Isbitirici, Abdurrahman
(2025)
Data-driven mass estimation of heavy-duty vehicles, [Dissertation thesis], Alma Mater Studiorum Università di Bologna.
Dottorato di ricerca in
Automotive per una mobilità intelligente, 36 Ciclo. DOI 10.48676/unibo/amsdottorato/11739.
Documenti full-text disponibili:
![Thesis_AMS_AI.pdf [thumbnail of Thesis_AMS_AI.pdf]](https://amsdottorato.unibo.it/style/images/fileicons/application_pdf.png) |
Documento PDF (English)
- Richiede un lettore di PDF come Xpdf o Adobe Acrobat Reader
Disponibile con Licenza: Salvo eventuali più ampie autorizzazioni dell'autore, la tesi può essere liberamente consultata e può essere effettuato il salvataggio e la stampa di una copia per fini strettamente personali di studio, di ricerca e di insegnamento, con espresso divieto di qualunque utilizzo direttamente o indirettamente commerciale. Ogni altro diritto sul materiale è riservato.
Download (7MB)
|
Abstract
This study investigates mixed model- and learning-based approaches to the mass estimation problem in heavy-duty vehicles. Effective mass estimation allows for more precise adjustments to engine power, braking systems, and suspension settings, leading to improved vehicle handling, fuel efficiency, and overall safety. Although direct mass measurement using sensors is a viable option, the substantial costs and complexities associated with sensor maintenance, integration, and calibration drive the search for alternative solutions.
The objective of this thesis is to introduce a methodology to mass estimation problem in heavy-duty vehicles, which harnesses the advantages of model-based and learning-based estimation methods. The proposed methodology builds upon the integration of Long-Short Term Memory (LSTM), which is a type of recurrent neural network, that supervises a Recursive Least Squares (RLS) vehicle mass estimator. The RLS estimator relies on a longitudinal vehicle dynamical model. The supervisory LSTM network is offline trained to recognize when the vehicle is operated such that the RLS estimator leads to an estimate with the desired accuracy while online enables the mass estimate update by the RLS estimator based on signals that include vehicle speed, longitudinal acceleration, engine torque, and engine speed.
The LSTM network is trained and tested in this thesis work using datasets artificially generated by a widely used simulation environment called TruckMaker. The simulation results demonstrate that the proposed methodology effectively forecasts the reliability of the RLS mass estimator, showcasing the potential of LSTM networks in enhancing the accuracy and trustworthiness of mass estimation of heavy-duty vehicles.
This thesis also presents a benchmark learning-based approach to mass estimation in heavy vehicles, that uses an LSTM network. In this case, a two-layer LSTM network is designed that utilizes vehicle speed, longitudinal acceleration, engine speed and engine torque to estimate the vehicle mass.
Abstract
This study investigates mixed model- and learning-based approaches to the mass estimation problem in heavy-duty vehicles. Effective mass estimation allows for more precise adjustments to engine power, braking systems, and suspension settings, leading to improved vehicle handling, fuel efficiency, and overall safety. Although direct mass measurement using sensors is a viable option, the substantial costs and complexities associated with sensor maintenance, integration, and calibration drive the search for alternative solutions.
The objective of this thesis is to introduce a methodology to mass estimation problem in heavy-duty vehicles, which harnesses the advantages of model-based and learning-based estimation methods. The proposed methodology builds upon the integration of Long-Short Term Memory (LSTM), which is a type of recurrent neural network, that supervises a Recursive Least Squares (RLS) vehicle mass estimator. The RLS estimator relies on a longitudinal vehicle dynamical model. The supervisory LSTM network is offline trained to recognize when the vehicle is operated such that the RLS estimator leads to an estimate with the desired accuracy while online enables the mass estimate update by the RLS estimator based on signals that include vehicle speed, longitudinal acceleration, engine torque, and engine speed.
The LSTM network is trained and tested in this thesis work using datasets artificially generated by a widely used simulation environment called TruckMaker. The simulation results demonstrate that the proposed methodology effectively forecasts the reliability of the RLS mass estimator, showcasing the potential of LSTM networks in enhancing the accuracy and trustworthiness of mass estimation of heavy-duty vehicles.
This thesis also presents a benchmark learning-based approach to mass estimation in heavy vehicles, that uses an LSTM network. In this case, a two-layer LSTM network is designed that utilizes vehicle speed, longitudinal acceleration, engine speed and engine torque to estimate the vehicle mass.
Tipologia del documento
Tesi di dottorato
Autore
Isbitirici, Abdurrahman
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
36
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
mass estimation, long-short term memory, recursive least squares
DOI
10.48676/unibo/amsdottorato/11739
Data di discussione
30 Gennaio 2025
URI
Altri metadati
Tipologia del documento
Tesi di dottorato
Autore
Isbitirici, Abdurrahman
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
36
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
mass estimation, long-short term memory, recursive least squares
DOI
10.48676/unibo/amsdottorato/11739
Data di discussione
30 Gennaio 2025
URI
Statistica sui download
Gestione del documento: