In general the inputs of automated construction are:

- the input problem for which we need to construct a method (given as: data and knowledge, e.g. a particular case to diagnose);
- the assumptions under which the method will have to operate;
- the goals that the resulting method will have satisfy.

The outputs are:

- the description of the constructed method;
- the solutions computed by the method
- the possibly slightly adjusted versions of the input problem, the goals and the assumptions.

The input/output relation of the construction process is as follows:

- the output has to be a representation of a method;
- it must not conflict with the (possibly adapted) assumptions;
- it must satisfy the (possibly adapted) goals
- the slightly adapted inputs (assumptions, goals, problem) have to be closely related to the original ones.

Examples of the inputs in the context of diagnosis are (1) the diagnostic problem containing the observed behaviour and the behaviour model, (2) the single fault assumption, and (3) a goal such as a maximal size of the diagnosis.

The goal of automated construction of methods is to construct a method
that produces acceptable solutions for a given problem under
particular assumptions and desired goals.
Our approach is to first configure and then validate a method, and, if
this validation fails, to iterate the configuration step.
We call the construction before validation * static
configuration* and the configuration using the validation results *
dynamic configuration*. The question in static configuration is
``Which PSM is optimal?'' and in dynamic configuration ``What should
be done if the PSM does not give the desired solution?''.
In line with the distinction of static and dynamic configuration we
distinguish static and dynamic goals.
* Static goals* are requirements (of the
solution or of the method) that can be guaranteed solely on the
basis of the description of the method.
For example, the goal that a method always produces singleton diagnoses.
* Dynamic goals* are requirements of the solution that can only be
validated
after executing the method.
For example, the goal of a maximal number of diagnoses.
This
distinction between static and dynamic goals is not fixed. With more
knowledge a dynamic goal might be established statically.
It depends on the knowledge that is available about methods, whether goals are
static or dynamic.

The method description that we have to construct has to satisfy both types of goals. The construction process proceeds in two steps. The first step of the construction process concerns the configuration of a method that satisfies the static goals. If there is no such a method, the second step occurs: we adapt the problem, assumptions or goals slightly such that a method can be constructed that satisfies the static goals (possibly slightly adjusted). If this method also satisfies the dynamic goals, a suitable method has been constructed, otherwise we try to adapt the method in such a way that is does. However, when this is impossible we again adapt the problem, assumptions or goals slightly and configure a method for these new inputs. The basic idea is that we construct the method that computes the ``best'' possible solutions for the given problem and assumptions and desired goals. For computing these solutions, the constructed method possibly has to apply to a problem which is a slight modification of the original problem, and under possibly slightly modified assumptions and for possibly slightly modified goals.

In all this, the object of the construction is the method description. The possibly slightly adjusted assumptions, goals and problem are side effects of configuring an appropriate method for a given problem under particular circumstances.

Fri Oct 4 13:40:35 MET DST 1996