Guidelines for the verification and validation of expert system software and conventional software: Project summary. Volume 1
- Science Applications International Corp., McLean, VA (United States)
This eight-volume report presents guidelines for performing verification and validation (V&V) on Artificial Intelligence (Al) systems with nuclear applications. The guidelines have much broader application than just expert systems; they are also applicable to object-oriented programming systems, rule-based systems, frame-based systems, model-based systems, neural nets, genetic algorithms, and conventional software systems. This is because many of the components of AI systems are implemented in conventional procedural programming languages, so there is no real distinction. The report examines the state of the art in verifying and validating expert systems. V&V methods traditionally applied to conventional software systems are evaluated for their applicability to expert systems. One hundred fifty-three conventional techniques are identified and evaluated. These methods are found to be useful for at least some of the components of expert systems, frame-based systems, and object-oriented systems. A taxonomy of 52 defect types and their delectability by the 153 methods is presented. With specific regard to expert systems, conventional V&V methods were found to apply well to all the components of the expert system with the exception of the knowledge base. The knowledge base requires extension of the existing methods. Several innovative static verification and validation methods for expert systems have been identified and are described here, including a method for checking the knowledge base {open_quotes}semantics{close_quotes} and a method for generating validation scenarios. Evaluation of some of these methods was performed both analytically and experimentally. A V&V methodology for expert systems is presented based on three factors: (1) a system`s judged need for V&V (based in turn on its complexity and degree of required integrity); (2) the life-cycle phase; and (3) the system component being tested.
- Research Organization:
- US Nuclear Regulatory Commission (NRC), Washington, DC (United States). Div. of Systems Technology; Electric Power Research Inst., Palo Alto, CA (United States). Nuclear Power Div.; Science Applications International Corp., McLean, VA (United States)
- Sponsoring Organization:
- Nuclear Regulatory Commission, Washington, DC (United States); Electric Power Research Inst., Palo Alto, CA (United States)
- OSTI ID:
- 42512
- Report Number(s):
- NUREG/CR-6316-Vol.1; SAIC-95/1028-Vol.1; ON: TI95010250; TRN: 95:011048
- Resource Relation:
- Other Information: PBD: Mar 1995
- Country of Publication:
- United States
- Language:
- English
Similar Records
Guidelines for the verification and validation of expert system software and conventional software: Volume 5, Rationale and description of verification and validation guideline packages and procedures. Final report
Guidelines for the verification and validation of expert system software and conventional software. Volume 7, User`s manual: Final report