| On January 24, 2007, Environmental Protection Agency (EPA) signed a proposed regulation to require an On-Board Diagnostics (OBD) system to monitor malfunctions in large highway diesel and gasoline trucks over 14,000 lbs from 2010 onwards. A sensor validation scheme, known as SFDIA which has been used in aircraft controls were modified and developed to not only monitor sensor malfunctions but also isolate the faulty sensor signals. The isolated signals would then be accommodated using an estimated value obtained from predictive sensor models. The models for this study were developed using a statistical modeling technique known as Multivariate Adaptive Regression Splines (MARS). Using 14 engine parameters on a Volvo MD11 diesel engine, this study demonstrated the feasibility of creating models of a target parameter based on the input of the other 13 parameters. Additionally, this study also investigated the feasibility of a virtual NOx sensor methodology, based on 14 parameters available on a typical diesel engine. The data collected in this study was obtained by conducting tests using simulated on-road cycles on a dynamometer in the West Virginia University (WVU) Engine and Emissions Research Laboratory (EERL).; The SFDIA scheme, which was written in Simulink, was tested by injecting 14 different types of artificial failures into the system during simulation. In each failure, the SFDIA was able to detect, isolate, and accommodate the faults in an accurate and timely fashion. Additionally, the virtual NOx sensor proved highly successful as it was able to predict the NOx emissions with an average root mean squared error of less than 4%. |
