Sophisticated modeling and analysis methods are being developed in academic and industrial research labs for reliability engineering and other domains. The evaluation and evolution of such methods based on use in practice is critical to research progress, but few such methods see widespread use. A critical impediment to disseminating new methods is the inability to produce, at a reasonable cost, supporting software tools that have the usability and dependability characteristics that industrial users require, evolvability to accommodate software change as the underlying analysis methods are refined AR?) enhanced. The difficulty of software development thus emerges as a key impediment to advances in engineering modeling and analysis. Today, producing sophisticated software tools is costly and difficult, even for capable software developers. One problem is that when common design-methods, such as object-oriented programming, are used to build such tools, the results are often large, complex, and thus costly programs. Tools on the order of a million lines of code are typical, with much of the code devoted to tool interoperability, human-computer interface, other issues not directly related to modeling and analysis. Making matters worse, domain experts, such as reliability engineering researchers, often lack skills in modern software development, while software engineers and researchers lack knowledge of the application domains. All too often the results of tool-development efforts today are thus costly, hard to use, not dependable, essentially unmaintainable. This paper presents an approach to tool development that attacks these problems. Progress requires synergistic, interdisciplinary collaborations between application-domain and software-engineering researchers. We have pursued such an approach in developing Galileo: a fault tree modeling and analysis tool. These innovations are described in 2 dimensions 1) The Galileo core reliability modeling and analysts function. 2) Our work on software engineering for high-quality, low-cost modeling and analysis tools. In the reliability engineering domain, Galileo supports precise, modular, dynamic fault-tree analysis using techniques developed primarily by Dugan and her colleagues. This approach addresses the problem that a single analysis technique seldom applies to an entire system. A good reliability engineer uses different techniques to analyze different parts of a system decomposing a complex model into smaller pieces, applying different analysis techniques to submodels, integrating partial results into a system-level result. Manually decomposing systems into parts, developing submodels, analyzing them with different tools and techniques, and integrating the partial results is tedious and error prone at best. By contrast, Galileo- automatically detects independent(1) sub-trees; translates them into appropriate submodels based on Markov chains, Boolean decision diagrams, and other formalisms, analyzes the submodels; integrates the results. Galileo supports precise analysis while exploiting modularity for scalability in solving problems that require time and space that is exponential in the number of basic events in the worst-case. This software engineering approach centers on the component-based design techniques of Sullivan and his colleagues. A key element of the approach is the use of mass-market software packages as large components, viz, package-oriented programming. It achieves at low cost an effective human-computer interface, tool interoperability, considerable dependability for the function delivered. Low-cost means that the effort involves a small handful of graduate and undergraduate students and faculty. Sullivan's mediator-based design approach is also used at several scales to support an integrated, multi-view environment in which it is possible to edit fault trees in either textual or graphical form, while fostering dependability and evolvability. To help validate this modeling approach and to verify its implementation, both natural-language and formal specifications are being developed for the fault-tree gates and their interactions. Galileo has been evaluated against commercially available fault tree analysis tools. The results highlight the need for fidelity in analysis. Testing two tools popular in the reliability engineering community revealed the same algorithmic error in both, despite their claimed ability to provide exact solutions. At the intersection of software and reliability engineering, the redundancy inherent in the use of multiple analysis techniques in Galileo is used as an aid to testing this software. Galileo has been acquired by hundreds of sites. We are now building an enhanced version with NASA Langley Research Center.
