Guidelines for the verification and validation of expert system software and conventional software: Rationale and description of V&V guideline packages and procedures. Volume 5

This report is the fifth volume in a series of reports describing the results of the Expert System Verification C, and Validation (V&V) project which is jointly funded by the U.S. Nuclear Regulatory Commission and the Electric Power Research Institute toward the objective of formulating Guidelines for the V&V of expert systems for use in nuclear power applications. This report provides the rationale for and description of those guidelines. The actual guidelines themselves are presented in Volume 7, {open_quotes}User`s Manual.{close_quotes} Three factors determine what V&V is needed: (1) the stage of the development life cycle (requirements, design, or implementation); (2) whether the overall system or a specialized component needs to be tested (knowledge base component, inference engine or other highly reusable element, or a component involving conventional software); and (3) the stringency of V&V that is needed (as judged from an assessment of the system`s complexity and the requirement for its integrity to form three Classes). A V&V Guideline package is provided for each of the combinations of these three variables. The package specifies the V&V methods recommended and the order in which they should be administered, the assurances each method provides, the qualifications needed by the V&V team to employ each particular method, the degree to which the methods should be applied, the performance measures that should be taken, and the decision criteria for accepting, conditionally accepting, or rejecting an evaluated system. In addition to the Guideline packages, highly detailed step-by-step procedures are provided for 11 of the more important methods, to ensure that they can be implemented correctly. The Guidelines can apply to conventional procedural software systems as well as all kinds of Al systems