Research Projects
KTP
Project with GDS Instruments Ltd
Funding:
Knowledge Transfer Partnerships Programme (No. 6902).
Period:
2008-2011
V.M. Becerra and W. Holderbaum
The
objective of this project is to introduce nonlinear and adaptive
control techniques to soil testing equipment.
Investigating
the Computational Capacity of Cultured Neuronal Networks Using Maching
Learning
Funding:
EPSRC (EP/D080134/1)
Period:
2007-2010
K.
Warwick, B.J. Whalley S.J. Nasuto, V.M. Becerra
In this project neural cell cultures will
be
cultured locally in the University of Readings'
electrophysiological research laboratory allowing real-time access to
the recording and stimulation hardware via an intranet link-up.
In order to test the abilities of such cultured neural networks we
propose using them to control some of our existing mobile robots. This
is to be achieved by applying a number of machine learning and
artificial intelligence techniques in order to correctly translate
robot sensor inputs into suitable patterns of stimulation and interpret
the resulting patterns of neural activity as motor actions. In order to
measure the amount of computation the cultured "brain" is performing we
will use a surrogate (an artificial neural network that redistributes
the input signal to the output) in place of the the cultured "brain".
Both the cultured "brain" and the surrogate will be applied to various
behavioural tasks (such as obstacle avoidance and wall following) the
difference in performance between the cultured "brain" and the
surrogate will give us some measure of the processing capabilities of
cultured neural networks when used in this way.
Global
trajectory optimisation: Can we prune the search space when considering
deep space manoeuvres?
Funding: European Space Agency (Ariadna Contract No. 06/4101 )
Period: 2007
M.
Vasile, M. Ceriotti and G. Radice (for Glasgow)
V M
Becerra, S.J. Nasuto, J. D. Anderson (for Reading)
This project involved
the development of methods for the optimisation of spacecraft
trajectories involving multiple gravity assists and deep space
manoeuvres, paying particular attention to pruning the search space to
reduce the amount of computations needed to find good solutions. The
mathematical challenges introduced by going from the simple multiple
gravity assist problem to the multiple gravity assist with deep space
manouvre problem are enormous. The complexity of the problem increases
considerably due to the necessarily added dimensions and to the larger
number of local minima introduced.
The
project final report is available from this link.
KTP
Project with @UK plc
Funding:
Knowledge Transfer Partnerships Programme (No. 1554).
Period:
2006-2009
S.J.
Nasuto and V.M. Becerra
The
objective of this project is to develop a a product aware page ranking
system as part of an integrated internet search engine for
e-procurement purposes.
A
search for invariant relative satellite motion
Funding: European Space Agency (Ariadna Contract No. AO04/4104 )
Period: March-July 2005
V M Becerra, S.J. Nasuto, V.F. Ruiz, and J. Biggs
This
project employed a Hamiltonian formulation of relative satellite motion
and a variant of Newton's method to
locate periodic or quasi-periodic relative satellite motion. The
perturbations considered in the model included nonlinear gravitational
effects, the oblateness of the Earth (J2 effect) and eccentricity of
the reference orbit.
Advantages of using Newton's method includes simplicity of
implementation,
repeatability of solutions due to its non-random nature, and fast
convergence. In order to evaluate the effect of the quality of the
model used to generate the
periodic reference trajectory, a study involving closed loop control of
a
simulated chief/deputy satellite formation was performed.
The
project final report is available from this link.
Global
optimisation tools for mission analysis and design
Funding: European Space Agency (Ariadna Contract No. 18138/4101)
Period: April-October 2004
V M Becerra, S.J. Nasuto, J.M. Bishop and D. Myatt
Mission design involves consideration of many factors, one of the
most important being the design of optimal trajectories. Traditionally
this task has been accomplished using gradient methods, optimal control
theory or mathematical tools specifically dedicated to each particular
problem. These approaches can be generally classified as local
optimisation methods. Due to their nature, however, most mission
trajectory design problems exhibit local minima.
The
importance of having an effective and efficient global
optimisation approach is emerging also in the space field with studies
on procedure and optimisation methods to procure a solution or even
just a first guess solution to complex problems.
The
proposed research will investigate problems associated with
optimal mission design, leading to appropriate problem taxonomy
characterised by the properties of the respective search spaces. This
taxonomy will lead subsequently to a principled evaluation of a
selection of optimisation algorithms.
The
project final report is available from this link.
Predictive control of
nonlinear systems using
feedback linearisation based on dynamic neural networks.
