Digital models are used in a variety of applications and development workflows. They can vary in degree in how they match a physical device. Models can be used as a digital twin or in simulations in addition to models used to design a system.
Current design models are a representation of a physical entity, and it’s typically used to describe what a physical entity will look like. This can be a 2D or 3D architectural model of a building or a device such as a car. The model provides dimensions and possibly descriptions of materials that would be used in construction.
A digital twin is a model that has the latest sensor data associated with a matching physical device. A digital twin is often used in process control and product operations to help monitor or control a remote system. The model doesn’t necessarily need to exactly replicate the physical device. It may even be a 2D representation, but it’s typically combined with other models to provide a context for the information that can be presented or examined.
Most process-control systems that deliver sensor data to a control program provide at least a limited digital model of a component within the system. However, with new developments the model can now be more robust and combined with other tools like Virtual Reality. For example, a heat sensor might show what part of a device is hotter by showing that portion of a model using false coloring with red indicating higher temperatures.
Digital models used in simulations often have the same type of sensor information and controls of a digital twin, but the information is generated and manipulated as part of the simulation. The simulation may replicate what could happen in the real world, but not what’s currently happening.
A digital twin can be used as a starting point for a simulation model that extrapolates how a system would operate in the future. The degree and accuracy of these simulations can vary depending on the implementation of the simulation and what type of results are desired.
For example, a digital twin of a gas engine could simply track material consumption, power output, and heat output, but not the actual movement of components within the engine. This level of simulation may be sufficient for checking out how a vehicle would operate when using such an engine.
On the other hand, if the desired results involve how durable a particular part would be within the engine, then the level of detail with respect to the engine would have to be greater. Likewise, simulation of an autonomous vehicle needs to know the output and control characteristics of the engine, but not the details within the engine.
Digital twins, and simulation models may share all of these aspects, depending on their function, although often a specific tool will create and manipulate a model. For example, a drawing package may be used to create a digital model, and then a process control system would use that model as the basis for a digital twin to provide the linkage between the digital twin sensors and controls with those in the real world.
Likewise, a model used in simulation may have characteristics added so that physical simulation is possible. This might include details about the virtual materials used in the model, which in turn would enable the simulation tools to replicate how the model will react during the simulation.
Any system component may incur different models that vary in the degree they replicate the actual component, as well as how they react and what kind of information can be associated with them.
Models may have different purposes, but they may also share common descriptions such as details about dimensions, material attributes, etc. Many models will be used by multiple applications for different purposes, from showing the status of a current system to simulating a device that has yet to be constructed.
The concept of a Digital Twin is mostly applied to the case where conclusions are drawn from a physical operating product and network platform. But in the case of virtual prototyping, the Digital Twin concept is applied to an old practice in mechanical hardware.
One approach to developing a smart connected product, one that hooks up with an network platform, is to just build it. Just put it together piecemeal. Throw some sensors on a product. Wire that to some kind of embedded system. Wire that to your antenna. Start sending data to the network platform. You and your organisation can actually learn a lot from going through that exercise.
While that needs to be done, you will quickly run into limitations on the experiments you can conduct with physical prototypes. Swapping out a sensor isn’t easy when it’s soldered in place. There might not be room, physically, for the sensor you really need for accurate measurement. You might run into too much electromagnetic interference for the antenna you planned to use.
Working through these issues is new to some organisations as they transition traditionally mechanical products to smart connected ones. However, the problems associated with resolving issues through physical prototyping aren’t new.
In fact, physical prototyping is an old concept when it comes to hardware. Long ago, mechanical and electrical engineers figured out that modeling and simulating a design virtually means you are more likely to get it right the first time when you get to prototyping and testing.
The benefits of an approach utilising virtual prototyping with digital twins are many. You have fewer rounds of prototyping, saving money and time. You stay on schedule. You stay on budget.
So while virtual prototyping is new to some organisations, this approach has advantages when applied to the development of linked smart, connected products and network platforms. In fact, Digital Twins are a key enabler.
You first need to set up the digital model component of a Digital Twin .
Numerical Models use machine learning and artificial intelligence. These applications or agents either extrapolate that data and/or correlate data to existing events. Both are an effort to predict future behaviour.
1D Simulation models are a combination of flow diagrams with equations or formulas behind the blocks that simulate the performance of embedded multi-disciplinary engineering systems. These models can provide deeper insights into ongoing operation.
3D Simulation models, often in the form of multi-body dynamics, are commonly used to predict the dynamics and structural performance of products. to help characterise requirements of ongoing operations.
