Digital twins are defying the limits of space & cheating the inertia of time
- A digital twin is a virtual model of a real-life system, coupled with that real-life system
- Digital twins can cut time-to-market, reduce operational costs and risks, and improve return on investment
- Building and using digital twins takes you on a journey of IT modernization and business transformation - Digital twins force digital maturity
- To mirror more of the complex real world, digital twins have to be very smart, very powerful and very connected
- The best-established uses for digital twins are the four Ps – prediction, prototyping, prevention and practice
- Digital twins change value assumption and asset consumption
- Improving the dimensions and detail of a digital twin helps a company understand the causes and effects of business change
- Tighter coupling of digital twins and the enterprise brings digital twins into business and IT operations; they can now get inside the tightest tactical decision windows
- Looser coupling between digital twin and enterprise can reduce digital complexity
- Digital twins are defying the limits of space – they are going beyond simulating large assets to replicating smart cities. Digital twins are cheating the inertia of time – they can now predict a machine’s future performance in real time, and they can train algorithms faster in a digital twin than in the real world
Digital twins accelerate digital change
Building and using digital twins takes you on a journey of modernization and transformation (Figure 1). Durable, scalable, affordable digital twins are built from fast-evolving but robust digital products and services. You must first modernize your IT foundations (e.g. with mature APIs, ML services, Big Data and ubiquitous instrumentation before you can use them. As the twin is coupled to the product, asset, process or person, it becomes an integral part of the organization’s operating model. Then as the twin evolves, it facilitates change, transforming the enterprise (e.g. virtual commissioning of industrial plant, Autonomous IT operations, sub 1 hr pick/delivery in grocery). The gradual integration or coupling of the twin into the operating model as it is built and used drives the organization through a digital transformation journey. And increasingly, twins can also disrupt the industry – when they are highly evolved and tightly coupled into an organization’s operations, they enable that organization to work at scale, and learn at scale and outrun competition (e.g. Evolving industrial products into on-demand services, accelerated algorithm training – Deepmind Go has digital twins, ‘no checkout’ stores such as Amazon Go).
Figure 1 – Building and using digital twins takes you on a journey of modernization and transformation
The parallel lives of digital twins
The simplest way to describe a digital twin is a virtual version of a physical space containing assets, processes or people - a system - that is coupled with the physical space. We replicate the characteristics of that system then test out scenarios in its digital twin.
The decades-long evolution of digital simulation has given us sophisticated digital twins. Today there are many species, harnessing machine learning, processing data streams from sensors, and building detailed models of assets, processes and people that represent real-world behaviour accurately. Leaders in the field are starting to build digital twins of a whole organization, an ecosystem or a city, applying the machine learning of the twin to suggest improvements, within days.
Today, the best-established uses for digital twins are the ‘four Ps’ – prediction, prototyping, prevention and practice – typically applied throughout the lifecycle of a manufactured asset. Let's take the example of a cutting-edge connected car. Early in the design phase, before launch (after which it becomes more expensive and complicated to fix any issues), digital twins are used to predict performance. The models are tested for aerodynamic, mechanical and structural performance. Successful designs are then prototyped virtually in digital environments that reflect both normal and extreme driving scenarios, and the chosen design improved. As the new car is prototyped in the real world, the digital twin is updated with actual performance characteristics and run through parallel virtual tests that replace some of the time-consuming physical tests, shortening the time to reach production. In production, the manufacturing machines also have digital twins that help monitor performance, predict breakdowns and schedule preventative maintenance. Production engineers also use them to identify performance opportunities and predict manufacturing performance.
When the cars are in the market and being driven, sensors give a unique view into the ways customers are using them (learning at scale) and this data can be fed back to the digital twin. Dealerships can also update the twin with any performance or maintenance issues.
The twin is now acting as a sophisticated database, allowing departments to interrogate performance data and send change requests to manufacturing and software development (software is already typically 40 per cent of a car’s value). Updates to the car’s software are done through dealerships or even over the air, overnight. Third parties such as insurers can also access the data to provide personalized services.
Throughout the lifespan of the car, the digital twin also acts as a simulator allowing technicians (often using VR) to practice diagnostic and maintenance procedures on a virtual car, helping them get comfortable with new or changed procedures. And throughout, the twin with its rich history of changes and real-world performance is also available to the design teams of other models.
The bottom line is that the digital twins of the car and the manufacturing facility cut time-to-market, reduce operational costs and risks, and so improve return on investment.
To mirror the real world, digital twins must be very smart, very powerful and very connected
In the real world, small things make up bigger things. For example, as shown in Figure 2, valves and other components are assembled to make engines; an engine integrates with a body, drivetrain and electrics to make a car; a car utilizes radar, video, ultrasonics, a computer, GPS and maps to become an autonomous vehicle; an autonomous vehicle communicates with roadside sensors and edge computing on a smart highway to become part of the transport ecosystem.
As components become subsystems then a system, their behaviour and their digital models change from simple to complicated. As that system integrates with other complicated systems, and they communicate with a wider ecosystem, the behaviours and models become complex. In parallel, the maths changes from linear to nonlinear, needing more granular models, smarter model code, more compute power, a bigger volume and variety of data, and – critically – more trustworthy data.
In short, to mirror the complex real world, digital twins have to be very smart, very powerful and very connected.
Figure 2 – In the real world, small things make up the bigger things
The long evolution of digital simulation
Digital twins are not new – there has been a long evolution to their current sophistication. In the 1970s and 80s, we modelled simpler things, like valves and engines, machines and manufacturing lines, and chemical reactions (see the lower part of Figure 3). We used proprietary finite element analysis software running on Cray and IBM supercomputers with specialized pre- and post-processing graphics running on VAXs, whose data stores quickly became full. It was expensive and slow; learning curves were in months or years. These early digital twins were mostly in R&D or product design, or academic research. They were very lightly coupled with the enterprise.
