Anzeige
Erschienen in:
2021 | OriginalPaper | Buchkapitel
Approach for Online Design of Experiments with Additional Constraint Modeling
verfasst von : Adrian Prochaska, Julien Pillas, Klaus Lüpkes, Yagiz Dursun, Reiner Pätzold, Bernard Bäker
Erschienen in: Der Antrieb von morgen 2020
Verlag: Springer Fachmedien Wiesbaden
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
The model-based optimization with empirical models is the state-of-the-art method for handling complexity in powertrain calibration. The test plans, used for the identification of these black-box-models, are designed before the test execution. This procedure requires expertise and prior knowledge about the process to be identified (offline). The kind of planning algorithm and the number of test points are highly dependent on the expected process behavior. If the results are not as expected after the test execution, a new iteration of the whole process of test planning and execution is performed.
To prevent those time-consuming iterations, the test plans can be created adaptively during the test execution (online). In [1] a method to actively place measurement points based on hierarchical local model trees (HiLoMoT) was introduced. In [2] this approach was used to actively learn models for more than one output dimension. The method proved to be advantageous for automotive applications: The number of measurement points for identifying the models is reduced. Furthermore, the expertise and prior knowledge, needed to apply these methods, are decreased.
In [3] an approach permitting or forbidding measurement points due to a constraint model was introduced. This method is called Online-DoE with Constraint Modeling (ODCM). It uses all measured points to predict whether a constraint is exceeded and decides online whether the next offline-planned measurement point is skipped. The method was applied to an engine calibration problem.
The following contribution combines both approaches and shows a calibration use-case. During test execution, measurements of the dynamic behavior of the drivetrain are evaluated using objective static criteria. This enables the usage of described algorithms, which are restricted to systems with static input-output-behavior for dynamic calibration applications. The HiLoMoT-model actively learns and places new measurement points. An algorithm similar to the ODCM-Algorithm prevents placement of measurement points in domains where the constraints are exceeded. The results of the drivability calibration use-case are presented and discussed critically.