Research and application of multi-agent model for aircraft PHM

With the increasing scale of aircraft, the PHM (Prognostics and Health Management) structure of real-time sensor-based embedded aircraft software systems become more complicated, thus data-collection becomes inefficient and the system may be invalid due to varieties of failures. In this paper, the information collection and storage and subsystem fault diagnosis and forecasting methods in these real-time embedded software systems are investigated, and a new model base on multi-agent SMDP (semi-Markov decision processes) reinforcement learning is developed. The experimental results show the model reduces the rate of data loss efficiently, and increases the average utilization of CPU, and thus improves the information collection speed and detection accuracy.