Research Topics
Fault-tolerant control
Investigator: Prof. R J Patton
A fault-tolerant control system
is designed to retain some portion of its control integrity in
the event of a specified set of possible component faults or
large changes in the system operating conditions that resemble
these faults (Patton, 1997). A component fault will refer to a
change in the operating behaviour of a component such that the
new behaviour differs significantly from what is defined as
nominal behaviour for that component. Common examples of such
faults include bias errors in the output of a sensing device
and loss of function for an actuating device. Fault-tolerant
control involves automatically detecting and identifying the
failed components and then reconfiguring the control law
on-line in response to these decisions. A fault-tolerant
control system must have the ability to adjust off-nominal
behaviour, which occurs during sensor, actuator, or other
component faults. The fault-tolerant control system consists
of three primary parts: a controller, a fault diagnosis
scheme, and a control law reconfiguration mechanism. Key
challenges in the fault-tolerant control problem are to
design: (a) a sufficiently robust controller which is
reconfigurable, (b) a robust fault diagnosis scheme and (c) a
suitable reconfiguration mechanism. A fault-tolerant
controller can engage suitably fixed or varied structure to
perform its fault-tolerant functions. If the structure and/or
parameters of a controller are adjusted on-line according to
the real-time measurement of fault effects, the controller is
called active fault-tolerant controller.
In this topic both robust
performance and robust stability with respect to modelling
uncertainty are considered in robust fault-tolerant control
systems design. The simultaneous consideration of modelling
uncertainty and performances introduce a rank constraint in
the Linear Matrix Inequality (LMI) formulation leading to a
non-convex problem. The numerical tractability must be
considered and constrained optimisation methods are used to
build up suitable LMI system synthesis, assuming both fault
effect factors and uncertainties can be of
linear-fractional-transformation (LFT) parameter
dependence.
The research in this topic will
lead to a more general framework for fault-tolerant control,
based on the integartion of robust control and robust fault
detection and isolation (FDI) principles.
Dynamic systems, state estimation and applications
Investigator: Dr. M Hou
Observers may be implemented electronically, mechanically, hydraulically, and etc., or in most cases digitally on a microcomputer or a chip nowadays. Sometimes observers are called 'soft' or virtual sensors since no hardware sensors are used directly to measure internal physical variables of interest. Very often this brings economic, environmental and technical advantages.
The initial interest in this area focused on issues of input observability and input reconstruction, designing observers for linear dynamic systems with unknown inputs and for fault diagnosis in technical processes. Later the research moved to the development of observers for complex non-linear dynamic systems. The best conference paper prize was awarded for the work (with Professor A C Pugh, Loughborough University) on non-linear observer designs by UKACC at the Conference Control'98. Non-linear observers have been designed for systems represented by recurrent neural networks, and for a laboratory 3-tank water system.
Recent research is about adaptive estimation of multiple sinusoids which are as a directly measured signal or as disturbance acting on a dynamic system. This line of research has great potential in various applications in diverse disciplines. On-going studies include applications of the adaptive estimation in speech processing and clinical hearing tests of infants.
As for general system theory, test conditions for external equivalence of behavioural systems and generalised transfer functions of generalised systems have been proposed. Previous research areas included DC and AC motor control, numerical calculation of optimal control and multi-target tracking by using Kalman filters.
Publications:
-
M Hou, Estimation of sinusoidal frequencies and amplitudes using adaptive identifier and observer, IEEE Trans. Automat. Contr., 52(3), 493-499, 2007.
-
M Hou, Frequency and amplitude estimator of a sinusoid, IEEE Trans. Automat. Contr., 855-858, 2005.
-
M Hou, Impulsive-mode controllability and elimination of descriptor systems, IEEE Trans. Automat. Contr., 1723-1727, 2004.
-
M Hou, Y S Xiong and R J Patton, Observing a three-tank system, IEEE Trans. Contr. Sys. Tech., 1350-1355, 2004.
-
M Hou, P Zitek and R J Patton, An observer design for linear time-delay systems, IEEE Trans. Automat. Contr., 121-125, 2002.
-
M Hou, A C Pugh and G E Hayton, A test for behavioral equivalence, IEEE Trans. Automat. Contr., 2177-2182, 2000.
-
M Hou, K Busawon, and M Saif, Observer design based on triangular form generated by injective map, IEEE Trans. Automat. Contr., 1350-1355, 2000.
