Industry Innovation in the Mobility and Automotive Sector and the Importance of Collaboration to Stay Up-To-Date

Industry Innovation in the Mobility and Automotive Sector and the Importance of Collaboration to Stay Up-To-Date

The constant evolution of new technologies is well beyond any expectations that may have been held a few decades back. Artificial intelligence, machine learning, cloud computing, computer vision…The list goes on. In this regard, mobility is no exception. In this rapidly changing environment, the whole sector is struggling to keep up while craving to stay a step ahead by investing in innovation.

Machine Learning (ML), essential in many Artificial Intelligence applications, is a buzzword referring to different methodologies. These methodologies aim at extracting knowledge from data to make predictions, identify patterns, and eventually support decision-making. Scientific Machine Learning (SciML) complements standard ML by incorporating the physical knowledge of the systems to analyze. Taking into account the pertinent mathematical and physical models, SciML targets data-intensive applications, enhancing modeling and simulation, and devising intelligent automation and decision support systems. SciML main features are:

  1. Domain-awareness; that is using all knowledge available contained in the models
  2. Interpretability; that is producing results explainable and understandable
  3. Robustness and Credibility; that is assessing and controlling uncertainty, stability and accuracy.

LaCàN is a research lab at the Universitat Politècnica de Catalunya · BarcelonaTech (UPC) focused on Computational Engineering and with long expertise and capabilities in Mathematical and Computational Modeling. An important share of LaCàn activities focus on SciML solutions for mobility and automotive applications. This includes developing Reduced-Order Models for structural and fluid mechanics automotive applications.

Often, using these theoretical models to solve practical applications is still a challenge. While it is important to stay up to date, industry firms and public entities, even those that pride themselves in being pioneers, end up being caught up in everyday operations. Investing on research and innovation internally by developing specific departments is an undoubtedly robust strategy. However, relying solely on internal sources may limit the links to the research and innovation community and narrow down the potential for innovation possibilities in areas that are not an entity’s usual playfield. This is precisely where CARNET makes a difference. With deep ties to industry players and a direct link to academia, it provides a link between technological innovation developed at universities and industrial implementation. This is the case of projects in which LaCàN collaborated with SEAT and Volkswagen to produce tailored-modeling solutions supporting decision making.

Through its links, CARNET helps its industrial partners to stay informed of the latest developments in research. Being aware of the right information early and linking it to existing problems plays a fundamental role when developing better services and products.

Coming back to mathematical models, for instance during the design phase for automotive applications, parametric solutions for internal and external air flows, provide real-time predictions when the designer explores new configurations. Moreover, constructing surrogate models for crashworthiness responses provides means to quantify the uncertainty of the system at an affordable computational cost. Surrogates for external flow problems are also providing real-time input for advanced driving assistance systems to react against wind gusts, which are being implemented in all EOMs at the moment. These specific topics are part of the research carried out by LaCàN in the framework of three doctoral theses co-supervised with researchers from SEAT and Volkswagen: “Complexity reduction in parametric flow problems via Nonintrusive Proper Generalised Decomposition in OpenFOAM” by V. Tsiolakis, “Quantifying uncertainty in complex automotive crashworthiness computational models: development of methodologies and implementation in VPS/Pamcrash” by M. Rocas and “Static and dynamic global stiffness analysis for automotive pre-design” by F. Cavaliere.

These, the associated papers and all the scientific outcomes, including pieces of software already in use in the industry, are good examples of how Scientific Machine Learning is an important topic for mobility and automotive applications.