On September 30, 2010, the Integrated Software Health Management (ISWHM) project team completed second year NRA research of “Specific Techniques for V&V of IVHM Software,” led by principle investigator Dr. Johann Schumann. This work investigates how an ISWHM system can monitor a software system in order to detect and diagnose failures of the software while it is in operation. The research considers the approach of using a Bayesian network to model the behavior of the software under both nominal and off-nominal conditions.
Three advanced techniques that can be used to help perform V&V of ISWHM systems that use advanced modeling techniques and reasoning algorithms are described:
In the first method, system health models in the form of Bayesian networks are compiled into a simpler data structure called arithmetic circuits. The technique offers efficient processing advantages, enabling the ISWHM to be executed in an embedded environment. The utility of the arithmetic circuit models is that they support analysis that can reveal incorrect operation of the compiler used to create the executable code. The report presents results demonstrating the ability of this method to produce correctness proofs.
In the second method, the compiled software health management code is evaluated by the Ames Java PathFinder Model Checker. Using this method, it can be determined if the software can work properly in its target environment. The report illustrates using this method to automatically prove a number of safety properties of the system.
In the third approach, the software health management code is tested to verify that it meets sufficiently rigorous code coverage metrics, such as Multiple Condition/Decision Coverage (MC/DC) testing metrics. Since the ISWHM reasoning algorithms are massively data-intensive, the manual development of such test cases is possible only for codes of trivial size. In this report, it is demonstrated that Symbolic PathFinder can be used to automatically generate test cases for ISWHM reasoning algorithms.
The report provides results of preliminary experiments showing the application of these techniques to simple Bayesian Networks representative of SWHM systems and outlines the work remaining to be performed next year.
BACKGROUND: The objective of this NRA research project is to develop integrated software health management (ISWHM) techniques for the detection and diagnosis of faults in safety-critical software systems. The motivation for the project is the need for an ISWHM system that can monitor a safety-critical IVHM software system (e.g., aircraft GNC) and be at least as safe and reliable as the underlying software. False alarms and missed adverse events must be avoided. As traditional V&V cannot be used “out of the box” for systems using advanced modeling techniques and reasoning algorithms, the goal of this work is to develop advanced reasoning algorithms for ISWHM systems. All these V&V techniques must be applied — in addition to traditional review and validation — to provide sufficient evidence that the SWHM system is working safely and reliably.
NASA PROGRAM FUNDING: NASA Aeronautics Office of Aviation Safety, Integrated Vehicle Health Management Project
TEAM MEMBERS: Johann Schumann and Ole Mengshoel
COLLABORATORS: Adnan Darwiche and Arthur Choi (UCLA)
Contact: Stephen Jacklin