Onboard Prognostics and Health Management for UAM using Machine Learning Techniques

During this Phase I effort ESAero willnbsp;research and develop anbsp;prognostics and health management system (PHM) designed for electric motor inverters.nbsp;Thenbsp;finalnbsp;PHM systemnbsp;willnbsp;consist of a microcomputer or FPGA loaded with a fault detection algorithm connected tonbsp;thenbsp;sensorsnbsp;in an inverter.nbsp;This stage will focus on refiningnbsp;the algorithm.nbsp;Past ESAero PHM usednbsp;deep autoencoder, butnbsp;convolutional autoencoder,nbsp;a recurrent neural network (RNN),nbsp;long short-term memory (LSTM) units, andnbsp;gated recurrent units (GRUs)nbsp;will be investigated asnbsp;alternativesnbsp;and improvements.nbsp;ESAero will add additional software functions ofnbsp;fault classification and remaining useful life.nbsp;K-Nearest Neighbor (KNN),nbsp;support vector machines (SVMs), andnbsp;random decision forestsnbsp;are candidates for fault classification methods.nbsp;For remaining useful life, ESAero will explorenbsp;several physicnbsp;modelnbsp;based and data-driven approaches.nbsp;This effort will utilize ESAerorsquo;s large depository of X57 test data taking during prototype and acceptance testing.nbsp;PHM will ldquo;operaterdquo; on the datanbsp;and detect and predictnbsp;faults recorded in the X57nbsp;tests.nbsp;This will demonstratenbsp;PHMrsquo;s capability, performance, requirements,nbsp;andnbsp;reduce risks before entering hardware design.nbsp;Planning ahead, ESAero will investigate thenbsp;certification and regulations that will be applicable to PHM. ESAero will develop a risk mitigation plan to overcome regulatory barriers.nbsp;Thenbsp;product of this research will lead to the development of PHM requirements for UAM inverters.nbsp;In addition, ESAero willnbsp;begin prototype component selectionnbsp;which will verify currently available hardware can meet these requirements.