At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

The Institute of Mechatronic Systems in Mechanical Engineering is conducting research into energy-efficient drive concepts for electric robot vehicles. To better assess the potential for future drive and transport concepts, these must be tested using simulations. In order to be able to assess the influence of a traffic environment, a traffic simulation of the city centre of Darmstadt is therefore used at the IMS. This is implemented in the SUMO simulation software. An initial approach to parameterizing the simulation was developed in previous works and is now to be continued.

Im Projekt „DigiTain“ wird am IMS an der Verbesserung der Betriebsstrategie von Brennstoffzellen-fahrzeugen geforscht. Aktuelle Ergebnisse zeigen, dass mittels Geschwindigkeitsprädiktion eine Effizienzsteigerung erreicht werden kann.

In den bisherigen Untersuchungen wurden das Thermalsystem des Brennstoffzellensystems sowie die Alterung von Batterie und Brennstoffzelle (BZ) unberücksichtigt gelassen. Im Rahmen dieser Arbeit soll durch deren Ergänzung die Simulation realitätsnäher werden sowie Alterung von Batterie und Brennstoffzelle in der Betriebsstrategie berücksichtigt werden.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

At the IMS, the potential for increasing the efficiency of powertrains by knowing the speed to be expected in the next few seconds is being investigated on the basis of AI-based predictions.

The Kolmogorov-Arnold networks (KAN) approach presented in April 2024 promises faster training, improved accuracy and interpretability compared to conventional multi-layer perceptrons (MLP). So far, only a simple Matlab implementation of KANs with basic functionality exists.

The aim of this advertised work is to realise a KAN implementation against the background of the Matlab Deep Learning Toolbox and to use it for speed prediction.

A data set comprising over 29,000 kilometres is available as a data basis.

Shared test platforms can realize multi-functionality investigations at the early stage of the vehicle powertrain development.

At IMS, the CONNECT (powertrain test bench) and Driveception (driving simulator) are integrated for a distributed measurement of powertrain dynamics and their impact on subjective perception during the launch process. The student work will be focused on the realization of the test bench networking and its real-time performance for the launch process study.