Statistical Pattern Recognition for TWC Diagnosis

Ideas from statistical pattern recognition are applied to the problem of diagnosing degraded automotive three-way catalysts using oxygen sensors. A brief overview of statistical pattern recognition is given. A novel method of processing the oxygen sensor signals converts the time signals into ensembles of vectors. Using the tools of pattern recognition, the ensembles are described statistically, and classifiers are derived from the statistical descriptions. The resulting classifiers are tested against four different aging categories, and the performance is quantified via the operating characteristic. Techniques to achieve arbitrary levels of performance are discussed.