An Unsupervised Feature Selection Method Based on Information Entropy

Abstract

Brushless Direct Current Motor (BLDC) is a power supply unit of the Multi Rotor Unmanned Aerial Vehicle (Multi Rotor UAV). Whether it is safe and reliable directly affects the reliability level of the Multi Rotor UAV. By obtaining the BLDC operating state characteristics (including faults and failures), and accurately determining its working state, the safety, mission success and economy of the BLDC can be improved. At present, the research work on the feature extraction of operating state is mostly based on single-parameter uniaxial expansion. There may be redundant and irrelevant information between the features obtained by different feature extraction methods, which makes the BLDC running state features difficult to be accurately grasped. Therefore, this paper takes the BLDC of Multi Rotor UAV as the research object, and comprehensively utilizes feature extraction technology, unsupervised mutual information feature selection technology and kernel principal component analysis fusion technology to study multi-features, multiaxial comprehensive feature extraction method based on BLDC vibration data. This paper provides an effective method for BLDC operation status judgment, and provides data support for BLDC life-cycle health management work.