The concept of digital twins is a leading trend in enterprise strategy. It gets its name from the way that companies are building virtual equivalents, or twins, of physical objects. These digital copies are increasingly popular because they can be used to drive important simulations that haven’t been possible until now.
Take a wind turbine, as an example of where digital twin technology comes in handy. The turbine can be outfitted with sensors, which produce real-time data about the turbine’s performance, be it speed, energy output, or weather conditions. This data can then be used to make a digital copy of the turbine, including a 3D digital representation. Machine learning and other models can be applied to recognize patterns in this turbine — for example, whether it is working optimally. The digital copy can be used to run simulations without bothering the original turbine, and improvements can then be fed back to the original.
Simulation drives interest in digital twins
Observers see significant demand for multi-physics simulations that present a holistic view across different physical domains like electronics, structures, and heat. This is critical for areas like noise and vibration. Top simulation techniques include computational fluid dynamics (CFD), multi-body systems (MBS), or finite element analysis (FEA) technologies.
Simulation is increasingly relevant in the manufacturing industry. Simulation software is an insurance policy for manufacturers, ABI Research principal analyst Michael Larner told VentureBeat. There is an arms race in the supplier community regarding the algorithms that can be deployed, he said. This “insurance” allows them to respond to rapid changes in consumer demands and supply chain disruptions, such as the chip shortage currently hobbling the auto industry. ABI forecasts that simulation technologies for manufacturing could grow at a rate of 7.1% to $2.6 billion by 2030.
Others expect to see simulation advances used to improve various aspects of operations, particularly with the rise of the so-called “omniverse” for rendering models — referring to the use of things like VR and AR, automated data labeling, AI-powered physics, and improved supply chains.
12 ways simulation trends affect digital twins
1. Omniverse for collaboration
“The most exciting development in simulation and modeling tools over the next three to five years will be the evolution of the omniverse,” Blackshark CEO and cofounder Michael Putz said. Some more exciting improvements will involve AI-supported modeling for reconstructing buildings, live labeling, and AI frameworks. Top omniverse use cases will combine simulation and collaboration for urban planning, location scouting, architectural acceptance, logistics, UAV flight planning, and insurance.
2. Learning with less data
The Generative AI techniques used in deep fakes are also getting better at refining and optimizing the simulation models used for different digital twins. As NASA JPL chief technology and innovation officer Chris Mattmann explained to VentureBeat, “The key is balancing between the need for labeled training data and realistic environmental simulation for ground truth of the digital twin environment.” He predict there will be more adoption of synthetic data techniques to improve model accuracy and efficacy with less manual labeling.
3. Covering gaps in physics
Modeling and simulation tools are improving using AI to build physics models from live data captured from physical industrial processes. Nnaisense’s CEO and cofounder Faustino Gomez told VentureBeat that digital twins from conventional physics models are too slow to be used for complex processes involving chemistry and fluid dynamics in real time. For example, Nnaisense worked with EOS GmbH to develop a digital twin for modeling heat in additive manufacturing processes without explicit physics models. These models can predict important phenomena in real time instead of days. Top AI algorithms he sees bridging the physics gap include geometric deep learning, neural ordinary differential equations (ODE) models, and contrastive learning.
4. Inferential models simulate manufacturability
Digital twins simulations have traditionally focused on simulating product performance characteristics. Improvements in sensors embedded in the manufacturing process are enabling inferential models that can simulate manufacturability characteristics that affect quality, cost, and ease of assembly. Tempo Automation’s chief product officer Jeff Kowalski said that inferential modeling techniques automate the process of generating digital twin models through direct observation. This reduces the human effort in handcrafting the rules that go into a model. It also automatically updates models in response to changes in the environment.
5. Improving autonomous systems
Better digital twins could also improve models that guide autonomous cars, ships, forklifts, and even factories. Kalypso director of data science Jordan Reynolds told VentureBeat, “Major advancements in autonomous system performance are attributable to model predictive control (MPC), a digital twin methodology that simulates how a complex system will respond to operational inputs and changes in its environment.” These models are used to simulate dynamic system behavior and autonomously control these systems in the physical world. MPC is also used to simulate the spread of COVID-19 and determine the optimal interventions to accelerate its decline.
6. Simulation orchestration
Simulation containers promise to build on the success of application containers underpinning agile software development and deployment practices. Aveva chief technologist of XR Maurizio Galardo expects to see simulation tools move from finessing products designed to solve specific tasks to enabling a container of features that allow users to synthesize complete product designs quickly. These simulation microservices could be reused across different design, simulation, and production workflows.
7. Generative design of systems
Generative design techniques automate design suggestions from a set of starting specifications. PTC’s vice president of product management, Paul Sagar, explained that engineers have traditionally used generative design to create and optimize single parts. He expects improvements in algorithms and processing capacity to solve broader problems around simulating complete assemblies, such as seeing how a carburetor might perform using a digital twin of the complete car.
8. Engineering business products
Improvements in computational horsepower and interoperability are ushering in digital twins that combine business and technical simulation techniques. Deloitte Consulting national emerging tech research director Scott Buchholz explained, “Digital twins can be very useful for simulating things like the change from selling widgets to selling as-a-service.” For example, Bridgestone uses digital twins to optimize fleets’ cost per mile, maintenance, and tire selection. This helps business teams sell tire miles as a service and align engineering decisions around longevity and maintenance strategies to improve this new business model.
9. Supply chain collaboration
Simulation tools providers like Synopsys are finding ways to simulate that span chip design and the software that runs on them. This promises to improve collaboration for products like automobile chips that have faced significant shortages owing to the integration challenges of more modern chip designs. Synopsys verification group vice president of engineering Tom De Schutter sees big promise in developing scalable digital threads that operate across the supply chain to reflect individual components through full system products. This includes digital twins of individual hardware designs, systems on a chip, electronic subsystems, and full systems. However, this will also require new infrastructure to capture, share, and track the fine-grained data powering these hybrid digital twins.
10. Smaller models
AI can also be honed to build smaller models that require less data and compute power than traditional approaches called reduced-order models, said Altair’s chief technology officer, Brett Chouinard. He expects this to increasingly support the provisioning of sophisticated digital twin models on remote devices like edge gateways and equipment. These smaller digital twins will increasingly add value to new products and services. Chouinard said, “While this is already happening, it will only get more center-stage with newer applications built around it and resulting in increased sophistication and demand for more capacities at the edge.”
11. Multi-domain digital twins
Simulation integration techniques are also opening opportunities for multi-domain simulations. These build on multi-physics and integration techniques to support working across domains such as security and physical infrastructures like power grids and gas pipelines, Scalable Network Technologies CEO and founder Dr. Rajive Bagrodia said. For example, in the power grid, attacks that delay control of a circuit breaker or falsify sensor load reporting may cause a series of cascading effects with potentially catastrophic outcomes to a regional power grid. Multi-domain digital twins that couple physical system simulators with network emulators could improve resilience, detection, and response to these kinds of scenarios.
12. Democratization of simulation
The democratization of simulation could open new planning and development opportunities for less technical business users. Today the simulation market principally addresses industrial designers and engineers in R&D departments. More accessible tools will lower the barriers to adoption for business users, purchasing departments, and subject matter experts. “Design decisions will be much smarter because, instead of new products being selected purely based on aesthetics or performance, they will be chosen based on a full range of factors,” Roger Assaker, president of Hexagon Manufacturing Intelligence’s MSC Software division, explained.