An Implementation of Prognosis with Dynamic Bayesian Networks

We present a probabilistic approach to reasoning in diagnosis and prognosis. The approach represents a mathematically rigorous way of handling uncertainty, which is often present in diagnosis, but inherent to prognosis. The approach is based on a novel form of layered dynamic Bayesian network models, which is used to perform Bayesian inference. It coherently integrates evidence on component usage, environmental conditions of operation, as well as component health history. The approach has been tested on several examples of health prognosis for electromechanical and electronic subsystems in aviation.