Long before there were digital twins or the internet of things, Ansys was making simulation tools to help engineering teams design better products, model the real world, and expand the boundaries of science research.

VentureBeat caught up with Ansys CTO Prith Banerjee, who elaborated on why interest in digital twins is taking off, how modeling and simulation are undergoing key developments, and how AI and traditional simulation approaches are starting to complement one another. His view is that of a foundational player surveying a robust set of new applications.

This interview has been edited for clarity and brevity.

VentureBeat: What do executive managers need to know about modeling and simulation today? They both allow us to peer deeper into things, but how do these underlying technologies serve in various contexts to speed up the ability to explore different designs, trade-offs, and business hypotheses?

Prith Banerjee: Simulation and modeling help companies around the world develop the products that consumers rely on every day — from mobile devices to cars to airplanes and frankly everything in between. Companies use simulation software to design their products in the digital domain — on the computer — without the need for expensive and time-consuming physical prototyping.

The best way to understand the advantages of simulation is by looking at an example: One blue chip customer is leveraging simulation technology to kickstart digital transformation initiatives that will benefit customers by lowering development costs, cutting down the time it takes to bring products to market. A more specific example would be a valve in an aircraft engine that regulates pressure in a pipe, or a duct that needs to be modeled in many ways.

Through digital modeling, engineers can vary the pressure and temperature of the valve to gauge its strength and discover failure points more quickly. As a result, engineers no longer need to build and test several different configurations. In the past, engineers would build multiple prototypes in hardware, resulting in long times and cost. Now they can build the entire virtual prototype through software simulation and create an optimal design by exploring thousands of designs.

VentureBeat: How would you define a digital twin, and why do you think people are starting to talk about them more as a segment?

Banerjee: Think of a digital twin as a connected, virtual replica of an in-service physical entity, such as an asset, a plant, or a process. Sensors mounted on the entity gather and relay data to a simulated model (the digital twin) to mirror the real-world experience of that product. Digital twins enable tracking of past behavior of the asset, provide deeper insights into the present, and, most importantly, they help predict and influence future behavior.

While digital twins as a concept are not new, the technology necessary to enable digital twins (such as IoT, data, and cloud computing) has only recently become available. So, digital twins represent a distinct new application of these technology components in the context of product operations and are used in various phases — such as design, manufacturing, and operations — and across various industries — like aerospace, automotive, manufacturing, buildings and infrastructure, and energy. Also, they typically impact a variety of business objectives. That could include services, predictive maintenance, yield, and [overall equipment effectiveness], as well as budgets. They also scale with a number of monitored assets, equipment, and facilities.

In the past, customers have built digital twins using data analytics from data gathered from sensors using an IOT platform alone. Today, we have demonstrated that the accuracy of the digital twins can be greatly enhanced by complementing the data analytics with physics-based simulation. It’s what we call hybrid digital twins.

VentureBeat: In what fundamental ways do you see modeling and simulation complementing digital twins and vice versa?

Banerjee: Simulation is used traditionally to design and validate products — reducing physical prototyping and cost, yielding faster time to market, and helping design optimal products. The connectivity needed for products to support digital twins adds significant complexity. That complexity could include support for 5G or increased concerns about electromagnetic interference.

With digital twins, simulation plays a key role during the product operation, unlocking key benefits for predictive and prescriptive maintenance. Specifically, through physics, simulation provides virtual sensors, enables “what-if” analysis, and improves prediction accuracy.

VentureBeat: AI and machine learning models are getting much press these days, but I imagine there are equally essential breakthroughs in other types of models and the trade-offs between them. What do you think are some of the more exciting advances in modeling for enterprises?

Banerjee: Artificial intelligence and machine learning (AI/ML) have been around for more than 30 years, and the field has advanced from concepts of rule-based expert systems to machine learning using supervised learning and unsupervised learning to deep learning. AI/ML technology has been applied successfully to numerous industries such as natural language understanding for intelligent agents, sentiment analysis in social media, algorithmic trading in finance, drug discovery, and recommendation engines for ecommerce.

People are often unaware of the role AI/ML plays in simulation engineering. In fact, AI/ML is applied to simulation engineering and is critical in disrupting and advancing customer productivity. Advanced simulation technology, enhanced with AI/ML, super-charges the engineering design process.

We’ve embraced AI/ML methods and tools for some time, well before the current buzz around this area. Physics-based simulation and AI/ML are complementary, and we believe a hybrid approach is extremely valuable. We are exploring the use of these methods to improve the runtimes, workflows, and robustness of our solvers.

On a technical level, we are using deep neural networks inside the Ansys RedHawk-SC product family to speed up Monte Carlo simulations by up to 100x to better understand the voltage impact on timing. In the area of digital twins, we are using Bayesian techniques to calibrate flow network models that then provide highly accurate virtual sensor results. Early development shows flow rate correlation at multiple test points within 2%.

Another great example where machine learning is meaningfully impacting customer design comes from autonomous driving simulations. An automotive customer in Europe leveraged Ansys OptiSLang machine learning techniques for a solution to the so-called “jam-end” traffic problem, where a vehicle in front changes lanes suddenly, [impacting] traffic. According to the customer, they were able to find a solution to this 1,000 times faster than when using their previous Monte Carlo methods.

VentureBeat: So, Ansys has been in the modeling and simulation business for quite a while. How would you characterize some of the significant advances in the industry over this period, and how is the pace of innovation changing with faster computers, faster DevOps processes in software and in engineering, and improvements in data infrastructure?

Banerjee: Over time, model sizes have grown drastically. Fifty years ago, simulation was used to analyze tiny portions of larger components, yet it lacked the detail and fidelity we rely on today. At that time, those models were comprised of dozens –at most hundreds — of simulation “cells.” Today, simulation is solving massive models that are comprised of millions (and sometimes even billions) of cells.

Simulation is now deployed to model entire products, such as electric batteries, automobiles, engines, and airplanes. As a result, simulation is at the forefront of advancing electrification, aerospace, and key sustainability initiatives aimed at solving the world’s biggest problems.

The core concepts of simulation were known a decade ago; however, customers were forced to run their simulations using coarse meshing to approximate their simulations to get the results back overnight. Today, with advances in high-performance computing, it is possible to accomplish incredibly accurate simulation of the physics in a very short amount of time. Furthermore, by using AI/ML we are exploring another factor of ten to one hundred times the speed and accuracy that was previously possible, all enabled by HPC on the cloud.

VentureBeat: What do you think are some of the more significant breakthroughs in workflows, particularly as you cross multiple disciplines like mechanical, electrical, thermal, and cost analysis for designing new products?

Banerjee: The world around us is governed by the laws of physics, and we solve these physics equations using numerical methods such as finite element or finite volume methods. In the past, our customers used simulation to model only a single physics — such as structures or fluids or electromagnetics — at a given time since the computational capabilities were limited. But the world around us is not limited to single physics interactions. Rather, it has multiphysics interactions.

Our solvers now support multiphysics interactions quickly and accurately. Ansys Workbench, which allows cross-physics simulation tools to integrate seamlessly, was a key breakthrough in this market. Workbench opened new simulation capabilities that, prior to its inception, would have been nearly impossible. Our LS-DYNA tool supports multiphysics interactions in the tightest manner at each time step. Beyond Workbench, today the market is continuing to expand into areas like model-based systems engineering, as well as broader systems workflows like cloud.

Finally, with the use of AI/ML, we are entering a world of generative design, exploring 10,000 different designs to specification, and rapidly simulating all of them to give the best option to the designer. A very exciting future indeed!