Model Based Engine-Off Natural Vacuum Leak Detection Monitor
2017-01-1020
03/28/2017
- Features
- Event
- Content
- Engine-Off Natural Vacuum (EONV) principles based leak detection monitors are designed to determine the presence of a small leak in the fuel tank system. It was introduced to address the ever more stringent emission requirement (currently at 0.02”) for gasoline engine equipped vehicles as proposed by the Environmental Protection Agency (EPA) and California Air Resources Board (CARB) in the United States [2, 3]. Other environmental protection agencies including the ones in EU and China will be adopting similar regulations in the near future. Due to its sensitivity to known noise factors such as the ambient temperature, barometric pressure, drive pattern and parking angle, it has been historically a lower performing monitor that is susceptible to warranty cost or even voluntary recalls. The proposed new model based monitor utilizes production pressure signal and newly instrumented temperature sensors [15]. Compatible with existing strategy’s execution phases, the well-known Antoine’s Equation motivated a diagnostic feature calculation associated with the ratio between vapor pressure changes to corresponding vapor temperature changes. In addition, we developed a optimization procedure to obtain threshold(s) that maximizes detection rate (performance) and separation between healthy and faulty data (robustness). A “decision buffer region” scheme (2 thresholds setup) is applied similarly to the technique in SVM [1, 13] to boost the monitor’s performance by inhibiting decision making in the feature space associated elevated risk of mis-classification. Finally, we summarized our analysis with experimental data collected from test vehicles (MY 2012 Explorer and MY2014 Focus) including performance comparisons among various sensor configurations.
- Pages
- 8
- Citation
- Tseng, F., Makki, I., Kumar, P., Jentz, R. et al., "Model Based Engine-Off Natural Vacuum Leak Detection Monitor," SAE Technical Paper 2017-01-1020, 2017, https://doi.org/10.4271/2017-01-1020.