Three-Way Catalyst Diagnostics and Prognostics Based on Support Vector Machines

2017-01-0975

03/28/2017

Features
Event
WCX™ 17: SAE World Congress Experience
Authors Abstract
Content
A three-way catalytic converter (TWC) is an emissions control device, used to treat the exhaust gases in a gasoline engine. The conversion efficiency of the catalyst, however, drops with age or customer usage and needs to be monitored on-line to meet the on board diagnostics (OBD II) regulations. In this work, a non-intrusive catalyst monitor is developed to diagnose the track the remaining useful life of the catalyst based on measured in-vehicle signals. Using air mass and the air-fuel ratio (A/F) at the front (upstream) and rear (downstream) of the catalyst, the catalyst oxygen storage capacity is estimated. The catalyst capacity and operating exhaust temperature are used as an input features for developing a Support Vector Machine (SVM) algorithm based classifier to identify a threshold catalyst. In addition, the distance of the data points in hyperspace from the calibrated threshold plane is used to compute the remaining useful life left. To further improve the monitor robustness and reduce the number of support vectors, clustering techniques are proposed, implemented and evaluated. The model was tested and validated on multiple vehicles with differently configured catalyst systems and was found to be robust and accurate for on-board implementation.
In addition, this approach for catalyst monitor is generic and has been successfully extended for other vehicle diagnostics applications such as universal exhaust gas oxygen (UEGO) sensor diagnostics and vehicle rollover detection.
Meta TagsDetails
DOI
https://doi.org/10.4271/2017-01-0975
Pages
7
Citation
Kumar, P., and Makki, I., "Three-Way Catalyst Diagnostics and Prognostics Based on Support Vector Machines," SAE Technical Paper 2017-01-0975, 2017, https://doi.org/10.4271/2017-01-0975.
Additional Details
Publisher
Published
Mar 28, 2017
Product Code
2017-01-0975
Content Type
Technical Paper
Language
English