Ontology-Supported Scaffolding for System Safety Analysis

System Safety Analysis is a valuable task used when trying to ensure that any thing
that can be represented with the systems-model does not behave in some manner
that is undesirable to the stakeholders in that system. It's a creative task,
with no known correct solution, with limited tool support. This thesis
investigates the possibility of providing support to analysts undertaking this
task through the use of ontology and pedagogy in an artificially intelligent tool.

An ontology to capture the system-model as understood by System-Theoretic
Accident Model and Processes (STAMP) was authored, building on an existing
set-theoretic representation. This required the authoring of underlying
ontology-modules, including one for Control Systems and one to capture
sufficient information for use with Situation Calculus. Together these capture
information to be used in reasoning about system behaviour. During System Safety
Analysis a user extends this ontology to model their system, and the intelligent
support tool interprets it to offer its advice.

The intelligent support tool uses Contingent Scaffolding to tailor its support
to the user, this pedagogical strategy was chosen as it's been shown to enable
the learner to produce a better quality product than they would be capable of
alone. Contingent Scaffolding depends upon knowledge of past behaviour of the
learner so that interventions can be pitched at the correct level for the
learner. Typically ontology authoring tools use a synchronic view of the
ontology, and so don't capture the required history. This tool uses
Situation Calculus to capture a diachronic view of the ontology such that the
history of authorship can be reasoned with to apply the Contingent Scaffolding
framework defined herein.

To evaluate the practicability of this approach the ontology and scaffolding
were implemented in software. This surfaced an issue with the inability to
inverse dependencies in Prolog, which was important to make the tools reuseable
and shareable. These were overcome by Protocols provided in Logtalk. The code
was then applied to other domains, such as robotics planning by a third-party,
demonstrating generalisability of the intelligent support tool.

A user study was conducted to evaluate the effectiveness of the intelligent
support tool, in which novices undertook a System Safety Analysis. The tool was
able to effectively provide support where definitions were missed and additional
patterns of behaviour were identified that are indicitive of the user needing
support.

The thesis makes a number of contributions including: a systems ontology with a
focus on capturing hypothetical and realised behaviour, a formal definition of a
contingent scaffolding framework that can be used with ill-defined tasks, and
the use of dependency inversion in Prolog to enable sharing of libraries. The
primary contribution is in the use of a diachronic view of ontology authoring to
provide support, which has been successfully exploited and has scope for
providing a platform for many more applications.