For scenarios based on engineering physics or asset operation, no prototype or operating product exists. As such, there is no sensor data to feed this digital model. However, the model can be fed historical sensor data from prior products or even from past physical tests or operational data.
In the worst case, a set of inputs can be modeled using statistical analyses or even a higher level simulation, such as a multi-body dynamics model. This creates a set of input data that can be fed to the digital model.
The combination of the digital model and the input is enough to get started. You run the model as a simulation, generating data from virtual sensors, which are points of measurements from the simulation.
In this application, that virtually generated data is used instead of physically recorded data from sensors and fed to the network platform as if it were receiving streaming data from a running product. Only, in this case, there is no physical product. There is only a virtual product that is running in a simulation.
In this scenario, you overcome many issues that you might experience when trying to physically prototype a smart, connected product. You can change anything related to the sensor configuration, including placement or type.
You are not limited by network bandwidth other than the limitation between the compute resource running the Digital Twin and the one running platform. You can change the product’s design in terms of mechanical or electrical hardware, embedded systems and more. There is a tremendous amount of flexibility with this approach.
Engineering organisations have been using the “Get It Right the First Time” principle to avoid all sorts of difficult development disruptions for decades. At this point, applying it to mechanical and electrical hardware is a no-brainer.
Digital twin is becoming increasingly relevant to systems engineering and, more specifically, to model-based system engineering. A digital twin, like a virtual prototype, is a dynamic digital representation of a physical system. However, unlike a virtual prototype, a digital twin is a virtual instance of a physical system “Twin” that is continually updated with the latter’s performance, maintenance, and condition status data throughout the physical system’s life cycle.
Here we present an overall vision and rationale for incorporating digital twin technology into systems engineering and examine the benefits of integrating digital twins with system simulation and networks and provide specific examples of the use and benefits of digital twin technology in different industries. We are making a strong recommendation to make digital twin technology an integral part of system engineering model experimentation testbeds.
While virtual models tend to be generic representations of a system, part, or a family of parts, the digital twin represents an instance i.e., a particular system or process. Digital twin technology has the potential to reduce the cost of system verification and testing while providing early insights into system behaviour.
The digital twin is ultra-realistic and may consider one or more important and interdependent vehicle systems such as airframe, propulsion, avionics and thermal protection. The extreme requirements of the digital twin necessitate the integration of design of materials and innovative material processing approaches.
The context of operation of the digital twin involves an instrumented testbed in which model-based systems engineering tools e.g., system modeling and verification tools and operational scenario simulations e.g., discrete event simulations, agent-based simulations are used to explore the behaviour of virtual prototypes in a what-if simulation mode under the control of the experiment.
Insights from the operational environment are used to modify the system models used in the virtual prototype. Data supplied by the physical system is used by the virtual prototype to instantiate a digital twin. Subsequently, the digital twin is updated on an ongoing basis, so it mirrors the characteristics and history of the physical twin with high fidelity.
The virtual system model can range from lightweight models to full-up models. The lightweight models reflect simplified structure e.g., simplified geometry and simplified physics e.g., reduced order models to reduce computational load especially in upfront engineering activities. These lightweight models allow simulations of complex systems and system-of-systems with fidelity in the appropriate dimensions to answer questions with minimal computation costs.
A digital twin is first created when the physical system can start providing data to the virtual system model to create a model instance that reflects the structural, performance, maintenance, and operational condition characteristics of the physical system. The physical system can be said to be ”fit for purpose” if the digital twin’s behaviour is analyzed and can be appropriately adjusted for a variety of contingency situations.
For example, computer simulations of the braking system of a car can be run to understand how the vehicle model would perform in different real-world scenarios. This approach is faster and cheaper than building multiple physical vehicles to test. However, computer simulations tend to be confined to current events and operational environments. In other words, they do not have the ability to predict vehicle responses to future/envisioned scenarios. Also, braking systems are networked systems, not merely a combination of mechanical and electronic subsystems.
Being a virtual representation, a digital twin is easier to manipulate and study in a controlled testbed environment than its physical counterpart in the operational environment. This flexibility enables cost-effective exploration of system behaviours and sensitivities to various types of system malfunctions and external disruptions.
Today any digital version of a system, component, or asset is called a digital twin. With this larger interpretation, there are several questions that arise including: 1) Does a physical system have to exist before a digital twin is created? 2) Does the physical system with onboard sensors and processor need to report performance, condition, and maintenance data to the virtual system model before the latter can be called a digital twin? 3) Does the definition of a digital twin have to change because any physical asset today can be made smart with the advanced network technology?
