Analysis and Choice of Input Candidates for a Virtual NOx Sensor by a Mutual Information Approach
Sprache des Titels:
SAE 2016 World Congress and Exhibition
Abatement and control of emissions from passenger car combustion engines have been in the focus for a long time. Nevertheless, to address upcoming real-world driving emission targets, knowledge of current engine emissions is crucial. Still, adequate sensors for transient emissions are seldom available in production engines. One way to target this issue is by applying virtual sensors which utilize available sensor information in an engine control unit (ECU) and provide estimates of the not measured emissions. For real-world application it is important that the virtual sensor has low complexity and works under varying conditions. Naturally, the choice of suitable inputs from all available candidates will have a strong impact on these factors. In this work a method to set up virtual sensors by means of design of experiments (DOE) and iterative identification of polynomial models is augmented with a novel input candidate selection strategy. Therefore, a systematic approach to analyze and determine possible input candidates based on mutual information (MI) analysis is introduced. As application example a virtual NOx sensor for a passenger car Diesel engine is presented. For input selection, sensor setup and validation, experiments on a hardware-in-the-loop (HIL) test bench under varying operating conditions are used. The validation of the sensor, performed during transient cycles varying environmental conditions, led to promising results.