How the Inference Engine Works › Forward Chaining

Forward Chaining

The inference engine starts by posting rules, which may exist directly in the INFER block or in rule methods called from the INFER block. Initially, all rules have a state of Ready.

The engine visits the rules in priority order. This is not necessarily the order in which they were posted, but the order specified by their priority keyword.

• If the premise can be evaluated to TRUE, the rule fires.
• If the premise can be evaluated to FALSE, the rule fails.
• Whether TRUE or FALSE, the rule is finished. Its state is changed to Fired or Failed, and the engine will not visit it anymore.
• If any attributes in the premise are UNKNOWN, the rule is pended-that is, the engine changes the rule's state from Ready to Pended. The engine adds the attribute to a list of attributes to be watched.

Then, the engine proceeds to the next Ready rule in the priority order and fires, fails, or pends it.

Whenever a rule fires and its action makes an assignment to an attribute for a pended rule, the engine takes notice. It changes the rule dependent on the attribute from Pended to Ready and, thus, readies the rule for further consideration. The engine then goes back to the top of the posted-rule list and takes another pass at the Ready rules, visiting them in priority order.

Forward chaining can end in three ways. You can:

• Allow inferencing to continue until all rules have fired, failed, or pended, and no Ready rules remain.
• Define a goal attribute, which will halt the inferencing process as soon as it has been assigned a value. Goal attributes are optional for forward chaining.
• Use a stopchain command.