SOME BASIC DIGITAL TWINNING CONCEPTS
Digital twinning has been identified by Gartner as one of 2019’s top 10 strategic technology trends. It has plenty of applications:
- HAZOP & virtual commissioning: studies of probability and risks
- Production monitoring and optimization
- Predictive maintenance
- Quality management
- Inventory optimization and supply-chain optimization
- Product monitoring
- Product development
Some literature is available on the topic but few references or established examples.
What do we understand by a digital twin?
A Digital Twin is a model capable of rendering the state and behavior of a real asset in (close to) real-time. Digital twinning has therefore been taken up by Airbus to build its A-320, by McLaren to optimize its Formula-1 racing and by Masseratti to design its new models.
Which are the components of a digital twin?
Three basic components:
- Digital SIMULATION (Modelization) : a model that defines the system
- Smart automation and Internet of Things (IoT)connected sensors: inter-relation between the real system and the model through IoT devices
- BIG DATA Analytics & AI: layer to exploit the data generated by the model
Another important factor is the advanced human-machine interface with mainly two technologies:
- Virtual Reality and Augmented Reality (VR/AR)
- 3D Computer-Aided Design (CAD)
As for the classification of digital twins, there are two possible approaches. The first one distinguishes three types of digital twin:
- Equations-based (first principles): A model-based digital twin uses equations to describe the behavior of the industrial systems
- Data-driven(statistics, ML, black box): System behavior is rarely observed during engineering. During the production process, however, data becomes available that allows the creation of a data-driven digital twin.
- Hybrid (reduced models): It is often impossible to simulate the full physical behavior of a system in real-time, so reduced models must be used.
Moreover, the most common classification depends on the type of process to be virtualized. So we can speak about
- Continuous (& Batch) process: for example in oil or gas plants, which is modelled by differential & algebraic equations
- Discrete system: for example a bottling plant where modelling can be discrete events, agent-based or system dynamics
How can artificial intelligence be used in digital twins?
Traditionally machine-learning techniques have been carried out to perform these activities:
- Producing synthetic data for supervised machine-learning techniques to solve predictive maintenance problems
- Data validation and reconciliation
- Detect outliers and data anomalies
- Equipment or sensor faults
But other artificial-intelligence approaches look set to become significant in the future:
- Production systems are embedded in organizations. This introduces Digital Twin of an Organization (DTO). It allows us to analyze and optimize a product, service or process in a digital instance to replicate what works or address issues before they turn into real-world problems. Process mining could become a data-driven approach to building a DTO.
- In addition to this, digital twins can also be used in conjunction with Reinforcement Learning to acquire knowledge about the process and thus be able to bring it up to its optimal control value.
In sum, to build digital twins you have to combine knowledge of the domain, models and computational overload and ensure full maintainability over the full life-cycle of their physical counterpart. It should also be borne in mind here that, in order to ensure maintainability, you need to have a model that can be operated and then calibrated and validated against observed behavior.
Authors: Nuria Gómez Rojo
Pedro José Hernández Ariznavarreta
Las opiniones vertidas por el autor son enteramente suyas y no siempre representan la opinión de GMV
The author’s views are entirely his own and may not reflect the views of GMV