[Show abstract][Hide abstract] ABSTRACT: In this paper Fuzzy Region Simulation (FRenSi), which provides solutions for fuzzy dynamical models, is applied to qualitative adaptive control. FRenSi, combines fuzzy model parameters and fuzzy initial conditions to form a fuzzy region. A simulation algorithm generates the evolution of this fuzzy region. This region is described by vertices and cubic splines linking these corner points. The vertices are integrated numerically and new cubic splines are generated after each integration step. FRenSi provides solutions for fuzzy models without ambiguity and spurious behaviour. The FRenSi method is applied successfully to Qualitative Model Reference Adaptive Control (QMRAC) to form a variable precision control system. 1 Introduction In control engineering, simulations are used to gain understanding of a physical system and its dynamic behaviour. Qualitative simulation, a branch of Qualitative Reasoning (QR), is based on the use of qualitative (less precise) but accurate models, derived f...
[Show abstract][Hide abstract] ABSTRACT: The authors consider models to be executable descriptions of the
real world, that is a model can be used to predict or analyse properties
of the system. Simulation and reasoning systems, which may be derived
from traditional or AI approaches, are used to execute these models.
Given the plethora of modelling techniques available which cope well
with certain, but not other, contexts, it is evident that there is no
`best model' covering all situations: a model is correct if it satisfies
its purpose no less and no more. The desires of the user of a modelling
system are always moderated by the availability of techniques permitting
these desires to be met. To alleviate the difficulties associated with
this requires a methodology to guide the user to the best model and
simulation technique to meet his needs. A primary requirement in the
construction of such a methodology is a comprehensive and understandable
classification of the choices inherent in the construction of a model
No preview · Article · Oct 1999 · IEE Proceedings - Control Theory and Applications
[Show abstract][Hide abstract] ABSTRACT: This paper demonstrates the use of multiple models in intelligent control systems where models are organised within a model space of three primitive modelling dimensions: precision, scope and generality. This approach generates a space of models to extend the operating range of control systems. Within this model space, the selection of the most appropriate model to use in a given situation is determined through a reasoning strategy consisting of a set of model switching rules. These are based on using the most efficient, but least general models first and then incrementally increasing the generality and scope until a satisfactory model is found. This methodology has culminated in a multi-model intelligent control system architecture that trades-off efficiency with generality, an approach apparent in human problem solving. The architecture allows learning of successful adaptations through model refinement and the subsequent direct use of refined models in similar situations in the future. Examples using models of a laboratory-scale process rig illustrates the adaptive reasoning and learning process of multi-model intelligent control systems.
No preview · Article · Apr 1999 · Artificial Intelligence in Engineering
[Show abstract][Hide abstract] ABSTRACT: Intelligent control of complex industrial processes requires that
knowledge from a variety of sources be used to maintain control over an
extended set of operating conditions. It is asserted that to maintain
the integrity of such heterogeneous knowledge, it should be encoded in
distinct models. Control then consists of selecting and executing the
most appropriate model for a given situation. Adaptive intelligent
control can then be implemented by developing switching strategies that
allow a trade-off between the various model properties. A prototype
system (MuRaLi) is presented that has multiple models based on three
primitive dimensions: precision, scope and generality. Generality is
realised through three different knowledge representation mechanisms:
procedures, rules and equations. Homogeneous control consists of
constant generality and variable precision and scope to generate the
most appropriate control action. Heterogeneous control consists of
potential variations in all three dimensions. Simple switching
strategies are investigated for both forms of control. The system has
been applied to the control of a simulated 800 MW thermal power plant.
