Site Loader

In this chapter
we shall discuss in detail the operations of an Engine Management System,
conventional methods used, their shortcoming and scope and methods for
improvement.

 

Engine
management system is a broad term for many similar components used in modern
day automobiles, namely ECU, DME, VECS, etc. are all examples of EMS Systems.
They are usually the central computing units in a car and govern engine tuning,
input and output signals as well as environmental input signals. In today’s modern
age of advanced engines with carious electrically controlled components and
precision metrics, the advancement in EMS plays an even more vital role in
increasing the efficiency of classical IC engines. By optimizing various
aspects of how input and output signals are taken and generates as well as how
they are processed, their speed as well as accuracy, the performance of said
engines can be greatly improved, without having to invest in additional
hardware.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

 

Electronic
engine controls was introduced in the 1980’s, significant improvements in fuel
economy and emissions reductions have been achieved by the development of this
area. Look-up table is the dominant method used by ECU in automotive industry.
The reasons for this are the simplicity, low computation load, and reli-
ability. However, the difficulty on the implementation of look-up tables is
that it will take a huge effort and labors for engine calibration to obtain
control data to fill the tables. Besides, the increased number of control and
sensor signals, the nonlinear characteristics, time dependents of engine
processes, and unavoidable time delays all make the calibration process more
complicated. In modern ECUs, around 9000 parameters and more than 600 control
and diagnostic functions are implemented3.

 

The ECUs in next
generation are expected to have the ability to overcome these difficulties
brought by the increasing requirement on engine performance and the limitation
of look-up table method, and improve the control performance by adopting
advanced control technologies. Moreover, online adaptation is another
attractive characteristic for future engine control system, which means that
the control system can alter settings to take account of changes in the
condition of components, such as engine wear, components replacement, air
leakage in manifold, etc. With the increasing computational powers that are
becoming available for engine control units, more advanced methods for
modelling and control can be implemented to increase the engine efficiency and
reduce the fuel consumption

 

4.1 Neural Network Models

Neural networks
are inspired by biological brain connectionism. From the past research in this
area, neural networks have proved to provide very powerful solutions to a large
variety of engineering problems ranging from modelling over prediction up to
classification. They can not only provide a simple model structure for
nonlinear system, but also capture the nonlinearity and dynamics with
satisfactory accuracy. However, one of the major obstacles to engineering application
of neural networks is the heavy computational burden for their online training.
With the evolution of electronics, reconfigurable device such as field programmable
gate arrays (FPGA) have make feasible online training for the parameters of the
neural network. Therefore, NN based modelling and control have been a very
active research area in recent years.

A lot of
researches on engine modelling using neural network has been done over past
years, most of them are based on feed-forward model including the multi-layer
perception(MLP) network, the radial basis function (RBF) network, the pseudo-linear
radial basis function (PLRBF) network678. In these researches, neural
networks have been used as a modelling method for air path dynamics, emission
gas recirculation (EGR) and even AFR dynamics. The results shows that NN
modelling is relatively easy and inexpensive. As some research have proved that
since time series data may have autocorrelation or time dependence, the
recurrent neural network models which take advantage of time dependence may be
useful. In other words, feedback allows recurrent networks to achieve better
predictions than can be made with a feed-forward network with a finite number
of inputs. Therefore, recurrent neural network has come to be very popular as
an engine modelling method recently, especially for the modelling of AFR
dynamics. The recurrent neural network is trained using back propagation with a
history stack learning algorithm. The advantage of RNN based engine models is
that they can make a relative accurate prediction on air fuel ratio with
limited training data set.

Post Author: admin

x

Hi!
I'm Myrtle!

Would you like to get a custom essay? How about receiving a customized one?

Check it out