RELIABILITY ESTIMATION FOR KNOWLEDGE BASED SYSTEMS

MANY KNOWLEDGE BASED SYSTEMS ARE DESIGNED AND BUILT WITH LITTLE ATTENTION PAID TO RELIABILITY. FAILURE TO PERFORM RELIABILITY ANALYSIS MEANS THAT FIELDED SYSTEMS MIGHT FAIL MORE FREQUENTLY THAN EXPECTED, CUSTOMERS MIGHT NOT BE WILLING TO TRUST KNOWLEDGE-BASED SYSTEMS FOR CRITICAL TASKS, AND DEVELOPERS MAY NEVER FULLY UNDERSTAND THE CAPABILITIES OF THE KNOWLEDGE-BASED TECHNOLOGY THEY ARE USING. THERE ARE TWO MAJOR REASONS BY KNOWLEDGE-BASED SYSTEMS ARE NOT SUBJECT TO RELIABILITY ANALYSIS: THERE ARE NO AVAILABLE METHODS FOR PERFORMING THE ANALYSIS, AND THE RAPID PROTOTYPING STYLE OF DEVELOPMENT MAKES IT DIFFICULT TO FORMULATE NEW RELIABILITY ESTIMATION OF A KNOWLEDGE-BASED SYSTEM DURING ANY STAGE IN A RAPID PROTOTYPING DEVELOPMENT PROCESS. THE APPROACH HAS THREE OBJECTIVES: ESTIMATE RELIABILITY FOR THE CURRENT PROTOTYPE, PREDICT THE FINISHED SYSTEM RELIABILITY, AND AID IN DESIGN DECISIONS DURING PROTOTYPING. THE FOUNDATION FOR OUR APPROACH IS A PARTITIONING OF BOTH THE RULES IN THE KNOWLEDGE-BASED SYSTEM AND THE INPUT SPACE TO THE SYSTEM TO ACCOMPLISH THESE OBJECTIVES.