1. Abdul Rahim Ahmad MITM 613 Intelligent System Chapter 2: Rule-based Systems

2. Contents  Rules and facts  A rule-based system for boiler control  Rule examination and rule firing  Maintaining consistency  The closed-world assumption  Use of variables within rules  Forward-chaining (a data-driven strategy)  Conflict resolution  Backward-chaining (a goal-driven strategy)  A hybrid strategy  Explanation facilities Abdul Rahim Ahmad 2

3. Rules and facts  Rule-based system is a knowledge-based system where  The knowledge base is represented in the form of a set, or sets, of rules.  Rules are an elegant, expressive, straightforward, and flexible means of expressing knowledge.  Simplest type of rule is the production rule IF THEN  Example of a production rule is IF the tap is open THEN water flows  Rules need facts to be applied, for a rule-based system to be of use.  Facts are unconditional statements which are correct at the time that they are used. Abdul Rahim Ahmad 3

4. Facts  Facts can be:  looked up from a database;  already stored in computer memory;  determined from sensors connected to the computer;  obtained by prompting the user for information;  derived by applying rules to other facts.  The collection of facts which are known to the system at any given time is called the fact base.  In the rule IF the tap is open THEN water flows  the tap is open is a fact. Therefore, the fact could also be written as the rule: IF TRUE THEN the tap is open  Facts can be thought of as special rules, where the condition part is always true. Abdul Rahim Ahmad 4

5. Facts  Given the rule  IF the tap is open THEN water flows  and the fact  the tap is open  the derived fact water flows can be generated.  The new fact  is stored in computer memory  and can be used to satisfy the conditions of other rules, thereby leading to further derived facts. Abdul Rahim Ahmad 5

6. Rule Writing  Rule-writing is a type of declarative programming because rules represent knowledge that can be used by the computer, without specifying how and when to apply that knowledge.  The ordering of rules in a program should ideally be unimportant, and it should be possible to add new rules or modify existing ones without fear of side effects. (however, cannot always take it for granted).  For the declared rules and facts to be useful, an inference engine for interpreting and applying them is required.  Inference engines are incorporated into a range of software tools, such as  expert system shells  artificial intelligence toolkits  software libraries  Prolog language. Abdul Rahim Ahmad 6

7. Example rule-based system Abdul Rahim Ahmad 7 A rule-based system to monitor the state of a power station boiler and to advise appropriate actions. Boiler produce steam to drive a turbine and generator. Drum separate the steam from the water. Water is heated in the boiler tubes to produce a steam and water mixture that rises to the steam drum, which is a cylindrical vessel mounted horizontally near the top of the boiler. Steam is taken from the drum, passed through the super heater and applied to the turbine that turns the generator Sensors are fitted to the drum in order to monitor: the temperature of the steam in the drum; the voltage output from a transducer, which in turn monitors the level of water in the drum; the status of pressure release valve (i.e., open or closed); the rate of flow of water through the control valve.

8. Power Station Boiler Control Abdul Rahim Ahmad 8 The following rules have been written for controlling the boiler: /* Rule 2.1 */ IF water level low THEN open control valve /* Rule 2.2 */ IF temperature high AND water level low THEN open control valve AND shut down boiler tubes /* Rule 2.3 */ IF steam outlet blocked THEN replace outlet pipe /* Rule 2.4 */ IF release valve stuck THEN steam outlet blocked /* Rule 2.5 */ IF pressure high AND release valve closed THEN release valve stuck /* Rule 2.6 */ IF steam escaping THEN steam outlet blocked /* Rule 2.7 */ IF temperature high AND NOT(water level low) THEN pressure high /* Rule 2.8 */ IF transducer output low THEN water level low /* Rule 2.9 */ IF release valve open AND flow rate high THEN steam escaping /* Rule 2.10 */ IF flow rate low THEN control valve closed Rules 2.4 and the rest …. all involve taking a low-level fact, such as a transducer reading, and deriving a higher-level fact Rules 2.1, 2.2 2.3 consists of recommendations to the boiler operators Low-level vs high-level rules Deep vs shallow rules shallow rules - rules specific to one particular arrangement. Deep rule - expresses a fundamental rule - valid under any circumstances. Low-level rules - dependent on a reading. Represent the first stages toward reaching a conclusion. High-level rules use more abstract information. Closest to providing a solution to a problem.

9. Example rule-based system Abdul Rahim Ahmad 9 Rules for controlling the boiler: /* Rule 2.1 */ IF water level low THEN open control valve /* Rule 2.2 */ IF temperature high AND water level low THEN open control valve AND shut down boiler tubes /* Rule 2.3 */ IF steam outlet blocked THEN replace outlet pipe /* Rule 2.4 */ IF release valve stuck THEN steam outlet blocked /* Rule 2.5 */ IF pressure high AND release valve closed THEN release valve stuck /* Rule 2.6 */ IF steam escaping THEN steam outlet blocked /* Rule 2.7 */ IF temperature high AND NOT(water level low) THEN pressure high /* Rule 2.8 */ IF transducer output low THEN water level low /* Rule 2.9 */ IF release valve open AND flow rate high THEN steam escaping /* Rule 2.10 */ IF flow rate low THEN control valve closed

10. Rule examination and rule firing  The task of interpreting and applying the rules is done by the inference engine  The application of rules can be broken down as follows:  selecting rules to examine (available rules)  determining which are applicable (triggered rules making the conflict set)  selecting a rule to fire.  Rule examination and firing can be forward-chaining or backward-chaining. Abdul Rahim Ahmad 10

11. Maintaining Consistency  A key advantage of rule-based systems is their flexibility.  New rules can be added at will, but only if each rule is written with care, consistent and without assuming the behavior of other rules.  Make each rule an accurate statement in its own right  Rules can contain combinations of conditions and conclusions. Abdul Rahim Ahmad 11

12. Closed world assumption  All propositions are assumed either TRUE or FALSE.  Sometimes the rule-writer intends a rule differently.  Measures that could be taken to avoid this ambiguity-  modify the rules - so that they do not contain any negative conditions  modify the inference engine - ensure that a rule is examined before another Abdul Rahim Ahmad 12

13. Use of variables within rules  In real-world systems, variables can be used to make rules more general  reducing the number of rules needed  keeping the rule set manageable. Abdul Rahim Ahmad 13 /* Rule 2.18 */ IF control valve ?x is open THEN flow rate in tube ?x is high /* Rule 2.13 */ IF control valve 1 is open THEN flow rate in tube 1 is high /* Rule 2.14 */ IF control valve 2 is open THEN flow rate in tube 2 is high /* Rule 2.15 */ IF control valve 3 is open THEN flow rate in tube 3 is high /* Rule 2.16 */ IF control valve 4 is open THEN flow rate in tube 4 is high /* Rule 2.17 */ IF control valve 5 is open THEN flow rate in tube 5 is high

14. Forward-chaining  Inference engine applies a strategy for deciding which rules to apply and when to apply them.  Forward-chaining - a data- driven strategy:  rules are selected and applied in response to the current fact base. The fact base comprises all facts known by the system, whether derived by rules or supplied directly Abdul Rahim Ahmad 14

15. Forward-chaining  The key points of the scheme are as follows:  Rules are examined and fired on the basis of the current fact base, independently of any predetermined goals.  The set of rules available for examination may comprise all of the rules or a subset.  Of the available rules, those whose conditions are satisfied are said to have been triggered. These rules make up the conflict set, and the method of selecting a rule from the conflict set is conflict resolution.  Only one rule in a conflict set is fired on a given cycle. (once a rule has fired, the stored deductions have potentially changed, and it cannot be guaranteed that other rules in the conflict set still have their condition satisfied). Abdul Rahim Ahmad 15

16. Alternative form of Forward- chaining Single instantiation of variables  conclusions may be performed using just the first set of instantiations that are found Abdul Rahim Ahmad 16 Multiple instantiation of variables  conclusions may be performed repeatedly using all possible instantiations

