Russell Almond, Educational Testing Service
Friday, September 20
2:00 PM
Science and Technology II, Room 330B

Risk prediction and decision analysis are important in a variety of fields including engineering, medicine and management science. The program Graphical-Belief provides an interactive environment for exploring risk prediction models based on graphical belief functions. A graphical belief function is a network model (more general than a fault tree) which uses belief functions (a generalization of probability which allows both exact values and upper and lower bounds) to describe relationships between variables. Graphical-Belief can explicitly model uncertainty about key parameters like failure rates and dependencies between component and system parameters. Thus, Graphical-Belief supports a very flexible and expressive class of models for system level reliability analysis or other applications.

Graphical-Belief provides a mechanism for sharing knowledge in the form of model fragments between engineering groups and projects. For example, reliability specialists can easily share their knowledge with design engineers ensuring that reliability and safety concerns are addressed throughout the product lifecycle. The graphical belief model adapts and learns as the system proceeds through its life cycle. In particular, the graphical belief model provides a way to encapsulate valuable field experience as model fragments and incorporate those lessons in current maintenance plans and future system designs.

Part 1: Model Analysis . Although the primary purpose of building reliability models is to see if a proposed system meets safety requirements, risk assessment and decision analysis techniques are useful throughout the lifecycle of a product. The nature of the Graphical-Belief environment allows engineers to interactively perform "what if" analyses to test system alternatives and to investigate the sensitivity of the overall system to the behavior of components. In this talk we will explore uses of Graphical-Belief reliability models: (1) during product planning (setting reliability targets and allocating research and development resources), (2) during system design (exploring design alternatives, identifying critical failure paths, and assessing the sensitivity of estimates to uncertainty about key parameters and assumptions), (3) during system operation (exploring maintenance schedules and producing diagnostic systems), and (4) during evaluation and testing (model improvement and knowledge capture).

Part 2: Model Construction. Constructing the risk model is a key step in the analysis process. Typically, the analyst must integrate information from diverse sources to build the model. Graphical-Belief supports this model construction and integration with an innovative library concept based on an object system for model fragments. Engineering groups separated in space and time can share information and knowledge via this medium. Thus reliability engineers can use this environment to produce a collection of building blocks for reliability models. Design engineers can draw on that knowledge, assembling those building blocks to evaluate and improve current system designs. Maintenance technicians can use the reliability models built during system design as a diagnostic expert system and as part of maintenance scheduling. Testing and evaluation engineers can update the models and model fragments with field experience, making that knowledge immediately available to other engineering groups. The model learns and is improved as information is added throughout the system life cycle. This talk will show how Graphical-Belief uses the library to support construction of system reliability models in a concurrent, distributed engineering applications.