-
M Hou, A C Pugh and G E Hayton, General solution to systems in polynomial matrix form, Int. J. Contr., 733-743, 2000.
Identification, Modelling &
Canonical Structures for Neuro-Fuzzy Networks
Investigators: Dr. C
Kambhampati and Prof. R J Patton, Faisel J Uppal
This topic involves two main
themes:
(1) Control of complex
systems. A pair of hands lifting an object, a human
walking, remote surgery are examples of multi-robots
co-operating to perform a given task. These systems consist of
a number of interconnected sub-systems and the mathematical
models are complex and multi-dimensional. It is often
difficult to design control systems, so that these robots can
perform their task efficiently and in a stable manner, using
large and complex models. However, the human motor system is
able to accomplish this task very efficiently, in a
decentralised fashion. The topic focuses on various aspects of
human motor control, and involves training decentralised
neuro-fuzzy models for these systems. The main aims are to
develop (a) a decentralised framework for training
single-input-single-output Neuro-Fuzzy Models for
multi-input-multi-output systems (b) design efficient training
algorithms.
(2) Fault detection and
isolation (FDI) A range of new challenges has
developed in all spheres of industry with an increasing demand
for reliability, cost-effectiveness, safety and environmental
protection. To meet these demands there has been a growing
interest in monitoring methods including fault detection and
diagnosis and system reconfiguration, based on redundancy
principles. Prompt detection and diagnosis of process
malfunctions are strategically important due to economic and
environmental demands required for companies to remain
competitive in world markets. This topic focuses on recent
approaches to FDI for non-linear dynamic systems using
model-based and neuro-fuzzy modelling methods, the focus being
on combined data-driven and model-based FDI within the context
of an industrial application study. The model-based methods,
little known twenty or thirty years ago, have recently found
increasing favour for a great variety of industrial
applications. In this study, the use of neuro-fuzzy method is
considered an important extension to the model-based approach
for residual generation. When quantitative models are not
readily available, a correctly trained neuro-fuzzy model can
be used as a non-linear dynamic model of the system.
Neuro-fuzzy rule-based inference methods enhance the FDI
diagnostic reasoning capabilities. In this topic a general
framework for FDI methods is developed and effective and
reliable neuro-fuzzy based FDI methods are implemented to
solve complex FDI problems.
Comparison of Fuzzy Modelling
Approaches
Investigators: Professor
R J Patton, Dr Letitia Mirea and Faisel J Uppal
This topic comprises a study of
the comparsion and integration of modelling approaches known
as Tensor Product Transformation (TPT) and Identification
Package (IP), respectively. The common primary goal of these
modelling approaches is to represent a dynamic system by the
fuzzy inference modelling strategy introduced by Takagi and
Sugeno. TSK. This startegy has exponential computational
complexity with approximation accuracy. Furthermore, in
general the TSK fuzzy model is nowhere dense in the modelling
space if the number of rules is bounded. This implies that in
pursuit of good modelling accuracy exponential explosion of
the TSK fuzzy modelling must be faced. The TP and IP
approaches have different properties in various implementation
scenarios. The main task is to find an optimal trade-off
between the modelling accuracy and processing complexity. This
plays an important role in many complex real-time
applications. The processing time in FDI for example can be
crucial in preventing and avoiding system shutdown, breakdown
and even catastrophes. The topic also includes a study of how
effectively the TPT and the IP methods can be connected to
Parallel Distributed Compensation (PDC) methods as all the
TPT, IP and PDC designs are executed numerically by computer
without analytic and heuristic derivations. If an immediate
link between these can be determined then the design of
self-developing and self-learning systems can be achieved.
This in turn will play an important role of the development of
autonomous agents for FDI and fault-tolerant control.