In the 1990s, as the cost of computing and storage dropped, and coding evolved, models were able to represent the behaviour of more complicated things, like cars, factories and molecules. What had been done in a specialist bureau was brought in-house; analyses could run within the longer decision timescales so we began to see annual budgeting, product roadmap refresh machine design and new material evaluation. These more sophisticated digital twins were more closely coupled into the enterprise and they started to steer decisions that drove the bottom line.
In the 2000s and 2010s, we modelled yet more complex things, often by expanding the scope to the larger environment (see the top of Figure 3). We developed models of autonomous vehicles on roads or in-flight; mainstream modelling of stocks and flows in supply chains or banking systems has evolved from spreadsheets developed by external experts to algorithms coded by internal business people.
Figure 3 – The evolution of digital simulation to digital twin
During this time, expensive compute power was transformed into inexpensive compute on demand; data became shared and traded; accurate algorithms became marketed or open-sourced. As complicated models became simplified, modularized then industrialized, models or twins could be assembled like Lego blocks to represent genuinely complex systems with unguessable emergent behaviours. These complex mechanical, business and biological systems could be approximated, refined then mastered. We can now model drug reactions in tissue, the behaviour of smart roads and highways in smart cities, and even the development of markets.
As the digital twins and their infrastructures evolved, the leaders began to forecast behaviours beyond the enterprise into the customer and supplier base. Better still, they can anticipate how operations should change if customer behaviour changes. Digital twins are now getting inside the tightest operational OODA loops. They can change delivery drivers’ routes in real-time, adjust the merchandising on shelves based on that day’s buying patterns, alter the priorities of ticket resolution in IT operations, change financial trading tactics on the same day and adjust the suspension on a leading-edge car to allow for updated road condition information. These tighter couplings between digital twins and the enterprise show digital twins finally becoming part of operations.
Digital twins can see change, learn change and guide change
Now, we can model the complex structures and systems of the physical world – molecular structures, engine structures, transportation systems, city structures, weather systems, business ecosystems – accurately, in all their complexity, scale and detail. Today’s digital twins reflect the nuances, non-linearities and counter-intuitive vagaries of these systems. We no longer need to model a simplified world and extrapolate conclusions; we can model the real world – people, cities, populations, businesses, trade, markets, economies – and show behaviours as varied as drug reactions, system hacks, shopping habits, viewing preferences, driving routes, flight profiles, manufacturing changes and supply chain responses.
These models can describe changes to the business that the business can understand (product differences that create manufacturing variability, environmental conditions that affect jet engine performance, demographics that influence drug trials). They create such resolution of their environment that they foresee the effects of small changes (Formula 1 car parts change every two weeks; Google analytics products change every day). They can predict whether a change will introduce side effects such as product faults, component failures or bad reactions to a drug.
Four more things you may not know about digital twins
- Digital twins make things add up. There is an additive value of digital twins that describes ever-larger systems. For example, the future of cars is to sense-drive-predict-optimize better than humans. So, we first model autonomous technology to see how it could change car behaviour. Then we model autonomous vehicles to see how they could drive better. Next, we model autonomous vehicles on a normal highway to see how their driving improves when they communicate. Finally, we model autonomous vehicles on a smart highway to see how driving improves when the highway communicates with the vehicles.
- Digital twins model smaller things so they can model bigger things. In order to accurately model big environments (e.g. digital ecosystems such as autonomous vehicles on a smart highway) to handle variability and change, digital twins have to be up integrated – you can’t model an ecosystem unless you understand its basic entities. Smaller digital twins must be built under the larger digital twins to get the right dimensions and behaviours. Continuously improving the dimensions and granularity of a digital twin helps a company understand the causes and effects of change.
- Digital twins force digital maturity and reduce digital complexity. As more of the real world is instrumented with sensors connected to the Internet of Things (IoT), we create applications to analyze and respond to the data they collect. This creates three sorts of complexity: first, the volume, velocity, variety and veracity of IoT data from the ‘thing’ creates complexity; second, complexity arises when every new application establishes a new channel to access the data; and finally, if either the ‘thing’ or the application changes, the other often has to change too. Digital twins reduce complexity by decoupling the ‘thing’ from the enterprise – sensors send all the data to the digital twin and business units connect to the twin to access the data. That decoupling also means changes to sensors or applications don’t necessarily mean changes to the other. Finally, the twin also processes the data to reduce its complexity before applications consume it. The twins that do this are a sophisticated integration of leading-edge digital products and services, which requires modernization of an organization’s digital skills and technology capabilities.
- Digital twins change value assumption and asset consumption. Digital twins change the overall asset lifecycle experience for buyers. For example, when a buyer newly appreciates the value created by the digital twin beyond procurement and into the manage and operate stages of the lifecycle, they often want to pay for that value differently. One example is power by the hour instead of a capital purchase of something like a jet engine – the output of the asset is consumed on-demand; another is an asset operated by third parties or vendors and paid for via transactions or subscriptions.
The future is happening now
Digital twins are defying the limits of space – they are going beyond simulating large assets to replicating smart cities. We are starting to see the crowdsourcing of digital twin data to create digital models of national infrastructure.
Digital twins are cheating the inertia of time – they can now predict a machine’s future performance in real-time, and they can train algorithms faster in the digital twin than in the real world. We are starting to see digital twins help companies gain competitive advantage by learning faster, then immediately automating that learning in their IT and business operations.
The challenge is, can you modernise your IT to evolve a twin? Do you know where to start? Can you transform your business by coupling the twin into operations? Can you disrupt your sector by putting the twin at the centre of a new business model?