Data-driven models are obtained purely from data without complex theoretical modeling and without explicit model knowledge. This results in black box models whose traceability and reliability constitute a major challenge. This contribution addressed this issue and presents a novel hybrid estimation method, which enhances the reliability of artificial neural networks. Within this method, a simple model based on physical knowledge secures a model based on an artificial neural network. An unscented Kalman filter realizes the interaction of the two individual models. Thereby, a confidence level determines which model is trusted to a greater extent or even entirely. As part of the method for adjusting this confidence level, the input variables of the artificial neural network are related to the data used in training. The more often the artificial neural network has encountered a situation in the training process, the greater the confidence level will be. Finally, the confidence level is used to set the covariances of the unscented Kalman filter. In this contribution, the method is presented using the application of roll angle estimation for passenger cars. By using the hybrid method the reliability of the estimation is increased in comparison to the artificial neural network. For this purpose, sensor malfunctions as well as a sensor failure are simulated. These disturbances are compensated by the introduced method. In addition, the hybrid state estimator increases the estimation quality compared to the individual estimators. The proposed method can be applied to any problem, where knowledge-based models are available to secure data-driven models.
Software sensors are playing an increasingly important role in current vehicle development. Such soft sensors can be based on both physical modeling and data-based modeling. Data-driven modeling is based on building a model purely on captured data which means that no system knowledge is required for the application. At the same time, hyperparameters have a particularly large influence on the quality of the model. These parameters influence the architecture and the training process of the machine learning algorithm. This paper deals with the comparison of different hyperparameter optimization methods for the design of a roll angle estimator based on an artificial neural network. The comparison is drawn based on a pre-generated simulation data set created with ISO standard driving maneuvers. Four different optimization methods are used for the comparison. Random Search and Hyperband are two similar methods based purely on randomness, whereas Bayesian Optimization and the genetic algorithm are knowledge-based methods, i.e., they process information from previous iterations. The objective function for all optimization methods consists of the root mean square error of the training process and the reference data generated in the simulation. To guarantee a meaningful result, k-fold cross-validation is integrated for the training process. Finally, all methods are applied to the predefined parameter space. It is shown that the knowledge-based methods lead to better results. In particular, the Genetic Algorithm leads to promising solutions in this application.
Long pulse modulator systems like the electron-positron compact linear collider (CLIC) with a pulse length of 140μs require a bouncer for voltage drop compensation. In this paper, an active bouncer system, consisting of four bouncer modules, in series to the main capacitor bank, is investigated. The system's transfer function including the matrix pulse transformer is analysed and the output voltage ripple induced by the bouncer system is designed to be smaller than 5 ppm due to demanding repeatability requirement of CLIC. Based on a loss analysis is conducted resulting in a total bouncer system volume of 52dm 3 and a reduction in the global efficiency (grid to klystron) of the modulator system of only 0.45%.