(Digital Presentation)Empirical Methods for Modeling NOx Sensor Performance Errors in Combustion Applications

Abstract

Combustion technologies require gas sensors for monitoring and controlling the exhaust and to meet increasingly stringent environmental regulations. NOx compounds (NO and NO2) in diesel exhaust can lead to poor air quality and act as both pollutants and greenhouse gases. Aftertreatment strategies have been developed to mitigate NOx and usually require information about the NOx sensor output. The design and operation of aftertreatment strategies rely on an understanding of sensor behavior and deviations. However, sensor specifications provided by suppliers typically only reflect the minimum requirements necessary to meet basic application needs. Supplier information about sensor performance can therefore be too general whereas more detailed information could allow more flexibility and potentially cost savings during early development of the overall aftertreatment system and strategy.The approach and application of a methodology for predicting nominal sensor output and standard deviation is presented. The prediction is in the form of an empirical NOx sensor performance model that relies on experimental data for a specified population of sensors. Inputs include the concentration of exhaust components (e.g., NO, NO2, O2, etc.), physical state (e.g., temperature, pressure, etc.), as well as additional information about sensor condition and location (e.g., as-received brand-new sensors, tailpipe position, etc.). The output would be a dynamic real-time estimate for the NOx sensor broadcast signal and standard deviation for the defined population. Experimental data are primarily collected using a laboratory high-flow test bench that simulates exhaust conditions.We present an approach that includes understanding exhaust conditions to determine the initial factors of interest as well as reasonable ranges of values for those factors. We then investigate the correlation of the factors with the measured error of the NOx sensor broadcast value using a design of experiments (DOE) response surface methodology (RSM) with ten factors. The population was defined as nine as-received brand-new NOx sensors tested simultaneously in the laboratory high-flow test bench. Least squares fitting with a linear regression model seemed to adequately describe the correlation of mean sensor signal shifts with changes in the ten factors. Methods for validation of the model and planned next steps are discussed.