Abstract
Following the Volkswagen Dieselgate scandal in 2015 where the emission tests were ‘defeated’, it was revealed that for several manufacturers there was a divergence between real world emissions and emissions emitted in a lab of over five times the emission limits, even when the vehicles were considered legal. Soon after, a new emission test was announced which aimed to bridge the gap between road and laboratory emissions by performing on road testing. This was named the real driving emissions (RDE) test, and it imposed a challenge to the automotive industry and especially calibration engineers to make sure that their vehicles are both ‘legal’ and ‘cleaner’ in terms of emissions which was not legally the case before. This was due to the test being non-reproducible because of the different cycles and ambient conditions the car would experience.Different methods were investigated by vehicle manufacturers to verify a vehicle’s compliancy over the full RDE conditions including experimental-based, simulation-based and model supported methods. It was understood that the experimental-based methods were reliable but too costly to perform and could hardly cover the full RDE testing range conditions. The simulation-based methods could cover the full RDE range but required a lot of data to train the models and was considered to be less reliable. The model supported methods required less data and was placed somewhere in the middle of the other two methods. This is because it used the real vehicle parts for testing, but simulations could be used to cover unexplored conditions. In all methods, the effect ambient conditions and driver style have on the emissions has not received much attention in the literature and is potentially an important factor affecting the emission results since the alteration of a single variable of ambient conditions could alter the WLTC emissions by 10%. In addition, a methodology where a limited amount of data was used and a methodology in designing a ‘worst-case’ RDE cycle was not investigated enough.
For this thesis and for the RDE+ project of Horiba-MIRA, RDE tests were performed in the UK, Austria, and Spain with vehicles of different powertrains to get some hand-on experience and to collect driving data for different environments and driver styles. By analysing the data, different trends were identified between ambient conditions, driver styles, and the different emission gases. Higher emissions were noted with an increase in altitude, a lower ambient temperature, or a more aggressive driver behavior with a maximum difference of 100%. However, it was still uncertain if the vehicles would pass the RDE test over the whole boundary conditions. Therefore, it was considered important to get a better understanding on how the actual load conditions change in different environments. This in effect would provide a clearer relationship between the road load conditions and the real driving emissions. Hence the coastdown test was analysed and was used as a basis to develop a method to predict how the road load coefficients would be affected in different environments. In addition, tyre tests were carried out on the chassis dynamometer using wheel torque transducers to check the effect tyre pressure has on emissions and rolling resistance. It was verified that a 5-psi increase would lead to an 11% decrease in Rolling resistance and a 6% longer coastdown test.
Following the experimental studies, cycle generating technologies were investigated to produce a method in generating valid, representative, and random RDE cycles using a limited amount of data by focusing on the driver behavior characteristics of the cycle. By using the Markov Chain method and the use of several proportional integral derivative (PID) controllers, RDE cycles with specific trip dynamics were generated with 80% of the tests yielding results within 5% of the targeted driver aggressiveness while only 1 test out of 24 having a difference larger than 10%. This variability enabled the expansion of the vehicle’s operating region by 24% which in effect expanded the RDE tested boundaries and emissions as well.
Emission modelling techniques were then researched and applied to the cycle generating methodology including Local Model Networks and the Volterra Series models. It was identified that to obtain a good model fit several variables were required for the training process of the models which required extra models including an engine and turbocharger model and better-quality data. Using a limited amount of training variables, CO2 was the only emission gas which exhibited a good fit and was used for the next stages of the thesis.
The research and developed methodologies throughout the thesis complete a process of identifying if a vehicle is legal within the RDE rules. This is done by using a limited amount of data to both produce emission models and generate several artificial RDE cycles. The emission models are then applied to the generated cycles while the highest emitting generated cycles are transformed into even ‘worse’ cycles in terms of emissions by the alteration of the road loads to simulate different ambient conditions and the adjustment of the gear shifting strategy. This cycle can be then used inside a laboratory for calibration testing. The specific process would reduce the resources required in ensuring that a vehicle is compliant with the new RDE legislation while also reducing the carbon footprint of the process overall.
| Date of Award | 22 Jun 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Sponsors | Horiba Mira Ltd |
| Supervisor | Richard Burke (Supervisor), Edward Chappell (Supervisor) & Chris Brace (Supervisor) |