to drive better business outcomes. This hasn’t been on their mind because simulation power was not powerful enough for use in near real-time operational settings, but that has now changed dramatically.