Schmiedt, Marius (2024)
Machine Learning Based Calibration of Dual Clutch Transmissions for Optimizing the Launch Behavior of Passenger Vehicles.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00027365
Dissertation, Erstveröffentlichung, Verlagsversion
Kurzbeschreibung (Abstract)
The drivability of a vehicle is strongly affected by its transmission. Especially dual clutch transmissions (DCT) offer the chance of a comfortable drivability through its ability of blending the torque during gear shifts from one clutch to the other (jerkless shifting). Another advantage is the higher efficiency compared to torque converters. These advantages come with the drawback of a high control effort for the clutch engagement of the two clutches. The control effort is handled with software functions (developed using model-based programming languages) deployed on the transmission control unit (TCU) with adjustable control parameters (calibration parameters). With these control parameters e.g., the driving behavior is adjustable for different vehicle, engine combinations. Calibration engineers set these parameters at different ambient conditions to comply with customer requirements in an iterative time-consuming process on costly test trips. Therefore, costs are increasing with increasing control opportunities. An approach for decreasing these costs is to automate the optimization of the calibration parameters. Several approaches have already been introduced but some suffer from lack of stability or time efficiency. Hence, to optimize these parameters the target state optimization (TSO) algorithm is illustrated where a target state is approached with a hybrid solution of reinforcement learning (RL) and supervised learning (SL) to overcome existing drawbacks. Since particularly at low speeds the transmission behavior must meet the intention of the driver (drivers tend to be more perceptive at low speeds) the control of the launch behavior is crucial and is therefore investigated in this study. The algorithm is applied in different environments e.g., in a software in the loop (SiL) environment as well as in different test vehicles to optimize the launch behavior and to verify if a deployment in existing development processes is possible. Further the application in different environments such as in different test vehicles proves the ability of the TSO algorithm to generalize.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2024 | ||||
Autor(en): | Schmiedt, Marius | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Machine Learning Based Calibration of Dual Clutch Transmissions for Optimizing the Launch Behavior of Passenger Vehicles | ||||
Sprache: | Englisch | ||||
Referenten: | Rinderknecht, Prof. Dr. Stephan ; Konigorski, Prof. Dr. Ulrich | ||||
Publikationsjahr: | 17 Juni 2024 | ||||
Ort: | Darmstadt | ||||
Kollation: | XIII, 110 Seiten | ||||
Datum der mündlichen Prüfung: | 28 Februar 2024 | ||||
DOI: | 10.26083/tuprints-00027365 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/27365 | ||||
Kurzbeschreibung (Abstract): | The drivability of a vehicle is strongly affected by its transmission. Especially dual clutch transmissions (DCT) offer the chance of a comfortable drivability through its ability of blending the torque during gear shifts from one clutch to the other (jerkless shifting). Another advantage is the higher efficiency compared to torque converters. These advantages come with the drawback of a high control effort for the clutch engagement of the two clutches. The control effort is handled with software functions (developed using model-based programming languages) deployed on the transmission control unit (TCU) with adjustable control parameters (calibration parameters). With these control parameters e.g., the driving behavior is adjustable for different vehicle, engine combinations. Calibration engineers set these parameters at different ambient conditions to comply with customer requirements in an iterative time-consuming process on costly test trips. Therefore, costs are increasing with increasing control opportunities. An approach for decreasing these costs is to automate the optimization of the calibration parameters. Several approaches have already been introduced but some suffer from lack of stability or time efficiency. Hence, to optimize these parameters the target state optimization (TSO) algorithm is illustrated where a target state is approached with a hybrid solution of reinforcement learning (RL) and supervised learning (SL) to overcome existing drawbacks. Since particularly at low speeds the transmission behavior must meet the intention of the driver (drivers tend to be more perceptive at low speeds) the control of the launch behavior is crucial and is therefore investigated in this study. The algorithm is applied in different environments e.g., in a software in the loop (SiL) environment as well as in different test vehicles to optimize the launch behavior and to verify if a deployment in existing development processes is possible. Further the application in different environments such as in different test vehicles proves the ability of the TSO algorithm to generalize. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Freie Schlagworte: | machine learning, reinforcement learning, supervised learning, deep learning, dual clutch transmission, calibration, target state optimization, TSO | ||||
Status: | Verlagsversion | ||||
URN: | urn:nbn:de:tuda-tuprints-273659 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 600 Technik, Medizin, angewandte Wissenschaften > 620 Ingenieurwissenschaften und Maschinenbau | ||||
Fachbereich(e)/-gebiet(e): | 16 Fachbereich Maschinenbau 16 Fachbereich Maschinenbau > Institut für Mechatronische Systeme im Maschinenbau (IMS) 16 Fachbereich Maschinenbau > Institut für Mechatronische Systeme im Maschinenbau (IMS) > Fahrzeugantriebe |
||||
TU-Projekte: | Magna Powertrain|4500554480|Untersuchung von Anf | ||||
Hinterlegungsdatum: | 17 Jun 2024 12:06 | ||||
Letzte Änderung: | 18 Jun 2024 05:37 | ||||
PPN: | |||||
Referenten: | Rinderknecht, Prof. Dr. Stephan ; Konigorski, Prof. Dr. Ulrich | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 28 Februar 2024 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |