The revise inference is the most complex inference of the method. Some models of P&R decompose the revise step into another inference structure [Motta et al., 1994,Fensel, 1995]. Here, we choose to keep the revise inference as a primitive, without decomposing it. We make this choice because in making explicit how the knowledge used by the inference should be organized, it is not necessary to decompose the carrying out of the inference.
The revise inference repairs a violated constraint. Figure 3 illustrates the inference with its knowledge roles: SELECTED VIOLATED CONSTRAINT, FIXES, FIX PROCEDURES, and PARAMETER VALUES.
Revise repairs the violation of a constraint by applying FIXES separately or by combining them, until the constraint is no longer violated. The FIX PROCEDURES accomplish the modifications on the PARAMETER VALUES. All fixes have a unique procedure that is represented by an expression. The order of application of fixes is based on their cost. In the VT application, the constraint maximum hoist cable traction ratio has three fixes associated with three different costs:
In trying to repair the violation of the constraint maximum hoist cable traction ratio, the fixes will be applied in the following order: 1, 2, 1 2, 3, 1 3, 2 3, 1 2 3.
Besides its cost, each fix has a type that characterizes the type of modification achieved by it. There are four types of fixes: (1) change-value fixes, which change the value of some parameter only once; (2) increment fixes, which increment the value of some parameter along some step; (3) decrement fixes, which decrement the value of a parameter along some step; and (4) upgrade fixes, which upgrade the model of a parameter value according to a list of pre-defined values. Fixes of type increment, decrement and upgrade are recurrent fixes, because they can be applied repeated times. Moreover, fixes of type increment or decrement have an associated step. In the VT task, the types of fixes are illustrated next:
Figure 9: The formalization of the revise inference and its knowledge roles.
When a fix combination contains one or more recurrent fixes in which a value should be stepped or upgraded along some dimension, all combinations of steps or upgrading for the fix combination should be tried before attempting the next combination. To limit the times the step or the upgrade must be applied, the modified values of the recurrent fixes have a maximum value, a minimum value and a list containing all the possible upgrade models. The maximum, minimum and upgrading values are related to the fixes of type increment, decrement and upgrade, respectively. The association of limiting values for the modified parameters of the recurrent fixes prevents the infinite application of these fixes. In Yost's description, to limit the number of times recurrent fixes are applied, special constraints are defined for the modified parameters of the recurrent fixes. These constraints are special, because no fixes are associated with them. Here, we propose associating limiting values to the modified parameters to make this distinction clear.
Figure 9 represents the formalization of the
revise inference and its knowledge roles. Definitions
(24)-(30) depict the knowledge roles. In (31), the
revise inference repairs the SELECTED VIOLATED