These questions and many more need to be answered about the digital twin and its various uses. Here we present several levels of a virtual representation. Each level has a specific purpose and scope and helps with decision making and answering questions throughout the system’s lifecycle.
Pre-Digital Twin Virtual Prototype is created during upfront engineering. It supports decision-making at the concept design and preliminary design. The virtual prototype is a virtual generic executable system model of the envisioned system that is typically created before the physical prototype is built. Its primary purpose is to mitigate technical risks and uncover issues in upfront engineering.
Like most model-driven approaches, virtual prototyping involves a model of the system early in the design process. However, a virtual prototype is not usually used to derive the final system. This is because a virtual prototype can be a “throwaway” prototype or a “reusable” prototype. The latter can be used to derive the final system. A virtual prototype is mostly used to validate certain key decisions about the system and mitigate specific technical risks early in the design process.
For example, a model could consist of ideal wheels with dry friction contact patch rolling on a surface. It employs a simple i.e., low fidelity model of differential gear to distribute torque equally to the wheels, and reflects properties such as inertia, mass, fixed translation and torque to realise a basic structure of vehicle with mass properties defining gravity and global coordinates system.
The sensors measure absolute position, velocity and acceleration. The trajectory control module provides torques values to the steering and differential gear mechanism. Such low fidelity models can be used, for example, in testing, planning, and decision-making algorithms related to, for example, trajectory control in autonomous vehicles performing lane changes.
Digital Twin virtual system model is capable of incorporating performance, condition and maintenance data from the physical twin. The virtual representation, an instantiation of the generic system model, receives batch updates from the physical system that it uses to support high-level decision making in conceptual design, technology specification, preliminary design and development. Data collection from the physical sensors and computational elements in the physical twin includes both condition status data e.g., battery level, mission performance data e.g., flight hours.
The data is reported back to the digital twin which updates its model including the maintenance schedule for the physical system. Since interaction with the physical system is bidirectional, there is sufficient opportunity for the physical twin to use knowledge acquired from one or more digital twins to improve its performance during real time operation.
At this level, the digital twin is used to explore the behaviour of the physical twin under various what-if scenarios. Being an executable digital representation, it is easy to manipulate when exploring the behaviour of the system in the controlled simulation environment of the testbed. Any deficiencies discovered are used to modify the physical twin with the changes reflected in the digital twin.
For example, a model could include a passenger car with power split hybrid power train. The chassis model has single degree of freedom with mass-and speed-dependent drag properties. The braking subsystem which uses brake pedal position resulting from driver action to calculate brake torque. This affects the driveline. The driveline model consists of four wheels with front wheel drive and ideal differential. The power split device consists of ideal epicyclic gear without losses. An ideal battery with constant voltage source powers DC motor model with inductor, resistor and emf component connected to shaft hub. The engine model with flywheel consists of drive-by-wire accelerator, where the accelerator inputs are converted to output torque. Road models are used to define the inertial frame, gravity, air temperature, wind velocity, gas constant for air, and air pressure.
With respect to developing smart, connected products linked to network platforms, we’re in the early stages. But soon, some organisations are going to want to sidestep all those difficult development problems they’re experiencing.
Leaders are going to realise that building and testing multiple rounds of prototypes is unacceptable. They’re going to realise that many delays completely undermine their competitive market position.
As a result, organistions are going to want to adopt more proven and standardised practices. Virtually prototyping with Digital Twins is not there yet. But given the rush toward new networks, the demand for this practice is expected to only increase.
So now we have this concept of using a Digital Twin to virtually prototype a smart, connected product linked to network platforms. What does that get you? Interestingly, it allows organisations to answer a set of serious questions.
The powerful aspect of this use of a Digital Twin is that you can answer these questions without any physical prototype or testing. Everything is digital. So you are getting smarter about the operation of this product without spending any money to build anything physical.
1. Is the right data being captured from the right physical/virtual sensors in product?
2. Do we need to use a physical sensor to capture this data, or can it be a virtual sensor?
3. Is edge processing required for the sensor data processed on product or network platform?
4. Is the mechanical design right for this product in the context of electrical systems configuration?
5. Are there changes that should be considered to improve placement of sensors?
6. Are there changes that should be made to avoid electromagnetic interference?
7. How will the connected product and network platform work to fulfill requirements?
8. What conclusions can be drawn from the data once it is in network platform?
9. What data event precursors trends must be crunched manually or machine learning be applied?
10. What data trends are precursors to events critical to the smart, connected product?