Simultaneous NOx and NH3 slip prediction in a SCR catalyst under real driving conditions including potential urea injection failures. (July 2022)

• Record Type: Journal Article
• Authors: Pla, Benjamin; Piqueras, Pedro; Bares, Pau; Aronis, André
• Is Part Of: International journal of engine research. Volume 23:Number 7(2022)
• Page Start: 1213
• Page End: 1225
• Publication Date: 2022-07
• Subjects: De NOx -- neural network -- Kalman filter -- vehicle emissions -- NOx sensor -- NH3 slip; Engines -- Periodicals; 629.25
• DOI: 10.1177/14680874211007646
• Languages: English
• ISSNs: 1468-0874

Abstract

To reach the emission limits imposed by governments and reduce the negative impact on the environment, the use of aftertreatment systems has become essential for internal combustion engine (ICE) based powertrains. In particular, the selective catalytic reduction (SCR) system is a widespread aftertreatment technology with high efficiency forN O x abatement which shows complex dynamics and requires urea injection as reducing agent. Current urea injection strategies usually rely on theN O x emissions feedback. This work presents a model for the on-line simultaneous prediction ofN O x andN H 3 emissions after the SCR catalyst, allowing the emissions estimation even in conditions of urea injector failure, when it is not possible to rely on the injector feedback signal. The proposed model is based on state of the art on-board after-treatment instrumentation and proposes an extended Kalman filter (EKF) to combine a data-based model and the analysis of sensor signals to provide a reliable estimation ofN O x andN H 3 slip. The proposed strategy is experimentally assessed in dynamic driving cycles, such as Worldwide harmonised Light vehicles Test Cycle (WLTC) and Standardised Random Test (RTS). The proposed method is evaluated in standard conditions (without failures) and with urea injection failures of 25% and 120% of the nominal injection amount. As a result, the prediction onN O x andN H 3 slip has been improved in all injection failure conditions, by an overall average of 47.8% and 61.8%, respectively, when compared to state-of-the-art control oriented models (physically based zero dimensional model or data-based).