The model-based approach to diagnosis emerged in the early 80's as a solution to some limitations
of the traditional expert system approach based on heuristic knowledge. Since then, research in
knowledge-based systems has very much focused on exploiting model based reasoning methods in order
to overcome limitations due to previously used expert systems technology based on empirical associations.
The core of the new technology is an explicit, declarative model of a technical system that can be composed
from context independent re-usable behaviour models of components taken from a domain library.
Methods for formulating and using such models at a conceptual and qualitative level have been developed in
the area of qualitative reasoning (Weld-de Kleer 90),
(Faltings-Struss 92) and several partners in
the project have made significant contributions, e.g.
(Dague et. al. 90),
(Lee-Ormsby 94),
(Struss 88,
89, 92).
Automated diagnosis is the task for which the most advanced work has been
carried out
(Console-Hamscher-de Kleer 92).
Today, model based diagnosis shows both a strong theoretical foundation in terms of logical
theories
(Reiter 87),
(de Kleer-Mackworth-Reiter 90),
(Console-Torasso 91),
(Struss-Dressler 89),
(Dressler-Struss 92,
94),
(Dague-Deves-Raiman 88),
(Tatar 94),
(Tatar-Iwanowski 94)
and considerable work on industrial problems, such as diagnosis of power networks
(Beschta et al. 93),
(Struss 92,
93), analogue circuits
(Dague et al. 87,
94), diagnosis and failure modes and effects
analysis of automotive subsystems
(Price et al. 92;
Hunt et al. 91,
93),
(PROMOTEX). The survey
indicates that the methods, techniques, and systems developed are mature enough for industrial
applications, at least for selected domains and tasks. The first tools for model-based diagnosis are appearing
on the market (IDEA, RODON).
In the automotive field diagnostic systems have been mainly developed for off-board applications.
The first diagnostic approach in this field was the exploitation of on-board diagnosis provided from
the ECU, using a hand-held system compatible with standard ISO 9141. Later, diagnostic tools employed
knowledge-based systems technology, mainly based on fault-tree analysis. Examples for such systems
are Strategy Engine from HP and TestBench
from Carnegie Group. To overcome these limitations second-generation
diagnostic systems have been developed: these systems are based on a structural and functional
approach, the so-called deep-model. Examples of such systems are IDEA (Fiat Research Centre) and
MDS (Daimler-Benz Research).
The main tasks to be solved for the application of model-based diagnosis at a broader scale are:
- the generation of model libraries covering the domain,
- the creation of environments that provide support for model building, maintenance of model
libraries, and customisation of diagnostic systems,
- the development of models and strategies appropriate for diagnosing dynamic systems.
From the point of view of the on-board diagnosis at the moment only emission related systems
have been taken into account for monitoring/fault detection. No other diagnostic provisions have
yet been taken, apart from SAE/ISO standardisation concerning the diagnostic interfaces
between on-board systems and out-board front-end equipment, which are mainly related to
electrical faults diagnosis (detection of short circuit, open circuit). Error codes generated by
current on-board diagnosis systems are more related to classes of symptoms and fault detection,
rather than to fault localisation. Centralised diagnostic systems are not present on the market.
Some prototypes of on-board systems have been demonstrated on Japanese concept cars
(Mitsubishi HSR-IV,
Toyota AXV-V), while
General Motors and Ford
have started research on model-reference diagnostic techniques.
The main innovative aspects of the project are:
- the use of model based diagnosis for vehicles,
- the strong integration between on-board and off-board diagnostic systems,
- definition of general guidelines for on-board vehicle diagnosis. where no standard
approaches have yet been defined (apart from the cited OBD II).
The expected results are innovative per se:
- a set of reusable model libraries for vehicle systems,
- an environment for building libraries and model-based diagnostic systems,
- a set of diagnostic systems, tested on a vehicle, that provide more coverage, guarantee
for completeness of their results, more discriminative power, and capabilities of fault prediction.
One of the major advances of the project is the development of model libraries covering entire
application domains and their evaluation in a variety of application examples. The scope of current
modelling and diagnostic techniques will be expanded by not only providing models of correct and
faulty behaviour but also by formalising knowledge about the transition from correct (or optimal)
to faulty (or suboptimal) behaviour ("wear models").
This project is closely related to
INDIA (intelligent diagnosis in application), proposed to
BMBF (German Federal Ministry of Education and Research). The INDIA project investigates
how the diagnosis relevant knowledge in an enterprise can be acquired automatically. This
requires the integration of knowledge from the development about system functions and knowledge
about potential faults of components, as it is recorded by quality assurance methods like FMEA.
In the UK-funded project "Jacquard", qualitative models of automotive electrical systems have
been built
(Lee, Ormsby 94) and used as part of
prototype diagnosis and failure modes and
analysis (FMEA) systems (Price et al. 92;
Hunt-Price-Lee 93).
Currently, two related projects, both funded by the UK Engineering and Physical Sciences
Research Council. The FLAME project, is developing further the FMEA work which was under
taken on the Jacquard project. The end result of this project will be model-based tools to support
FMEA for use by automotive design engineers in their daily work. The tools combine electrical
circuit descriptions derived from an ECAD tool and a model of system function in order to
reason about the effects of failure
(Price et al. 95).
Dream is a much smaller project, concerned with the development of models that can be used
to support the development of more traditional diagnostic systems. The aim here is to investigate
the extent to which model-based techniques can assist the developers of conventional automotive
diagnostic systems.