Research Application of DFSS: Study of the Impact of Accelerated Aging and Recovery on Low-Rh Three-Way Catalyst Activity for Catalyst Monitoring 2010-01-0702

Robust on-board diagnosis of emission catalyst performance requires the development of artificially damaged "threshold" catalysts that accurately mimic the performance of damaged catalysts in customer use. The threshold catalysts are used by emissions calibrators to determine fore-aft exhaust oxygen sensor responses that indicate catalyst failure. Rather than rely on traditional trial-and-error processes to generate threshold catalysts, we have used a DFSS (Design For Six-Sigma) approach that explores, at a research level, the relationship between oxygen storage capacity (OSC) of the catalyst (i.e., the fundamental property dictating the response of the aft oxygen sensor) and key process input variables: high-temperature exposure, phosphorus poisoning, and catalyst "recovery." Generating threshold catalysts which can demonstrate NO_x (nitrogen oxides) performance monitor capability is challenging due to the low levels of rhodium (Rh) yet high levels of oxygen storage materials in current three-way catalysts (TWC). Laboratory aging and recovery conditions were constructed using Design of Experiments (DOE) to determine their impact on catalyst activity and OSC. Subsequent response surface designs were generated and run in order to detect response curvature with the various factors. Transfer functions characterizing NO_x conversion efficiency and OSC as functions of aging temperature (T), phosphorus poison (P), and recovery levels had R-squared (adjusted) values of 89% and 97%, respectively. These equations were used in conjunction with a multi-component optimizer to identify catalyst aging and recovery conditions that could maximize catalyst NO_x performance (to 72%) while minimizing OSC (at 33 μmol-O/gm TWC). Research level understanding, combined with DFSS tools, offers a powerful approach for developing robust components and processes that function effectively over a much broader range of input factors and combinations of factors than typical of traditional trial-and-error methods where we evaluate one factor at a time.