17. Example of Forward-chaining Rules /* Rule 2.20 */ IF control valve ?x is open THEN flow rate in tube ?x is high /* Rule 2.21 */ IF flow rate in tube ?x is high THEN close control valve ?x Abdul Rahim Ahmad 17 Suppose that we start with two facts: control valve 1 is open control valve 2 is open In multiple instantiation, each rule would fire once, generating conclusions in the following order: flow rate in tube 1 is high flow rate in tube 2 is high close control valve 1 close control valve 2 multiple instantiation is a breadth-first process Under single instantiation, each cycle of the inference engine would result in a rule firing on a single instantiation of the variable x. After four cycles, conclusions would have been generated in the following order: flow rate in tube 1 is high close control valve 1 flow rate in tube 2 is high close control valve 2 single instantiation is a depth-first process

18. Rete (“ree-tee”) Algorithm  In forward-chaining, one inefficiency is that once a rule has been selected from the conflict set and fired, the conflict set is thrown away and the process starts all over again.  Rete algorithm is an efficient approach that examine only those rules whose condition is affected by changes made to the fact base on the previous cycle. Abdul Rahim Ahmad 18

19. Example of Rete Algorithm Abdul Rahim Ahmad 19 Rules /* Rule 2.22*/ IF ?p is a pipe of bore ?b AND ?v is a valve of bore ?b THEN ?p and ?v are compatible Condition parts of all rules assembled into a Rete network. Each node represents an atomic condition, i.e., contains a simple test. Two types of nodes: alpha nodes can be satisfied by a single fact beta nodes can only be satisfied by a pair of facts. Lets say a pipe is found in node α1 with bore 100, and so the fact would be passed on to node β1. However, node β1 would not be satisfied yet as it has not received info from node α2. Imagine, as a result of firing other rules, another pipe with bore 100 is found in α2. The Rule 2.2 can be added to conflict set:

20. Conflict Resolution  Choosing one rule to fire from those that are able to fire (conflict set).  Can choose 3 ways to fire rule in conflict set:  Fire immediately the first rule that qualifies – first come first serve. (conflict set is not assembled at all, and the order in which rules are selected for examination determines the resolution of conflict)  Fire according to explicitly stated priority value.  Use Metarules (knowledge about how that knowledge should be applied) Abdul Rahim Ahmad 20 /* Rule 2.1a */ IF water level low THEN open control valve PRIORITY 4.0 /* Rule 2.2a */ IF temperature high and water level low THEN open control valve AND shut down boiler tubes PRIORITY 9.0 /* Metarule 2.23 */ PREFER rules about shut-down TO rules about control valves

21. Backward-chaining (a goal-driven strategy)  Inference strategy that assumes the existence of a goal that needs to be established or refuted  Initially, only those rules that can lead directly to the fulfillment of the goal are selected for examination  Searching for what rules to apply next can use a depth- first or breadth-first strategy using a search tree where the nodes of the search tree are rules. Abdul Rahim Ahmad 21 Depth First Search

22. Depth First Backward chaining Abdul Rahim Ahmad 22 /* Rule 2.1 */ IF water level low THEN open control valve /* Rule 2.2 */ IF temperature high AND water level low THEN open control valve AND shut down boiler tubes /* Rule 2.3 */ IF steam outlet blocked THEN replace outlet pipe /* Rule 2.4 */ IF release valve stuck THEN steam outlet blocked /* Rule 2.5 */ IF pressure high AND release valve closed THEN release valve stuck /* Rule 2.6 */ IF steam escaping THEN steam outlet blocked /* Rule 2.7 */ IF temperature high AND NOT(water level low) THEN pressure high /* Rule 2.8 */ IF transducer output low THEN water level low /* Rule 2.9 */ IF release valve open AND flow rate high THEN steam escaping /* Rule 2.10 */ IF flow rate low THEN control valve closed One of the relevant rules is selected for examination The system searches the rule base for a rule which can satisfy this goal. If fails the system backtracks to the last place where a choice between possible rules was made and will try an alternative. If information is not yet known, it becomes the new goal. This process continues until a goal is satisfied by the information in the fact base. When this happens, the original goal is fulfilled, and the chosen path through the rules is the solution. If all possible ways of achieving the overall goal have been explored and failed, then the overall goal fails.