Examples of homogeneous and heterogeneous control are given, with
experimental results of adaptation based on the proposed switching
No preview · Article · Dec 1998 · IEE Proceedings - Control Theory and Applications
[Show abstract][Hide abstract] ABSTRACT: A perspective on situated learning is presented which proposes that navigating within a space of many descriptive models overcomes limitations inherent in mono-model approaches to learning contextualized knowledge. The situationist view may be interpreted as drawing attention to neglected aspects of knowledge such as non-verbal, tacit, sub-conscious, metacognitive and affective. Although these elements have evaded adequate modelling for the purpose ofsimulatinghuman behaviour, they can be “attended to” from descriptions tosupporthuman behaviour—however, the utility of a representation depends on the kind of knowledge so described. Different viewpoints of a situation, such as the learner's and a professional's, can be described with different models, which differ in fundamental dimensions. These facilitate communication of viewpoints between learners and professional members of the community, so that though negotiation a synthesis emerges which retains critical aspects of both view points—this is learning. Other interactions between viewpoints develop other affective and metacognitive skills. The many elements of situated action and knowledge are discussed and then a methodology for supporting situated learning with multiple descriptive models is presented. A situated learning environment is present, founded on multiple models, which demonstrates that switching between models is a metalevel process for changing viewpoints and this is the basis of integrating learners into new communities of practice. Learning problem solving, developing identities and refining roles are all introduced and consolidated by an example using multiple models to developaffectiveconciliation skills. The examples given illustrate how part of a professional's knowledge can be used by a learner for one particular approach to learning. However, the representation of the professional, using this multiple models approach, would include informal knowledge from the community of practice as well as formal knowledge. This combination of different knowledge types allows a variety of learning situations to be accommodated.
No preview · Article · Dec 1998 · International Journal of Human-Computer Studies
[Show abstract][Hide abstract] ABSTRACT: This paper will not discuss the philosophy of causality in depth, but will focus on its applications to the modelling of macroeconomic systems. There has been much discussion in econometrics about the extraction of structural parameters. This is the question of ‘identification’, and there are well-established procedures for this extraction. Equations estimated in econometrics usually provide the functional relations among variables or economic indicators. However, economists are not only concerned with mathematical connections, as the causal relations between variables are also considered to be important. Qualitative models in economics are usually specified in terms of variables and their relations. Morever, data sets are not always available in qualitative models. Therefore, one possible approach for economic prediction is to apply causal reasoning methods to qualitative models. In this paper, modifications to the causal ordering concepts of Simon and Iwasaki will be proposed. A discussion on recent approaches for different types of tests for causality on stochastic models in also provided.
No preview · Article · Nov 1998 · International Journal of Systems Science
[Show abstract][Hide abstract] ABSTRACT: The qualitative model reference adaptive control (QMRAC) system
described is based on using fuzzy qualitative models to describe the
physical system and the desired performance specification. This method
allows accurate representation of the performance specification and
knowledge of the physical system as an explicit model. Based on these
explicit models an accurate solution is derived. In this work the
extension of conventional MRAC methods, i.e. the gradient method, to a
QMRAC system is discussed. The analytical properties of a QMRAC system
makes it possible to asses the system performance, e.g. stability
[Show abstract][Hide abstract] ABSTRACT: This paper proposes a new approach for utilizing qualitative simulation techniques within a model reference adaptive system for the control of ill-defined and uncertain processes, typical of the process industries. It is argued that the practical specification of performance of industrial systems is very often imprecise and multi-valued leading to non-unique (numerical) descriptions of the reference behaviour and, further, that the lack of precise knowledge of the industrial process results in inaccurate (numerical) models of the process to be controlled. This can lead to significant deterioration in performance with respect to the desired specification necessitating empirical tuning and hence the loss of analytic properties. Qualitative simulation techniques are used to model imprecise specifications and process knowledge, and hence to generate the reference behaviour without a loss of accuracy with respect to the original specifications. The discrepancy between the actual and the reference behaviour is used to adapt a conventional control algorithm such that model-following behaviour is maintained in the face of significant disturbance to the normal behaviour. Results are presented for first- and second-order models of desired specifications. The results are very encouraging, demonstrating that accurate adaptive behaviour of ill-defined systems can be obtained without the need to corrupt, or approximate, the original specifications, and without necessitating the availability of accurate, high-order, numerical models.
