Knowledge of the rotor temperature is important for an efficient and secure operation of induction motors, especially in electric vehicles. However, in vehicle applications, measuring the rotor temperature is not possible, making a model-based estimation necessary. In this paper, the modeling of the rotor temperature is presented based on subspace identification, a black-box system identification method. The performed steps to achieve a high quality model are described in detail, including a systematic selection of inputs from the large amount of available signals. The resulting model is applied as an open-loop observer and evaluated using measurements of a specially equipped induction motor on a test bench, where the rotor temperature was measurable. Finally, some remarks are given regarding the on-board implementation of the identified model into a vehicle control unit for open-loop observation of the rotor temperature.

No abstract

No abstract

In literature, identification of models with uncertain parameters is restricted to parameter identification. This paper presents a method for system identification with unknown, but bounded measurement errors for linear, time-invariant systems. For each system output, the method determines a model order in a first step, and estimates the uncertain parameters in a second step, both from interval measurement data due to the unknown, but bounded errors. It results in a discrete-time model description with interval parameters. The method is demonstrated and discussed using simulated data as well as measurements.Zusammenfassung: Zur Gewinnung von Modellen mit unsicheren Parametern sind in der Literatur bisher nur Parameteridentifikationsverfahren bekannt. Dieser Beitrag stellt ein Verfahren zur Systemidentifikation von linearen zeitinvarianten Systemen bei Messdaten mit unbekannten, aber beschränkten Fehlern vor. Für jeden Ausgang wird dabei zunächst die Modellordnung ermittelt, und dann die unsicheren Parameter eines zeitdiskreten Modells in Form von Intervallen geschätzt. Nach der Vorstellung des Verfahrens wird es im Beitrag anhand von simulierten und echten Messdaten diskutiert.

No abstract