Prognostics of Induction Motor Shaft Based on Feature Importance and Least Square Support Vector Machine Regression

Autor:	Muhammad Nizam, Toni Prahasto, Achmad Widodo, Didik Djoko Susilo
Rok vydání:	2021
Předmět:	Superposition principle Noise (signal processing) Computer science Control theory Mechanical Engineering Automotive Engineering Least squares support vector machine Feature (machine learning) Prognostics Wavelet transform Regression Induction motor
Zdroj:	International Journal of Automotive and Mechanical Engineering. 18
ISSN:	2180-1606 2229-8649
Popis:	This paper aims to present a prognostic method for induction motor shafts that experience fatigue failure in the keyway area, using motor vibration signals. Preprocessing the data to eliminate noise in raw signals is done by decomposing the signal, using discrete wavelet transforms. Prognostic indicator candidates are obtained through the selection of features based on its importance, which involve the superposition of monotonicity and trendability parameters. The prognostics model is built based on the least squares support vector machine regression approach. Remaining useful life (RUL) estimates of motor shafts were performed by fitting the sum of two exponential functions to the regression results and extrapolating over time until the specified failure threshold hits. The results of the study show that the proposed method can work satisfactorily to estimate the RUL of motor shaft. The best prognostic indicator namely the RMS, can be used to predict the motor shaft RUL since 50% of the time step before the end of the motor shaft life is error bound within 20%.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::89aff9208ae1c17c5964f2ebfd6b7982 https://doi.org/10.15282/ijame.18.1.2021.06.0641 Zobrazit plný text záznamu