No preview · Article · Apr 1998 · Engineering Applications of Artificial Intelligence
[Show abstract][Hide abstract] ABSTRACT: We present a methodology for the selection of candidate generation and prediction techniques for model-based diagnostic systems (MBDS). We start by describing our taxonomy of the solution space based upon the three main functional blocks of a top-level MBDS architecture (the predictor, the candidate generator and the diagnostic strategist). We divide the corresponding problem space into user requirements and system constraints which are further subdivided into task and fault requirements, and plant and domain knowledge constraints respectively. Finally we propose a set of guidelines for selecting tools and techniques in the solution space given descriptions of diagnostic tasks in the problem space.
No preview · Article · Jan 1998 · Artificial Intelligence in Engineering
[Show abstract][Hide abstract] ABSTRACT: A novel approach to time constrained model-based fault diagnosis of ill-defined dynamic systems is presented. A fuzzy qualitative simulation algorithm is used to generate qualitative predictions of the dynamic behaviour of the faulty process. The predictions are then compared to the observed behaviour to identify the correct parameter values of the qualitative model. By comparing these identified parameter values to their known fault-free or nominal values, faults in the process can be detected and identified. Further, a control module for this qualitative parameter identification system is developed, which can trade-off aspects of the solution quality for time. In particular, the precision of the model and the completeness of the qualitative simulation can be traded-off to enable a solution to be produced within a user prescribed time constraint. Experimental results for a benchmark 3rd order dynamic system are given.
No preview · Article · Oct 1997 · Engineering Applications of Artificial Intelligence
[Show abstract][Hide abstract] ABSTRACT: The desires of the user of a modelling system are always moderated
by the availability of techniques permitting these desires to be met. To
alleviate the difficulties associated with this requires a methodology
to guide the user to the best model and simulation technique to meet his
needs. A primary requirement in the construction of such a methodology
is a comprehensive and understandable classification of the choices
inherent in the construction of a model and a categorisation of existing
techniques in the light of this classification. This paper has presented
such a classification. The modelling process was presented as consisting
of three modelling choices involving the definition or adjustment of a
number of model properties
[Show abstract][Hide abstract] ABSTRACT: We propose that to realise a trade-off between the generality of problems that can be solved and efficiency of response, intelligent systems require representations of different types of knowledge, and such heterogeneous knowledge can most effectively be represented through multiple models with heterogeneous representation formats. Three dimensions; generality, precision, and scope of models are suggested here to formulate a frame-work for the structuring of heterogeneous models in intelligent systems. We embed such models within an architecture for intelligent systems that supports adaptive responses to unfamiliar situations by switching between pre-specified models and learning of these for future use.
[Show abstract][Hide abstract] ABSTRACT: Intelligent training systems have been developed using techniques advanced within the AI in education community. Each new system developed, however, exhibits its own inherent idiosyncrasies and does not address the problems of high development costs. This paper describes two generic approaches adopted within the Mobit project for building Intelligent Training Systems. The first being our approach to acquiring domain specific knowledge which initially requires a decomposition of the training objective into primitive generic tasks. The second being our approach to training based on domain independent learning styles and training strategies.
[Show abstract][Hide abstract] ABSTRACT: Traditional quantitative methods of analysis and simulation are compared with recently developed techniques in qualitative simulation by using as a case-study a simple dynamic model of the interacting markets for housing and mortgages. Analysis by the different techniques shows that while the qualitative simulation requires less detailed models, of the precision normally available in practice, it results in ambiguous descriptions of behaviour that for certain initial conditions can obscure the true behaviour. By contrast, quantitative simulation produces a unique precise behaviour, but in requiring excessively specific information of the modeller it may produce an inaccurate if precise outcome.
