Leak Diagnosis in the Evaporative Emissions Control System Using Statistical Methods

摘要

Uncontrolled evaporative emissions contribute to air pollution and can cause public health issues, Environment Protection Agency and California Air Resources Board have evaporative emission standards to prevent gasoline vapors from freely escaping into the atmosphere. The standards require that every gasoline-powered vehicle be equipped with an Evaporative Emissions Control (EVAP) system that captures fuel vapors, and the corresponding on-board diagnostics to warn drivers when a leak is present for light- and medium-duty passenger vehicles [EPA, 2014][CARB, 2008][SAE, 2010]. Accurate small leak detection in the EVAP system is a challenging problem because of limited measurement capabilities, a wide range of operating conditions, and limited computing power on board the vehicle. In this study, we do not concern ourselves with data storage and computation limitation, and explores the possibility of using supervised classification algorithms to diagnose incipient small leaks. We show that without any physics-based knowledge of the EVAP system, a simple binary classifier can detect leaks, regardless of size. In addition, preliminary results show that a more advanced detector can offer improved performance.