| The important problem of validating models used in dependability analysis is addressed. The lack of model visibility makes model validation difficult and without adequate validation, there is a risk involved in using the results of dependability analysis. This risk has to be reduced considerably for both critical and non-critical applications.; A knowledge-based methodology for validation of dependability models is developed. The scope of this methodology is limited to conceptual validation with respect to model completeness and model consistency. Knowledge derived from the system specification is used for model validation. Knowledge concerning dependability models is also stored in the knowledge base, thus enhancing model visibility. The model validation process primarily involves: (1) Generating a reference model for a targeted system given the system knowledge in the knowledge base, (2) Translating the target dependability model into the reference model format given the model knowledge in the knowledge base, and (3) Comparing the two models in the reference model format to find incompleteness and inconsistencies in the target dependability model. Event trees are used as the common format for both reference and target models. Also, for validating complex system models, this methodology generates and compares hierarchical models.; The effectiveness of this methodology is verified by validating the dependability models of some example systems. Five simple systems and ten complex systems are selected, primarily from published literature. For testing purposes, errors are introduced in some of the dependability models of these systems. All errors introduced are detected, and for some models, more errors are detected than introduced. The reason for the latter is that these models had voluntary omissions or errors.; This effort is also generalized with the definition and specification of a knowledge-based framework for analysis of fault-tolerant systems. The major components of this framework are: (1) An intelligent model building aid, (2) A multilayered model evaluation methodology, and (3) A model validation and improvement mechanism. |
