Fraunhofer Institute for Industrial Mathematics ITWM
Fraunhofer Institute for Industrial Mathematics ITWM
Datenbasierte Mathematik in der Fahrzeugindustrie
© iStockphoto
Datenbasierte Mathematik in der Fahrzeugindustrie

Machine Learning for Vehicle Engineering

We use intelligent methods such as Machine Learning in vehicle development, for example to derive driver and operator models from corresponding data sources.

© iStockphoto

Data-Based Mathematics

Data Analysis, System Identification and Machine Learning

In addition to physics-based modelling techniques, data-based mathematics is an important part of our work in the department »Dynamics, Loads and Environmental Data«. For example, we apply classical data analysis methods to usage, load and component data in order to better understand usage and system behavior and draw conclusions for future design and development steps.

In the vehicle sector, a great deal of data is already collected and recorded during normal driving, and data volumes are also generated in planned measurement campaigns (for individual vehicles or vehicle fleets). In addition, large amounts of data from past campaigns are often also available. Here we develop and use modern methods of data analytics that efficiently process and analyze even very large data sets (»Big Data«).

A second important area of work in the field of data-based mathematics and Artificial Intelligence (AI) are methods of system identification and Machine Learning (ML). Here we use both classical techniques and modern approaches to learn dependencies, patterns and dynamic system behavior from existing data, in order to use the trained models for analysis and prediction purposes.

Dynamical Systems on a Vehicle
© Fraunhofer ITWM
Identification of dynamic systems on the vehicle using Machine Learning techniques

A special focus lies on applications in the automotive area, for example the data-based estimation of dynamic systems of vehicular subsystems or the entire vehicle. Among other things, we use recurrent neural networks and Gaussian processes. We use the models learned in this way for prediction, analysis, control, monitoring and predictive maintenance.

We also use Machine Learning methods to derive driver and operator models from corresponding data sources. In particular, reinforcement learning techniques are used, which we apply to driver and operator models as well as to derive control and action models for biomechanical systems.

Related Links: