During product development, companies often aim for faster, cheaper, more accurate, lower-power, and smaller designs. To accomplish these goals, designers have used step function innovations such as switching from through-hole components to surface-mount, moving from manual analysis on spreadsheets to sophisticated simulations, and conducting manual testing.
Yet, despite recent improvements, most companies still take several months to correlate test data with design, significantly delaying product launches and straining engineering resources.
Many companies have standardised and streamlined their design, test, and build processes, including implementing gate processes to ensure consistency and repeatability. However, the tools and methods remain complex. Research shows that organisations use various software tools for designing, testing, and verifying their products. Most organisations use 3 to 10 software tools for each process. This complexity can delay product introduction and cause performance and quality issues during the production and customer support processes.
But change is on the horizon. A new approach to designing and building workflows involves using digital twins. Digital twins mimic every aspect of the product as if it were real. This process eliminates the gap between theory and reality.
While the digital twin concept appears simple, it can be deceptively complex when unravelled and understood. Although it requires careful construction and deep thought, it is possible to model a single function over time.
Role of Digital Twins
A digital twin is a virtual model that accurately reflects a physical object, and these models afford a number of significant advantages in terms of the management of physical objects.
For example, using a digital twin it is possible to predict and optimise system behaviours. For example, using these models it is possible to predict future behaviours of a system and it is possible to then improve process productivity.
A digital twin can be used to perform continuous monitoring via real-time data acquisition and can provide information to make better business decisions and control a physical system. They can also provide a testing platform to verify different scenarios to choose the most efficient one and to increase the system performance.
A digital twin can continuously learn and update itself so as to better reflect the near-real-time status and operating conditions of the physical system it is twining. Updates may come from various sources, including a human expert familiar with the system’s operation. Embedded sensors within the large physical system can provide information about the system or its operating environment. It is also possible to get updates from a connected artificial intelligence (AI) and machine learning system. The digital twin may incorporate historical data from past operations.
One characteristic property of digital twins is the connection and real-time data exchange that allows for the continuous or periodic synchronisation of the Virtual Object and Physical Object and while the flow information is mainly from the Physical Object to the Virtual Object, it is quite possible for the Virtual Object to send data and information the other way.
During the product life cycle, digital twins play a crucial role in reducing cost and time as customers incorporate them into design and test workflows. Industry pioneers are constantly exploring new means to import and export digital twin data as well as to aggregate and share specific models, schemas, and data across the product spectrum. They are also developing the notion of bionic digital twins or integrated digital twins to represent complex network systems that incorporate live hardware operational devices or hardware emulators as part of the overall digital twin.
Design verification (simulation) and hardware validation take nearly two-thirds of product development time. This is an area where industries hope to see some significant improvements. There are many similarities between these two activities, and both strive to characterise a design based on a set of requirements. One activity occurs within the virtual digital twin domain, and the other takes place within the physical test domain. It is possible to connect these tasks through digital threads, allowing the digital twin to deliver tangible benefits.
AI Augments Digital Twins
Advancements in artificial intelligence and machine learning are playing an increasingly important role in making digital twins more powerful and capable. With AI and automated learning, a digital twin can dynamically improve the quality and validity of predictions under various operational scenarios. When machine learning-based models integrate into the digital twin, designers can scale these models to large networks with tens or hundreds of nodes and represent these diverse real-world situations.
Data analysis aims to connect the data and the context for the data — that is, the results the digital twin produces and the site conditions under which it operates. One can do this using faster-than-real-time execution of the models to create experiments with different variations of the operating parameters, then feed the results into a machine learning algorithm. The algorithm will then recommend configurations to optimise performance under the given scenario.
This type of AI-powered digital twin helps monitor and improve the physical system. So here is where one can take digital twins beyond the realm of product design or product testing into monitoring and improving the performance of operational systems.
In addition, a digital twin offers the advantage of generating synthetic data with the required parametrization while replicating previous tests. It is important that machine learning training generates similar data sets and has the informational complexity to fully excise underlining algorithms. With physical systems alone, this is nearly impossible. High-fidelity simulation models like digital twins are the only ones that allow simultaneous operation.
Customers often want to test early prototypes in this context ahead of compliance and conformance testing to understand how their designs will perform in the real world. The digital twin represents this virtual world where the design lives and the prototype will go through testing. So, there are two distinct focuses for AI and machine learning.
In the first case, machine learning models represent one or more of the components or subsystems of the virtual world. In these scenarios, AI and machine learning capture the scale, fidelity, and diversity of testing and analysis conditions and capabilities of the digital twin platform.
Second, and perhaps more important, is the concept of synthetic data generation, which involves generating data within a virtual world. One can use this context to train and assess how well a machine learning system performs under diverse situations in the real world. A physical system cannot do all these tasks because of the lack of time, money, or tractability. If one intends to widely deploy AI and machine learning, these types of network digital twins will be crucial.
Digital Twins are opening up new possibilities to optimise, simulate, predict, and control the behaviour of physical processes. Their growth will be based on complementary technologies, such as AI, IoT and big data analysis, while network connectivity will also be critical as it enables the data transfer from the physical object to be processed by the virtual counterpart.
Author details: Paris Akhshi, Product Marketing Manager, Keysight Technologies