Today's drivers of battery electric vehicles must deal with limited driving range in a sparse charging infrastructure. An accurate prediction of energy demand and driving range is therefore important and enables reliable routing and charge planning applications. Predictions of energy demand entail uncertainty, which can be considered directly with the use of probabilistic prediction algorithms. Machine learning algorithms are frequently applied in this context, but data used to train these algorithms are often distributed over a fleet of connected vehicles. Federated learning can be applied in this setting, but predictive uncertainty is typically not considered. We apply an extension of the federated averaging algorithm to learn probabilistic neural networks and linear regression models in a communication-efficient and privacy-preserving manner. We demonstrate the performance advantage of probabilistic prediction models over deterministic prediction models using proper scoring rules. Furthermore, we show that federated learning can improve the standard, driver-individual learning. Using probabilistic predictions, variable safety margins based on destination attainability can be applied, leading to increased effective driving range and reduced travel time.

Virtual powertrain analysis is widely applied in the automotive industry to cope with the increasing complexity and variance of future vehicle propulsion technologies. Since the vehicle-usage behavior has a strong impact on component loads, realistic computation results require accurate assumptions about these boundaries. In this context driving cycles (DCs) are used to represent the system boundaries in vehicle operation. The aim of this paper is to identify multiple characteristic driving cycles (CDCs) from extensive vehicle measurement data which represent the full variety of possible real-driving scenarios. Vehicle measurements are segmented and consumption-relevant features are extracted from each segment. These features are then used to apply clustering and classification techniques to identify characteristic groups that are consequently assigned to different driving environments and driving styles. In order to obtain even more realistic driving scenarios, a data-fusion approach is used to incorporate a road slope signal from a NASA digital elevation model for each segment. Lastly, a genetic mixed-integer optimization algorithm is proposed to efficiently generate representative DCs for each characteristic group of driving segments. The main contribution of this paper is a data-driven identification of the parameter space of real-driving scenarios from extensive vehicle measurements including the implementation of road slope information. The scenarios are represented via a constrained number of compact CDCs which enables comprehensive investigations of new powertrain technologies under average as well as extreme real-driving conditions to develop efficiency-robust powertrain systems.

Automotive radar sensors play a vital role in the current development of autonomous driving. Their ability to detect objects even under adverse conditions makes them indispensable for environment-sensing tasks in autonomous vehicles. As their functional operation must be validated in-place, a fully integrated test system is required. Radar Target Simulators (RTS) are capable of executing end-of-line, over-the-air validation tests by looping back a received and afterward modified radar signal and have been incorporated into existing Vehicle-in-the-Loop (ViL) test beds before. However, the currently available ViL test beds and the RTS systems that they consist of lack the ability to generate authentic radar echoes with respect to their complexity. The paper at hand reviews the current development stage of the research as well as commercial ViL and RTS systems. Furthermore, the concept and implementation of a new test setup for the rapid prototyping and validation of ADAS functions is presented. This represents the first-ever integrated radar validation test system to comprise multiple angle-resolved radar target channels, each capable of generating multiple radar echoes. A measurement campaign that supports this claim has been conducted.

Limited range and charging infrastructure leads to range anxiety of electric vehicle drivers. Current range estimation algorithms are deemed unreliable and large safety margins are reserved to prevent the risk of stranding. Range estimation in general depends on two factors: current battery energy content and the energy consumption forecast on the route to destination. This work aims at improving the latter by enhancing the forecast with a notion of uncertainty. The prediction algorithm itself learns from driver and traffic data in a training set to generate accurate, driver-individual energy consumption forecasts. Thereby, a central part of the algorithm is the explicit evaluation of the traffic situation by classifying the traffic phases. With the help of this methodology, individual forecasts can be made more precise since they are highly dependent on surrounding traffic. To demonstrate the validity of the algorithms, the performance is evaluated using real test drive data comprising multiple drivers. On the basis of the performance evaluation, both the superiority of stochastic algorithms over deterministic predictions and the improvement of predictive performance by evaluating explicit traffic phases can be shown. Implementing the proposed methodology in modern day electric vehicles could reduce range anxiety and ultimately increase acceptance of electric mobility worldwide.

This study focuses on particulate matter emissions from tire–road contact and their investigation using an internal drum test bench. The test bench is equipped with real-road surfaces and has been upgraded to enable real-time measurements of particulate matter. It was found that the road surface changes during the tests due to constant rolling over, influencing the level of emissions significantly. To account for this effect, the micro roughness was characterized before, during, and after the tests. Specific emission values consisting of particle mass and number were determined with summer, all-season, and winter tires for different road conditions, as well as specific longitudinal and lateral forces. It turned out that emissions increase disproportionately with load for both force directions. The winter tire led to the highest emissions across all loads, and the summer tire led to the lowest ones. While lateral forces caused emissions many times higher than longitudinal forces for the summer tire, forces in both directions led to comparable emissions for the all-season and winter tires. Regarding the ambient temperature, a lower one seems to be favorable for summer tires and a higher one seems to be favorable for winter tires. Lastly, particle size distributions during different load conditions show a dependence on load, such that larger particles are emitted with increasing load.

Probabilistic predictions with machine learning are important in many applications. These are commonly done with Bayesian learning algorithms. However, Bayesian learning methods are computationally expensive in comparison with non-Bayesian methods. Furthermore, the data used to train these algorithms are often distributed over a large group of end devices. Federated learning can be applied in this setting in a communication-efficient and privacy-preserving manner but does not include predictive uncertainty. To represent predictive uncertainty in federated learning, our suggestion is to introduce uncertainty in the aggregation step of the algorithm by treating the set of local weights as a posterior distribution for the weights of the global model. We compare our approach to state-of-the-art Bayesian and non-Bayesian probabilistic learning algorithms. By applying proper scoring rules to evaluate the predictive distributions, we show that our approach can achieve similar performance as the benchmark would achieve in a non-distributed setting.