Research
student: Mrs. J. Deng
Funding:
EPSRC
Period:
April 2002 - March 2005
Predictive
control is a technique that uses a model of the controlled system to
calculate the control action based on optimal predictions. Predictive
control has
been very successful due to its ability to accommodate constraints and
multivariable systems and also due to the use of empirical dynamic
models. Most conventional predictive controllers use linear
empirical models. Feedback linearisation is a well known nonlinear
control technique that consists of transforming a nonlinear system into
a linear system by means of state feedback and nonlinear
transformations. Dynamic neural networks are mathematical structures
that can be described by means of differential or difference equations
and have the ability, given appropriate training, to approximate
various types of nonlinear
dynamic systems. They are particularly suitable for efficiently
modelling
dynamic systems with multiple inputs and multiple outputs. The research
will
integrate predictive control with feedback linearisation based on
dynamic
neural networks, with particular emphasis on multiple-input
multiple-output systems, and focusing on the handling of constraints on
the input variables (which become nonlinear and state dependent with
the nonlinear transformations used), the structure of the empirical
models, the training of the dynamic neural networks, and the effects of
the nonlinear transformation used on control
performance. Real time experiments will be carried out on a laboratory
scale
crane system that is able to position a payload in three dimensions and
comparisons
will be made with conventional control schemes.
The final project report to the EPSRC is available here.
Investigation into the
use of optic flow and CMAC networks for robot balance
Research student:
Miss G.
Martinez-Hernandez
Funding: Mexican
Government
Period: May 2001 -
July 2005
The motivation for this work arised from research into
the psychology of human balance. It has been established that the human
sense of balance is an combination of visual, mechanical and vestibular
sensory information. This work has involved virtual reality
simulations and real time experiments of balancing an inverted pendulum
using optic flow and CMAC neural networks. Optic flow is the movement
of elements of the scene over time which result from changes in the
position of the vantage point. It has been demonstrated that
information derived from optic flow, when augmented by limited
information about the mechanical state of the system, is sufficient to
maintain the balance of an inverted pendulum.
The PhD thesis of Miss G. Martinez-Hernandez is available here
in PDF format with her permission.
Automatic tuning of PID
controllers using
neural networks.
Reseach
student: Mr. K. Pirabakaran
Funding:
Research Endownment Trust Fund (The University of Reading)
Period:
October 2000 - September 2003
The
use of neural networks is being investigated to automatically tune the
parameters of PID controllers. Automatic tuning consists of a number of
steps. Typically a user sends a signal to start the procedure. As a
consequence, the plant is perturbed for a certain period of time, the
resulting transient signals are analysed, the controller parameters are
calculated, and finally the
controller returns to normal operation. Conventional methods for
automatic
tuning, such as relay feedback, sometimes fail to provide a good tuning
due
to particular characteristics of the process (such as nonlinearities,
high
order dynamics or the precense of disturbances), and due to the
approximations
inherent to the methods used (for instance, neglecting higher harmonics
in
describing function analysis). Thus, although automatic tuning is a
useful
feature found in most commercial PID controllers nowadays, improvements
are
possible. The integration of conventional controllers with neural
networks
has immediate practical applications and is therefore relevant to
industry.
KTP
Project with Applied Weighing International Ltd.
Funding: Knowledge Transfer
Partnership Programme
Period: 2004-2006
Academic Leader: Prof. P.M.
Sharkey
Academic supervisor: Dr. V.
M. Becerra
Partnership
objective: To develop improved high speed weighing
systems for conveyor and vehicle weighing applications; to increase
flexibility of existing product and to move into new markets.
Knowledge
Transfer Partnership with Merris Development Engineers Ltd.
Funding: Knowledge Transfer
Partnership Programme
Period: 2005-2007
Academic Leader: Dr. V.M.
Becerra
Academic supervisor: Prof.
P.M. Sharkey
Partnership objective: To
develop an innovative and effective automated paint tinting system with
low maintenance requirements to expand the current product range.
Development and
integration of artificial
intelligence techniques for real time control.
Joint
research project with: Dr. J. Calado (ISEL, Lisbon, Portugal)
Funding:
The British Council
Period: April
2001-March 2002
This
project concerns the development of real time control algorithms based
on artificial intelligence techniques including neural networks,
genetic algorithms and fuzzy logic. The developments are being tested
on a pilot scale process that includes level, temperature and pressure
control loops. The project
will include an investigation into fault diagnosis methods for control
valves. They project will involve interfacing the pilot process to the
Internet to enable remote monitoring and control.
|
