Dynamic Bayesian Network Based Fault Prediction of High Power Switching-Mode Amplifiers

Most switching-mode power amplifiers are made up of multiple modules running in parallel redundant operation mode. The complex structure and operation mode make fault diagnosis and prediction difficult to design and develop. Effective fault prediction is required to improve operational reliability of amplifier systems, by achieving pre-maintenance of the system to avoid major accidents and reduce maintenance costs. This paper uses the special structure of dynamic decision networks and their ability to process time series data to perform probabilistic reasoning and to build a dynamic Bayesian network model. A numerical simulation model for the state variables that characterize the system fault is also established. This is to generate a number of simulation data to describe the change in fault state variables under different working conditions during different running modes. Using these data, the feasibility of the fault prediction model of the switch mode power amplifier is illustrated and verified.