Intelligent Vehicle Development through Scalable Data Collection Processes and Simulation
MetadataShow full item record
With current automotive trends in both vehicle electrification and intelligent features such as Advanced Driver-Assistance Systems (ADAS), there is a significant need for a modern vehicle development process which makes use of big data. In the following report, a scalable, phone-based, driving data collection system is developed and applied to powertrain design through a motivating example. Initial project efforts are directed towards the development of both a data collection platform and a system which is capable of interpreting and storing the collected drive data. The developed UWAFT Innovation Platform (UIP) and Monocular Vision Pipeline (MVP) are a functional system which attempt to precipitate crowdsourcing of data collection through a low system cost and open software approach. In an application of this platform data is collected by a test driver for a month in the form of a pilot project, with results evaluated in terms of geographical coverage and with the development of a statistical event profile detailing events of simulation value. The data collected contains over 6 million data points, and over 7.45hrs of driving. In evaluating MVP performance, the You Only Look Once (YOLO) multi-object detector and Markov Decision Process (MDP) multi-object tracker are implemented, with results demonstrating robustness to occlusions and the capability to detect far-away pedestrians and vehicles. With this data collection system functional, and the data from the pilot project experiment, a powertrain simulation environment for University of Waterloo Alternative Fuels Team (UWAFT) is developed. Given the Advanced Vehicle Technology Competition (AVTC) process, it is crucial to continue to explore and design novel powertrain configurations in an environment which is conducive to flexible configuration and with acceptable ease-of-use. Of the environments available, Simscape is selected and a novel Metal-Air Extended Range Electric Vehicle (MA-EREV) powertrain model is developed as a validation of the simulation tool. Upon validating simulated VTS against existing work, results are consistent excluding a 15% reduction in estimated range and a 41% decrease in 50-70 mph acceleration time. To provide an example of the data-driven approach, a winter-driving scenario where the pilot project driver demonstrated slipping is imported as a drive cycle in the MA-EREV model and simulated in an experiment. In analyzing results traction performance of the MA-EREV is evaluated. The MA-EREV weighs 677kg more than the pilot project vehicle, and has increased starting torque due to electrification. In analyzing the results of this scenario replication, the longitudinal slip on the tires reached a maximum of 41% slip (94% of available traction) during stopping and 84% slip (55% of available traction) during acceleration from stop, with more slipping overall during acceleration than stopping. This result indicates that the MA-EREV may need additional traction considerations for safe performance in winter conditions.
Cite this version of the work
Paul McInnis (2017). Intelligent Vehicle Development through Scalable Data Collection Processes and Simulation. UWSpace. http://hdl.handle.net/10012/12230