About Greydient
The GREYDIENT innovative training network aims at training a next generation of Early Stage Researchers (ESR) to fully sustain the ongoing transition of European personal mobility towards safe and reliable intelligent systems via the recently introduced framework of grey-box modelling approaches.
One of the main challenges that we currently face in this context is the integration of the data captured from the plenitude of sensors that are involved in a particular road-traffic scenario, ranging from monitoring car-component loading situations to power network-reliability estimations. The aim is to fully exploit the potential of merging these data with advanced computational models of components and systems that are widely available in industry in order to fully assess the momentarily safety.
Grey-box models are an answer to this pressing issue, as they are aimed at optimally integrating (black-box) data driven machine learning tools with (white-box) simulation models to greatly surpass the performance of either framework separately.
Therefore, GREYDIENT will train its ESR’s in a wide spectrum of fields, including the modelling, propagation and quantification of the relevant variabilities, the application of big data and machine learning methods, as well as the optimal combination of data-driven approaches with numerical models.
All our ESR’s will obtain a PhD from an internationally respected university, build experience in communicating and disseminating their work, applying their research skills in a non-academic context and receive in-depth training in transferable skills such as commercialization, collaboration and entrepreneurship.
What is a grey-box model?
White box models are based on first-principle equations that describe the actual physics to represent the structural behavior of the model and so require a very deep insight into the nature of the problem. Perhaps the best-known example of a white box model is a finite-element model, but there are many other examples. Black box models, on the other hand, start from observational data on the structure or system and then construct a mathematical model based on, for example, machine-learning approaches. While these techniques are capable of making valid predictions about a structure or system, their abilities to extrapolate towards ‘unseen’ problems are often very poor. Further, large amounts of data are usually required for their accurate training.
Grey-box models combine both approaches to relax the need to exactly model the underlying physics, while requiring considerably less data than a pure black box approach. For instance, a white box model can be used to fill in the ‘gaps’ in a black box model when it has to make predictions about phenomena it has never seen before. Alternatively, a white box model can be applied during the training of a black box model to steer the convergence of the underlying machine-learning approaches, and so require less data to obtain an optimal model performance.
Finally, black and white boxes can be applied in a fully parallel context (e.g., as a grey-box virtual twin). Such a parallel approach yields both a continuous update and calibration of the white box model, as well as information on the system quality and its performance, including, e.g., variable operating conditions.
WP1
Developing the grey-box methodology
This relates to grey box models for uncertainty quantification (UQ) and reliability assessment (RA). Even though UQ and RA based on black box and white box models are well established, grey box modelling for UQ & RA is wide open for progress. GREYDIENT will advance the state of the art.
WP2
Designing vehicles that are reliable and safe
This involves the application of new grey-box modelling techniques in the context of designing reliable and safe components and structures (considered as an assembly of components). The grey-box models will be applied at two levels of simulation complexity: to assess the reliability of car components and to assess the safety of the whole car.
WP3
Continuous monitoring, optimizing and controlling of processes and systems
Continuous monitoring, optimizing and controlling of processes and systems applies grey-box approaches for reliability & resilience analyses and monitoring continuous systems such as production processes and energy grids, with a specific focus on those that are relevant for the uptake of intelligent mobility systems. Two specific types of applications are envisaged: real-time monitoring & performance assessment and robust optimization.
