Virtual NOx Sensor Using Machine Learning for SCR Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in accurately measuring NOx concentration and NH3 slip downstream from an SCR catalytic converter, particularly during critical operating states like cold starts, due to the lack of NOx sensors and unreliable sensor data, which affects emission control and monitoring.
Innovation Solution
The use of machine learning algorithms or stochastic models, such as convolutional neural networks and Gaussian process models, that incorporate state variables from the internal combustion engine and instantaneous NH3 fill level to calculate NOx concentration and NH3 slip, providing more accurate and adaptable results, even with limited data, and enabling real-time predictive control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If NOx sensors are installed downstream from the SCR catalytic converter to measure emissions, then measurement capability is improved, but system cost increases and reliability remains poor during critical operating states
Solution Approach 1:
The patent creates a virtual copy of the NOx sensor measurement capability through a machine learning model that replicates sensor functionality using alternative input data (engine state variables, urea injection data, exhaust gas temperature), thereby achieving measurement capability without relying on physical sensors that fail during cold starts
Solution Approach 2:
The patent replaces the mechanical/physical sensor-based measurement system with a data-processing-based system using machine learning algorithms that compute NOx concentration and NH3 slip from engine operating parameters, eliminating the reliability issues of physical sensors during critical operating states
2Measurement precision
If machine learning algorithms are used to calculate NOx concentration and NH3 slip, then measurement accuracy during cold starts is improved, but system complexity increases
Solution Approach 1:
The machine learning model serves multiple functions simultaneously: it calculates NOx concentration downstream of the SCR catalyst, determines NH3 slip, and adapts to different operating conditions (including cold starts), thereby achieving multiple measurement goals with a single system rather than requiring separate sensors for each parameter
Solution Approach 2:
The system uses existing data from the vehicle's own operational parameters (engine speed, torque, exhaust temperature, urea injection rate) to self-generate the emission measurements it needs, eliminating the need for external measurement devices and reducing overall system complexity despite the computational model
Data Source
AI summary
A method is provided for ascertaining a NOx concentration and an NH3 slip downstream from an SCR catalytic converter of an internal combustion engine of a vehicle. State variables of an internal combustion engine as first input variables and an updated NH3 fill level of the SCR catalytic converter as a second input variable cooperate with at least one machine learning algorithm or at least one stochastic model. The at least one machine learning algorithm or at least one stochastic model calculates the NOx concentration and the NH3 slip downstream from the SCR catalytic converter as a function of the first input variables and the second input variables and output the same as output variables.