Fault-Tolerance and Coordination in Networked Control Systems (NCS)
Investigators: Professor R J Patton, Supat Klinkhieo, Dr Cahit Perkgoz & Changqing Lin
The NCS concept began to attract the attention of academic researchers in the 1990s due to amazing advantages of flexibility, reconfigurability, etc. A considerable volume of theoretical tools have emerged which now extend to encompass FTC approaches to NCS, including the use of two-level hierarchical and distributed control (Patton et al, 2007). Traditionally, large inter-connected systems have been viewed within a distributed, over-lapping or large-scale systems framework (Singh et al, 1978). A number of tools for Control, FTC and FDI have been developed that can be applied to NCS if due care is given to the complex distributed system structure in terms of inter-connections, over-lapping decompositions and redundancy (Singh et al, 1983; Chen & Stankovic, 2005). The NCS problem can be partitioned into a distributed control design, based on a “physical” network and a communications network involving communication delays and bandwidth limitations etc. The large-scale systems theory of the 1970s was based only on the physical network, whereas in the complete NCS problem the communications network gives rise to additional complexity in terms of network constraints (delays and bandwidth limitations, etc). This research uses the distributed control systems concept considering first of all the physical network (i.e. assuming a network of infinite bandwidth), based on a suitable architecture for achieving reliable FTC in NCS, under autonomous system operation (Patton et al, 2007). The approach requires reliable local FDI, at subsystem level. Faults within subsystems must be detected and isolated reliably in the presence of subsystem interactions. Robustness in FDI must be addressed whenever a model-based method is used. For the complex and inter-connected NCS this is no exception and a robust FDI strategy must be based on the principle of diagnosis within a given subsystem. The interactions between subsystems can be considered as “unknown inputs”. The coupling effects can be de-coupled or minimized using a variety of recently developed robust FDI methods, based on H∞, Linear Parameter-Varying (LPV) systems (Bokor & Balas, 2004; Casavola et al, 2007, Henry et al, 2008) or the unknown input observer (applied in the context of robust FDI design) (Chen & Patton, 1996, 1999; Pertew et al, 2005). The determination of the network subsystems and an in-depth description of the NCS architecture and strategy for autonomous control in an FTC context has been described in (Patton et al, 2007).
Research on this subject is focused on methods for:
- Fault-tolerant Control architectures for distributed subsystems
- Autonomous Coordination and Control in NCS
- Robust methods for FDI within subsystems and at higher levels.
- Robust fault identification/estimation for NCS problems.
- Re-configuration and redundancy issues in networked systems.
References:
-
Bokor J & Balas G, (2004), Detection filter design for LPV systems–a geometric approach, Automatica, 40(3): 511-518.
-
Casavola A, Famularo D, Franze G & Sorbara M, (2007), A fault detection filter design method for linear parameter-varying systems, Proc. IMechE Part I: J. Systems and Control Engineering, vol. 221, (6), 865-874.
-
Chen J & Patton R J, (1996), Optimal filtering and robust fault-diagnosis of stochastic systems with unknown disturbances, IEE Proc.-D: CTA, 143 (1): 31-36.
-
Chen J & Patton R J, (1999), Robust Model Based Fault Diagnosis for Dynamic Systems, Kluwer Academic ISBN 0-7923-841-3.
-
Chen Xue-Bo & Stankovic S S, (2005), Decomposition and decentralised control of systems with multi-over-lapping structure, Automatica, 41: 1765-1772.
-
Chen Xue-Bo & Stankovic S S, (2005), A Method for Designing Fault Diagnosis Filters for LPV Polytopic Systems, J. of Control Science and Engineering, 2008.
-
Patton R J, Kamphampati C, Casavola A, Zhang P, Ding S & Sauter D, (2007), A Generic Strategy for Fault-tolerance in Control Systems Distributed over a Network, Eur. J. Control, 13(2-3), 280-296.
-
Pertew A M, Marquez H J & Zhao Q, (2005), H? synthesis of unknown input observers for nonlinear Lipschitz systems, Int. J. Control, 78 (15): 1155-1165.
-
Singh M G, Hassan M F, Chen Y L, Li D S & Pan Q R, (1983), New approach to failure detection in large-scale systems, IEE Proc., 130 (5): 243-249, Pt. D, Sept.
-
Singh M G & Titli A, (1978), Systems Decomposition, Optimisation and Control, Pergamon Press, Oxford, ISBN 0-08-022150-5.
Modelling the osteocyte network and its control of the mechanotransduction and remodelling of bone
EPSRC Grant Ref. EP/E057365/1
Bone is a remarkable material which goes through an initial phase of growth and development to maturity (known as ‘modelling’), followed by a continuous cycle of repair, renewal and optimisation throughout the rest of its life (known as ‘remodelling’). Surprisingly, the processes controlling bone modelling and remodelling are still a question of debate. But what is agreed, is that these are highly non-linear dynamic activities, which adapt and change with use, with age, with disease and with other factors. This EPSRC funded research is a collaboration between C&ISE and CMET. The aim of the research is to examine how the latest control and systems engineering methodologies might be applied to this problem, with the primary objective of developing a demonstrator control model system of remodelling that can predict the variable non-linear adaptive behaviour of bone.
|
