Unknown response processing in a diagnostic expert system

• US 5,018,075 A
• Filed: 03/24/1989
• Issued: 05/21/1991
• Est. Priority Date: 03/24/1989
• Status: Expired due to Term

• Current Assignee: Bull H.n. Information Systems Italia SpA (Atos SE)
• Original Assignee: Bull HN Information Systems Incorporated (Atos SE)
• Inventors: Pan, Andrew Y., Rolston, David W., Ryan, Charles P., Howell, Thomas H.
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Sheikh, Ayaz R.
• Application Number:
  US07/328,380
• Time in Patent Office: 788 Days
• Field of Search: 364/513
• US Class Current: 706/46
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A diagnostic expert system incorporating a cause-effect graph is disclosed in which "yes", "no" and "unknown" are valid possible responses to a query. Each node in the graph is assigned a local decision factor (LDF) based on a given node'"'"'s desirability for selection during best-first search and a collapsed decision factor (CDF) based on the average values of the LDFs for all its daughter nodes. For a current node being processed, a list of all its daughter nodes is obtained and examined to remove all daughter nodes that have been visited before in the current transition and all daughter nodes which have prerequisites that are not met. Then, the daughter node with the largest LDF is selected and its test function is executed. If the user response is "yes", the daughter node is made the current node, and a list of its daughter nodes is obtained to continue the process. If the user response is "no", the daughter node is added to the closed list, and the process continues with the next daughter node of the current node. If the user respoanse is "unknown", the daughter node'"'"'s CDF is compared with the LDFs for the other daughter nodes of the current node which are not on the closed list and which have their prerequisites met. If the CDF is larger than all the LDFs, the daughter node is made the current node to continue the process; otherwise, the daughter node with the largest LDF is selected, and its test function is executed to continue the process.

Claims

1. In a diagnostic expert system for solving a problem incorporating a cause-effect graph having a plurality of nodes and in which "yes", "no" and "unknown" are valid possible responsive inputs to a query, the method for processing in a computer which includes the following steps:

(A) assign to each node in the cause-effect graph a local decision factor based on a given node'"'"'s desirability for selection during a best-first search of the cause-effect graph;

(B) assign to each node in the cause-effect graph a collapsed decision factor based on the values of the local decision factors for all daughter nodes of a current node; and

(C) for the current node under examination having daughter nodes;

(1) obtain a first list of all of the current node'"'"'s daughter nodes;

(2) examine the first list of daughter nodes and remove from the first list all daughter nodes on a closed list, a list of daughter nodes that have been visited before in examining a parent node of the current node which did not provide a solution to the problem, or whose prerequisites were not met;

(3) remove from the first list all daughter nodes which have prerequisites that are not met, and add them to the closed list;

(4) from the daughter nodes remaining on the first list, choose the daughter node with the largest local decision factor; and

(5) identify a test function to be executed for the chosen daughter node, and query the result of executing the identified test function;

(a) if the responsive input to a query produced in executing step (5) is a "yes", make the chosen daughter node the current node and return to step (C);

(b) if the responsive input to a query produced in executing step (5) is a "no", delete the chosen daughter node from the first list and add it to the closed list and return to step (C); and

(c) if the responsive input to a query produced in executing step (5) is an "unknown", compare the chosen daughter node'"'"'s collapsed decision factor with the local decision factors for the other daughter nodes of the first list for the current node of the first list and which have their prerequisites met; and

(i) if the collapsed decision factor of the chosen daughter node is larger than the local decision factors of the other daughter nodes of the list, make the chosen daughter node the current node and return to step (C); and

(ii) if the collapsed decision factor of the chosen daughter node is not larger than the local decision factor of all of the daughter nodes of the first list, then choose the daughter node with the largest local decision factor and return to step (5).

2. The diagnostic expert system of claim 1, in which, in step (B), a given node'"'"'s collapsed decision factor is determined by averaging the local decision factors for all its daughter nodes.

3. The diagnostic expert system of claim 1 in which, in step (B), if a given node is a terminal node, its collapsed decision factor is made equal to its local decision factor.

4. The diagnostic expert system of claim 2 in which, in the event that a given node is a terminal node such that there are no daughter nodes to average, its collapsed decision factor is made equal to its local decision factor.