RESEARCH ON ISOMAP AND DEEP BELIEF NETWORK-BASED MACHINE HEALTH STATE ASSESSMENT

Abstract

This paper integrates Isomap that transforms data features from original high-dimensional space to projected low-dimensional space into DBN model for gear fault diagnosis. The run-to-failure experiment of gearbox was conducted for validation studies, through a series of comparison experiments with the original features without space transformation, PCA and LE, HMM and BPNN, the proposed method in this paper is proved more effective for gear fault diagnosis and state assessment.

Keywords

• Feature extraction;Isomap;Feature transformation;DBN;Gear fault diagnosis