ADAFI: Adaptive Fidelity Digital Twins for Robust and Intelligent Control Systems

We create new computational methods for digital twin development

Project duration

1.1.2021. – 31.12.2023

Source of funding

Academy of Finland

Total funding

857 196 €

The cyber-physical systems of the celebrated fourth industrial revolution, so-called digital twins – whereby accurate numerical simulation models are operated alongside their physical counterparts – are projected to constitute the backbone of modern industrial automation. While the promise of computer-aided engineering has traditionally been reduction of cost and time in product development, today, digital twins and ubiquitous wireless connectivity extend simulation capabilities into product operations across its lifecycle.

Indeed, condition-based optimization and predictive maintenance are facilitated today by digital twins that can be built based on input-output data or the underlying physics. The big promise of digital twins for the future is, ultimately, in them providing full system autonomy.

From the autonomous systems perspective, there are two key open challenges with digital twins. First, the level of fidelity of the digital twin should be optimal with respect to the control system’s architecture and objective, and adaptable during system operation. Second, the infrastructure embodying the digital twin should be capable of managing vast amounts of real-time data originating from simulation results and on-line measurements, and executing the digital twin at a sufficiently low latency.

Resolving these challenges is paramount to achieving robust control and autonomy across a spectrum of different operating conditions, and it is the objective of this research project.

The project proceeds in the following work packages (WP):

WP 1: Adaptive-fidelity digital twins for robust control systems. New simulation models and methods with flexible architectures will be devised for the large space between extreme computational simplicity and extreme model accuracy: Hybrid models coupling physics and machine learning, and reduced-order models with adaptive architectures. Automatic methods for simulation model creation will also be addressed.

WP 2: Robust control systems for adaptive-fidelity digital twins. New robust control algorithms, methods and principles that exploit adaptivity of simulation model fidelity will be developed. Extensions for multi-agent and multi- objective control systems are considered.

WP 3: Synthesis and system integration. The infrastructure required for practical implementation of WP1 and WP2 will be addressed. Issues such as wireless sensor fusion, cloud/edge computing, management of vast data streams, and ultra-low latency synchronous co-execution of digital twins and physical systems are to be considered.

WP 4: Industrial applications. The methods, models and algorithms developed in WPs 1-3 will be applied in industrial applications. Specific attention is paid on the TUAS eRallycross (eRx) race car power train and drone flight control problems. The usage profiles of the eRx race car are similar to those for modern battery-driven heavy mobile machinery. On the other hand, autonomous robust drone control is imperative for many such manufacturing activities that display significant potential in autonomization, such as welding.