23. Breadth First Backward chaining Abdul Rahim Ahmad 23 /* Rule 2.1 */ IF water level low THEN open control valve /* Rule 2.2 */ IF temperature high AND water level low THEN open control valve AND shut down boiler tubes /* Rule 2.3 */ IF steam outlet blocked THEN replace outlet pipe /* Rule 2.4 */ IF release valve stuck THEN steam outlet blocked /* Rule 2.5 */ IF pressure high AND release valve closed THEN release valve stuck /* Rule 2.6 */ IF steam escaping THEN steam outlet blocked /* Rule 2.7 */ IF temperature high AND NOT(water level low) THEN pressure high /* Rule 2.8 */ IF transducer output low THEN water level low /* Rule 2.9 */ IF release valve open AND flow rate high THEN steam escaping /* Rule 2.10 */ IF flow rate low THEN control valve closed All rules at the same depth would be examined. If all branch fails, then the system will not need to backtrack, as it will simply carry on down those branches which still have the potential to fulfill the goal. The first solution to be found will be the one with the shortest (or joint- shortest) chain of rules. Require large amounts of computer memory to keep a record of progress along all branches of the search tree, rather than just one branch.

24. Prolog  The inference mechanism built into the Prolog language is a depth-first backward-chainer.  There are two sets of circumstances under which backtracking takes place:  when a goal cannot be satisfied by the set of rules currently under consideration; or  when a goal has been satisfied and the user wants to investigate other ways of achieving the goal (i.e., to find other solutions). Abdul Rahim Ahmad 24

25. Implementation of backward-chaining Abdul Rahim Ahmad 25 Generalized flowchart for backward-chaining from a goal G1. We assume each rule has only one condition. Length of the chain of rules cannot be predetermined. A recursive definition is also available and is important in Prolog. Backward-chaining is part the Prolog language, expert system shells and artificial intelligence toolkits.

26. Recursive Definition of Backward Chaining Abdul Rahim Ahmad 26 define function backwardchain(G); /* returns a boolean (i.e., true/false) value */ /* G is the goal being validated */ variable S, X, C; result:= false; /* ‘:=‘ represents assignment of a value to a variable */ S:= set of rules whose conclusion part matches goal G; If S is empty then result:= false; else while (result=false) and (S is not empty) do X:= rule selected from S; S:= S with X removed; C:= condition part of X; if C is true then result:=true elseif C is false then result:=false elseif (backwardchain(C)=true) then result:=true; /* note the recursive call of ‘backwardchain’ */ /* C is the new goal */ endif; endwhile; endif; return result; /* ‘result’ is the value returned by the function */ /* ‘backwardchain’ */ enddefine;

27. Variations of backward-chaining  Use depth-first or breadth-first search for rules;  Decide whether or not to pursue other solutions  Different ways of choosing between branches to explore (in depth- first search only)  Deciding upon the order in which to examine rules (in breadth-first search only)  Whether to fire the sequence of rules or simply to conclude that the goal is proven (after having found a succession of enabling rules whose lowest-level conditions are satisfied). Abdul Rahim Ahmad 27

28. A hybrid strategy  Makes use of an inference engine that can be seen of as part forward-chaining and part backward-chaining.  Example :  ARBS (Algorithmic and Rule-based Blackboard System).  In hybrid strategy, a rule dependence network is built prior to running the system.  For each rule, the network shows which other rules may enable it (its antecedents), and which rules it may enable (its dependents) Abdul Rahim Ahmad 28

29. Rule dependance network Abdul Rahim Ahmad 29

30. Explanation facilities  Can be divided into two categories:  how a conclusion has been derived;  why a particular line of reasoning is being followed.  Example explanation for a recommendation to replace the outlet pipe Abdul Rahim Ahmad 30 Replace outlet pipe BECAUSE (Rule 2.3) steam outlet blocked steam outlet blocked BECAUSE (Rule 2.4) release valve stuck release valve stuck BECAUSE (Rule 2.5) pressure high AND release valve closed pressure high BECAUSE (Rule 2.7) temperature high AND NOT(water level low) NOT(water level low) BECAUSE (Rule 2.8) NOT(transducer output low) Release valve closed, temperature high and NOT(transducer output low) are supplied facts

31. END Abdul Rahim Ahmad 31