Search

The Extended Kalman Filter (EKF) is a well-known method for state and parameter estimation in vehicle dynamics. However, for tuning the EKF, knowledge about the process and measurement noise is needed, which is usually unknown. Tuning the noise parameters manually is very time consuming, especially for systems with many states. Automated optimization based on the filtering errors promises less application time and better estimation performance, but also requires computing resources. This work presents two approaches for estimating the noise parameters of an EKF: A particle swarm optimization (PSO) and a gradient-based optimization. The EKF is applied to a nonlinear vehicle model of a tractor-semitrailer for estimating the steering and articulation angle as well as lateral and vertical tire forces based on real measurement data with different trailer loadings. Both methods are compared to each other to achieve the best estimation performance.

Moderne PKW sind umfangreich mit Sensorik sowie darauf basierenden Fahrassistenzsystemen ausgestattet und entwickeln sich stetig weiter hin zu autonomen Fahrzeugen. Eine ähnliche Si- tuation ist bei der Zugmaschine von Sattelzügen zu vernehmen. Eine genauere Betrachtung des Sattelaufliegers offenbart, dass dieser lediglich in einem sehr geringen Maße mit elektronischen Komponenten ausgestattet ist. Dies hat zur Folge, dass aufgrund fehlender fahrdynamischer und sicherheitsrelevanter Informationen der Sattelauflieger sich potenziell zum Hindernis bei der Ent- wicklung autonom fahrender Sattelzüge entwickeln kann. Um dem vorzubeugen, werden in dieser Arbeit Ansätze erarbeitet, mit denen auf Grundlage von heute im Sattelauflieger serienmäßig verfügbarer Sensorik zusätzliche Informationen über fahrdynamische Zustände des Aufliegers generiert werden können. Da die direkte messtechnische Erfassung einiger Zustände teilweise nur mit enormem technischen Aufwand realisierbar ist, kommen für diese häufig Schätzverfahren in Betracht. Grundsätzlich eignen sich zur Schätzung dynamischer Zustände verschiedene Methoden. In der Fahrzeugtechnik wurden in der Vergangenheit überwiegend modellbasierte Ansätze herangezogen. Diese haben zur Eigenschaft, dass sie physikalisch interpretierbar sind, erfordern gleichzeitig jedoch ein tiefgreifen- des Systemverständnis. In der jüngeren Vergangenheit kommen vermehrt datenbasierte Methoden zum Einsatz, motiviert durch die erfolgreiche Anwendung in anderen Forschungsgebieten wie der Bildverarbeitung, der Signalverarbeitung und der Robotik. Datenbasierte Methoden sind in der Lage, komplexe Zusammenhänge auf Grundlage von Messdaten zu erlernen, ohne dass eine detaillierte physikalische Kenntnis über das zugrundeliegende System erforderlich wäre. Darunter leidet die Interpretierbarkeit des Lösungswegs. Im Fokus dieser Arbeit stehen die Analyse und der Vergleich von modell- und datenbasierten Methoden zur echtzeitfähigen Schätzung des Knick- winkels, Modern passenger cars are extensively equipped with sensors as well as driver assistance systems based on them and are constantly developing towards autonomous vehicles. A similar situation can be found in the truck unit of truck-semitrailer combinations. However, if the semitrailer is examined more closely, it can be seen that it is hardly equipped with electronic components. As a result, the semitrailer can potentially become an obstacle in the development of autonomously driving truck-semitrailer combinations due to missing dynamic and safety relevant information. In order to prevent this, in the present work approaches are being developed that can be used to generate additional information about the dynamic states of the semitrailer. These are solely based on sensors that are available as standard in a semitrailer today. As the direct measurement of some states can only be realized with enormous technical effort, state estimation methods are often considered. In general, various methods are suitable for estimating dynamic states. In the past, model-based approaches were mainly used in vehicle technology. These have the property that they can be interpreted physically, but at the same time require a profound understanding of the system. More recently, data-based methods have been increasingly used, motivated by their successful application in other research areas such as image processing, signal processing, and robotics. Data-based methods are able to learn complex physical relationships based on measured data without requiring detailed knowledge of the underlying system. As an outcome, the interpretability of the approach suffers. The focus of this work is the analysis and comparison of model-based and data-based methods for real-time estimation of the hitch angle and the lateral tire forces in a semitrailer which supports condition monitoring of safety relevant chassis components. These do not require additional beyond serial sensors or communication to the tru

Targeting a resource-efficient automotive traffic, modern driver assistance systems include speed optimization algorithms to minimize the vehicle’s energy demand, based on predictive route data. Within these algorithms, the required energy for upcoming operation points has to be determined. This paper presents a model-based approach, to predict the energy demand of a parallel hybrid electrical vehicle, which is suitable to be used in speed optimization algorithms. It relies on separate models for the individual power train components, and is identified for a real test vehicle. On route sections of 5 to 7 km the averaged root mean square error for the state of charge prediction results to 0.91% while the required amount of fuel can be predicted with an averaged root mean square error of 0.05 liters.

Physically motivated parametric models are the basis of several techniques related to control design. Industrial model-based controller tuning methods include pole placement, symmetric optimum and damping optimum. The challenge is that the resulting model-based controller is satisfactory only if the underlying model is appropriate. Typically, a set of potential models is known a priori, but it is not known, which model should be used. So, the critical question in model-based controller tuning is that of model selection. Existing approaches for model selection are mostly based on maximizing accuracy, but there is no reason why the most accurate model should also be the optimal model for control design. Given the overall aim to design a high-performance controller, in this paper the best model is considered as the one that has the potential to give a model-based controller the highest performance. The proposed method identifies parametric candidate models for control design. Then, a nonparametric model is used to predict the actual performance of the various controllers on the real system. A validation with two industry-like testbeds shows success of the method.