Semi-physical transient NOx modelling approach for gasoline engines

Abstract

In recent years, the emission legislation has become gradually more stringent in order to control the emission level produced by road vehicles. In response, both control of engine out emissions and enhancement of aftertreatment systems are employed to reduce the overall level of tailpipe emissions. Engine-out emissions are now commonly used as an objective for engine optimization strategies. Thus, during engine optimization process, the capability of accurately modelling emission level from engine system has become more critical than ever. However, engine emissions simulations, NOx emission in particular, has always been a challenging task as it involves complex chemical reaction that is highly integrated to combustion process.

An advanced combustion model plus a detailed chemical reactions mechanism can describe the combustion behavior and NOx formation process in an accurate manner but with the cost of computational time which is not desirable for optimization routines. On the other hand, several 1D engine simulation models have been developed with which can provide a relatively fast estimation of NOx emissions but in lack of accuracy without large amounts of training data. Faced with these challenges, the objective of this thesis is to develop a methodology that can improve the NOx prediction accuracy but without comprising computation time such that the approach remains suitable for use within optimization routines.

To achieve this objective, the proposed methodology is a “semi-physical” model structure: combustion model + NOx model + empirical tuning factor so that the physical characteristic during NOx formation can be captured. In this study, several commercially available software tools were combined in a new way to form the overall proposed approach:

1. Ricardo WAVE/WAVE RT (version 2019) was used to estimate the combustion behavior,
2. Cantera software (Version 2.4.0)integrated with a gasoline surrogate mechanism is selected to complete the NOx determination,
3. An empirical tuning factor is trained to offset any error between NOx prediction value and true experiment measurement.

The Ricardo WAVE /WAVE RT engine simulation also incorporates a phenomenological NOx emissions model which was used as a benchmark for the proposed modelling approach.

The accuracy of the purposed semi-physical methodology is first examined with steady state measurement, the validation dataset contains 408 measurements that are recorded across various engine speed, torque, spark angle and EGR operating conditions. In the structure of: Ricardo WAVE combustion model + Cantera NOx model + empirical tuning factor (47 samples based), the final NOx prediction performs a MAPE (Mean absolute percentage error) of 16.28% which is better than Ricardo WAVE RT software “in-built” NOx model (MAPE: 33.04%). When Ricardo WAVE RT is integrated to the “semi-physical” methodology and form a structure: Ricardo WAVE RT combustion model + Cantera NOx model + empirical tuning factor (47 samples based), an 82% reduction on simulation time can be achieved while maintaining a MAPE of 17.77% on NOx modelling accuracy.

The NOx prediction capability is then evaluated with dynamic measurements which contains a “load ramp” signal and a varying frequency sinusoidal signal. During both transient events, Ricardo WAVE (2019) “in-built” NOx model performs around 30% MAPE whereas the semi-physical methodology’s MAPE is only around 20%. Furthermore, the NOx prediction from semi-physical methodology demonstrates a better trend wise fitting to the experiment NOx measurement, a correlation coefficient of 0.72 and 0.76 is obtained for these two transient events obtained respectively.

Date of Award	29 Mar 2023
Original language	English
Awarding Institution	University of Bath
Supervisor	Richard Burke (Supervisor), Sam Akehurst (Supervisor) & Colin Copeland (Supervisor)