SAHARBAUGH@ROO.FIT.EDU (06/22/91)
I was too late for the summarized posting so here is some more information, last on the list so you can skip it easily. --- Here are some items regarding Expert Systems and Ada: --- Regarding CLIPS/Ada here are some excerpts from a note from CSC, the organization that distributes CLIPS for NASA. It may be no longer true. April 11, 1990 Dear CLIPS User: Regarding CLIPS Ada for MS DOS... We cannot get a reliable executable of CLIPS Ada created for the MS DOS environment, even with the latest versions of Alsys (4.3) and Meridian (3.0) compilers. Consequently, we cannot provide an MS DOS Ada executable and will not pursue the matter further until the next release of CLIPS Ada. <stuff left out> Alternately, you may copy the Ada source code to a UNIX workstation or a VAX and recompile it there where it should work fine. 16511 Space Center Boulevard Houston, Texas 77058 713.486.8153 ----------- Following are excerpts from an ISI project report which may help you to get started on "Expert Systems written in Ada". If you are interested in Beta testing AFES or in learning more about it and/or the research project, please contact me. We chose the AFES fuzzy expert system because it allows reasoning about inexact situations, such as "fuel is low" and "threat is high", and thereby would be a more realistic demonstration. Only Ada based expert systems were considered. AFES is a fuzzy, rule-based, expert system shell written in Ada. AFES is a fuzzy expert system, performs forward reasoning, accepts sensor inputs and has a blackboard and a rules editor. AFES is a full capacity expert system shell that can be used interactively or can be called from within a running application program. Two other expert systems were considered for use, a system from Software Artistry and a system from Louis Baker [BAKER]. [BAKER] "Artificial Intelligence with Ada", 1st edition (1989) Louis Baker, McGraw-Hill, ISBN 0-07-003350-1 It was difficult to get consistent information about the Software Artistry system(St. Louis, MO, circa 1989). Their brochure said it was written in Ada but upon inquiry it seemed that it would be written in Ada only if someone agreed to pay the $5000.00 asking price. It also was not forward reasoning. It used a linked-list data structure and little recursion; however, it did not utilize dynamic memory management and was judged unsuitable for an embedded real-time system ication. The system in Louis Baker's book performs both forward and backward reasoning. We obtained the source code from Dr. Baker and found it to be as "ugly" as Dr. Baker said it was. The code uses recursion and the Ada keyword "new" to allocate dynamic memory but it does not use "unchecked_deallocation" to tell the memory manager when it can reclaim storage. AFES was chosen for use because it performed fuzzy logic, was designed to be an embeddable component, the authors were available locally and we were able to influence the Ada code as it was being converted from Pascal. AFES employs a matrix of bits to determine if a rule is a concluding rule. A list of all factual data called the blackboard is maintained by the shell. The blackboard contains clauses in a compressed format to facilitate comparison. The certainty factor is a measure of how well a fact on the blackboard satisfies the premise of a clause. The certainty factor is computed from the overlap of the ranges of two clauses Every clause on the blackboard has a certainty factor, expressed as a real number between 0 and 1. AFES follows the conventions of fuzzy logic as defined by Lofti Zadeh. The logical functions "and" and "or" are evaluated left to right unless the order of operations is changed by parentheses. The logical "and" accepts two operands (the certainty factor on the right and the certainty factor on the left) and returns a minimum of the two values. Conversely, the logical "or" returns a maximum of the two certainty factors. The logical "not" negates (subtracts the certainty factor from 1) only the first clause immediately to its right. If several clauses are bounded by parentheses, the total certainty factor of all clauses is negated. The "not" returns the difference between 1 and the certainty factor of the clause(s) to its right. The actual inferencing process is very simple when described in high-level terms. The inferencing module is indeed very simple because it relies heavily on the supporting routines to compute and evaluate clauses. Passes are made through the knowledge base until it is exhausted. During each pass, each rule that has not already been fired is broken into clauses and then compared to the blackboard to compute its certainty factor. If the premise is true, the conclusion is then fired and the rule is added to the list of fired rules. If the rule is a concluding rule, then the rule is added to the list of fired conclusions as well. When using the interactive AFES shell, if the inferencing process has been completed and no concluding rules have been fired, the shell proceeds to ask the user for factual data in order to reach at least one conclusion. Most of the data used by the shell is stored dynamically in the form of linked lists of data structures. Some are straightforward such as the list of rules. Others are more complex, like the matrix and the blackboard. All lists are maintained by Ada's heap management procedures. An example rule is: if exposure is high and fuel-status is not-low and mission-status is return-to-base then speed is afterburner -- end of message sam harbaugh saharbaugh@ROO.FIT.EDU ----------- ISI Integrated Software, Inc. 1945 Palm Bay Rd #7 Palm Bay, FL 32905 ----- the end, really