Software Health Management with Bayesian Networks (Extended Abstract)

Most modern aircraft as well as other complex machinery are equipped with diagnostics systems for their major subsystems. During operation, sensors provide important status information about a subsystem (e.g., the engine), and this information is used to detect and diagnose faults. Typically, FDDR (fault detection, diagnosis, and recovery) or IVHM (Integrated Vehicle Health Management) systems are used for this purpose. Most of these systems focus on the monitoring of a mechanical, hydraulic, or electromechanical component of the vehicle or machinery. Only recently, health management systems that monitor software have been developed (for an overview see, e.g., [1]). In this paper, we will briefly discuss our approach of using Bayesian networks for software health management (SWHM) [2–4]. The field of system health management for hardware is quite mature; this includes the use of Bayesian networks for fault detection and diagnosis [5–12]. Many industrial systems use FDDR systems (e.g., the automotive and aerospace industries). However, the health management of software has to adhere to substantially different requirements. One striking difference is that software faults usually occur instantaneously, whereas faults in hardware systems tend to develop over time (e.g., an oil leak).1 Furthermore, many software problems are caused by software-hardware interactions, which means that both the software and the hardware must be monitored. At the same time, software has features that might make system health monitoring easier and more promising in some ways. First, the introduction of software redundancy does not increase the weight of a system, while hardware redundancy clearly does. Second, software can be debugged and fixed remotely, without need for human presence at the location where the system is deployed (say, a robotic vehicle on Mars). Based on the brief discussion above, it is clear that software has several unique features, making a dedicated research and development effort worthwhile. At the same time, it is important to utilize and extend existing results from more