Previous: Domain knowledge Up: Propose & revise Next: Inference knowledge

Task Knowledge

Figure shows a visualization of the function structure for the P&R approach at the first level of decomposition.

The top level approach of P&R is to construct a chain of design choices by selecting repeatedly from a set of requirement, design choice associations. The features of the design choices are then checked on consistency. If any constraint are is violated then a set of violation, fix pairs is consulted. The process terminates once either all constraints are satisfied or tall fixes have been tried unsuccesfully.

1. Propose-task(+A:attribute, -V:value)

   The goal of the propose task is to identify a set V of attribute values that satisfy the attribute, solution constraint pairs.

   This task takes an attribute and produces a value or a value definition in terms of the values of other attributes. The propose-task can use this value in different ways;

   0pt

2. sep 0pt

   • If propos-op returns a fixed value, then return this value.

   • If propose-op returns a definition in terms of other attribute values and the definition is not cyclic (values defined in terms of each other), then apply the propose task on the attributes of the definition.

   • If the definition is cyclic then ask the user (or another agent) to get an initial value.

3. check-task(+A:attribute, -OK:boolean)

   The check task is directly supported by a domain operation and can therefore also be classified as a kads-II PIA.

4. revise-task(+A:attribute*, +V:value-set*, +C:constraint*, -R:revision)

   The goal of Differentiate is to create a Revision from a set A (the attributes), a set V (values), a set C (Constraints) and a set R (The design revisions). Revise(A, V, C, R) holds if:

   0pt

5. sep 0pt

   • propose(A, V) holds, the values are propositions for this attribute.
   • constraint(A, R) holds, the revised values satisfy the constraints.

   Revision operators are included in the domain schemata. Different revisions can be available and then the revision task has to choose one. The revisions for the various attributes should be chosen in such way that the proposed design converges to a solution, i.e. both local constraint and global constraints are fulfilled.

   Sub-tasks of revise in the P&R approach are:

   (Control: for each of the violated constraints, for each of the values that contribute to the constraint)

   1. get-revisions-task(+A:attribute, +C:constraint*, -R:revision*). This task takes a set of violated constraints and tries to find the possible revisions. In principle any value that contributes to the constrain or the constrained value be proposed in order to be revised.

   2. get-preference-task(+R:revision*, -P:preference*) Get a preference set from the domain knowledge or the user of the different possibilities for revision.

   3. get-interactions-task(+C:constraint, -I:revision) Produce a listing of chains of interacting revisions. `Each chain originates from a constraint whose fixes make other constraints more likely to be violated'.

   4. select-revision-task(+R:revision*, +IR: revision*, +P:preference, -Fix:revision). Select a fix to apply based on the possible revisions, the interacting revisions and the preferences.

   5. apply-revision-task(+/-A:attribute, +R:revision*). Apply the revision on the attributes of the partial design.