Guidelines for the verification and validation of expert system software and conventional software: Survey and assessment of conventional software verification and validation methods. Volume 2
Nuclear Regulatory Commission, Washington, DC (United States). Div. of Systems Technology, S Mirsky, Nuclear Regulatory Commission, Washington, DC (United States), E.H. Groundwater, J Hayes, L Miller · 1995
By means of a literature survey, a comprehensive set of methods was identified for the verification and validation of conventional software. The 153 methods so identified were classified according to their appropriateness for various phases of a developmental life-cycle -- requirements, design, and implementation; the last category was subdivided into two, static testing and dynamic testing methods. The methods were then characterized in terms of eight rating factors, four concerning ease-of-use of the methods and four concerning the methods` power to detect defects. Based on these factors, two measurements were developed to permit quantitative comparisons among methods, a Cost-Benefit metric and an Effectiveness Metric. The Effectiveness Metric was further refined to provide three different estimates for each method, depending on three classes of needed stringency of V&V (determined by ratings of a system`s complexity and required-integrity). Methods were then rank-ordered for each of the three classes by terms of their overall cost-benefits and effectiveness. The applicability was then assessed of each for the identified components of knowledge-based and expert systems, as well as the system as a whole.