Predictive Powertrain Management through Driver Behaviour Recognition
MetadataShow full item record
With automotive trends leading towards electrification and inclusion of intelligent technology for advanced driver assistance systems (ADAS), there is a need to research the use of advanced control strategies. This report touches on the development of a powertrain model built in Simulink used for simulation testing and vehicle development. While addressing the needs of incorporating state-of-the-art technology, this report shows how a LiDAR and camera system can function together as ADAS sensors for vehicle detection and range estimation. Lastly, the main purpose of this report is to show how the UWAFT powertrain model and ADAS sensors, along with behaviour recognition software, can be used to reduce emissions and energy consumption while also increasing driveability. Machine learning techniques are used to classify a driver’s behaviour on a spectrum from aggressive to eco-cautious. 288 hours of driver behaviour data is simulated using the UWAFT’s vehicle model built in Simulink. The data is labelled as aggressive, normal, or eco-cautious depending on the scaling factor applied to the drive cycle inputted. Linear discriminant analysis is performed to maximize the separation between classes and reduce the dimensionality. Support vector machines are used to classify the driver’s behaviour. Lastly, fuzzy logic is used to assign a driver an aggressiveness value between 0 and 100. The classifier implemented achieved 81.53% accuracy; however, the aggression value assigned to the data via fuzzy logic is a more accurate representation. Vehicle testing is performed with the use of a closed-loop testing track and a chassis dynamometer. An acceleration test is conducted by applying a wide-open throttle in various operating modes. This identified drive traces that are only achievable in certain modes, thus concluding that if the driver’s behaviour is predicted prior to an acceleration event, the correct operating mode could be selected ahead of time, increasing the driveability. Additionally, a regenerative braking test is conducted on a chassis dynamometer to determine the optimal regen torque parameters for a given braking rate. It is concluded that using the best parameters for a stopping distance of 2 mph/s would result in a 0.003% state of charge gain per second. Therefore, by knowing a driver’s braking behaviour the UWAFT PHEV could select the best parameters for the current drive to decrease energy consumption.
Cite this version of the work
Patrick DiGioacchino (2018). Predictive Powertrain Management through Driver Behaviour Recognition. UWSpace. http://hdl.handle.net/10012/13959