No preview · Article · Oct 1995 · Decision Support Systems
[Show abstract][Hide abstract] ABSTRACT: The authors of this paper have investigated the potential of using the recently developed constraint logic programming (CLP) languages as an implementation engine for qualitative simulation. This was initiated by the recognition that both utilize constraints as a basic representational formalism and constraint propagation as the inference mechanism. The focus of the work was on an advanced qualitative simulation system that uses fuzzy sets to describe the values of the system variables (FuSim). This led to a number of technical innovations that allows the semi-quantitative quantity space to be represented on a finite computational domain, and an incremental algorithm that makes active (constructive) use of the constraints rather than the passive use for consistency checking employed in conventional qualitative simulation algorithms. However, the approach described is applicable to any semi-quantitative simulation system. The system, called CLP-FuSim, is implemented in the CLP language CHIP and has been tested and validated on a number of benchmark examples. The resulting performance is as least as good as the Lisp counterpart, however, the CLP version has the distinct advantage of declarative semantics and non-determinism.
No preview · Article · Aug 1995 · Engineering Applications of Artificial Intelligence
[Show abstract][Hide abstract] ABSTRACT: This paper presents details of an architecture for parameter
identification, based on qualitative modelling and reasoning techniques,
which is used as the basis for a time-constrained model-based diagnosis
system. Section 2 presents the motivations for this work and outlines
the QPID (Qualitative Parameter Identification for Diagnosis)
architecture. Section 3 gives details of the Strategist module of QPID,
and how this meta-level module exercises control over the reasoning at
the object-level to attempt to meet time-constraints. Finally, section 4
presents the evaluation and conclusions of this interim work into the
use of qualitative techniques in parameter identification for diagnosis
[Show abstract][Hide abstract] ABSTRACT: The development of application systems for fault diagnosis has
attracted worldwide interest in many different research areas. In
particular, advanced techniques for finding faults through the use of
explicit structural and/or behavioural models of the physical system to
be diagnosed have been developed both in the area of control engineering
and in the field of artificial intelligence. Although many approaches to
creating model-based diagnostic systems (MBDS) exist, as yet, no clear
methodology is available for the selection of an appropriate approach to
solve individual given diagnostic problems. We have therefore been
developing a specification methodology that essentially comprises a
taxonomy of diagnostic tasks, a taxonomy of model-based systems, and a
set of guidelines that provides a mapping from the former to the latter,
Our aim is to provide a method by which existing MBDS tools and
techniques may be combined within a generic architecture in a principled
manner to produce effective diagnostic systems for given applications.
The structure of the methodology has been derived in part from the top
level architectural design of ARTIST, a generic model-based diagnostic
toolkit that combines a wide variety of model based diagnostic (MBD)
tools. This architecture is based upon the three main types of knowledge
that are necessary for the construction of model based diagnostic
[Show abstract][Hide abstract] ABSTRACT: This paper presents several innovations in the development of model-based diagnostic systems for diagnosing faults in continuous dynamic physical systems. The approach utilises recent developments in qualitative simulation techniques to cope with the inherent lack of modelling knowledge and to provide a qualitative description of the dynamic behaviour. In particular, techniques for the synchronous tracking of the model-based predictions and the evolution of the physical system between equilibria are developed. A discrepancy metric is defined that allows for the continuous degradation of the system behaviour from normal to faulty to be detected. And, most fundamentally, a method for iteratively searching through the space of possible model variations is presented. This provides explicit feedback from detected discrepancies to model adjustments and has the important advantage of reducing the sensitivity to modelling errors and approximate fault models. In the limit, no fault models are required. However, if available these can be used to initialise the search. An example is included which outlines the basic approach discussed in this paper.
No preview · Article · Jan 1995 · Artificial Intelligence in Engineering
[Show abstract][Hide abstract] ABSTRACT: This paper presents a preliminary approach to providing an
explanation facility for model-based diagnostic systems by the use of
causal ordering. An explanation algorithm is described that allows
explanations to be automatically generated for diagnostic findings by
searching through the causal graph derived from the structural and
behavioural model of the physical system under diagnosis. Typical
experimental results are given, demonstrating the successful use of the