Use of Machine Learning to Define Optimum HUMS Acquisition Strategy

Health and Usage Monitoring Systems (HUMS) measure vibrations levels and compute Vibration Condition Indicators to monitor Helicopters dynamic systems. Modern avionics record continuously a wealth of flight parameters. The influence of flight parameters on Vibration Condition Indicators was assessed using machine learning methods. Machine learning was used to derive Vibration Condition Indicators from flight parameters only. This derivation yielded quantitatively the influence of flight parameters. An illustration of this methodology is presented on main rotor speed. Variable rotor speed contributes to better acoustic performance but is a daunting challenge for HUMS vibration monitoring. Several Vibration Condition Indicators were modelled and their dependency to rotor speed was determined by machine learning parameter weight output. This allowed to optimize the acquisition parameters, the filtering of Vibration Condition Indicators and eventually to classify and choose the most relevant one based on machine learning and extensive replay of fleet data.