Riassunto analitico
In the past decades, growing concerns about air pollution led to extensive
studies addressing the risks for human health. In order to stop this worrying trend,
the governments started posing new regulations over pollutant emissions. As a
consequence, automotive manufacturers have been forced to invest more and more
resources in regulations-compliant vehicles, creating the need for advanced technological
solutions.
This thesis, conducted within the gasoline engine calibration department of AVLItaly
Srl, aims to introduce a methodology for applying AI to build a virtual oxygen
sensor model. This virtual sensor is then introduced as the “plant” of a softwarein-
the-loop system, useful to improve and validate engine calibrations, with the
objective of complying with EURO 7 regulations . More specifically, the strategies
will be applied to ’high-load’ driving conditions, where emission control was less
stringent under previous regulations.
In the first part, a virtual ECU control model was developed, starting from data
acquisition of the real one.
In the second part, several machine learning algorithms were investigated to predict
the dynamic behavior of the λ before the three-way catalyst. These algorithms were
trained and tested using a dataset of “on the road” ramp tests, collected from a
vehicle equipped with a high performance V6 engine with double injection system
and passive pre-chamber. The prediction on the λ value is assessed with a Mean
Absolute Percentage Error (MAPE) being 0,6% and a correlation (R2) of 0,91 on
the testing dataset, while a 0,4% MAPE and a 0,99 R2 on training dataset.
In the end, an effective new calibration to improve drivability will be developed
virtually, and tested directly on-board.
|
Abstract
In the past decades, growing concerns about air pollution led to extensive
studies addressing the risks for human health. In order to stop this worrying trend,
the governments started posing new regulations over pollutant emissions. As a
consequence, automotive manufacturers have been forced to invest more and more
resources in regulations-compliant vehicles, creating the need for advanced technological
solutions.
This thesis, conducted within the gasoline engine calibration department of AVLItaly
Srl, aims to introduce a methodology for applying AI to build a virtual oxygen
sensor model. This virtual sensor is then introduced as the “plant” of a softwarein-
the-loop system, useful to improve and validate engine calibrations, with the
objective of complying with EURO 7 regulations . More specifically, the strategies
will be applied to ’high-load’ driving conditions, where emission control was less
stringent under previous regulations.
In the first part, a virtual ECU control model was developed, starting from data
acquisition of the real one.
In the second part, several machine learning algorithms were investigated to predict
the dynamic behavior of the λ before the three-way catalyst. These algorithms were
trained and tested using a dataset of “on the road” ramp tests, collected from a
vehicle equipped with a high performance V6 engine with double injection system
and passive pre-chamber. The prediction on the λ value is assessed with a Mean
Absolute Percentage Error (MAPE) being 0,6% and a correlation (R2) of 0,91 on
the testing dataset, while a 0,4% MAPE and a 0,99 R2 on training dataset.
In the end, an effective new calibration to improve drivability will be developed
virtually, and tested directly on-board.
|
