Predictive Modeling of a Two-Stage Gearbox towards Fault Detection

Edward J Diehl,J Tang

doi:10.1155/2016/9638325

Abstract

This paper presents a systematic approach to the modeling and analysis of a benchmark two-stage gearbox test bed to characterize gear fault signatures when processed with harmonic wavelet transform (HWT) analysis. The eventual goal of condition monitoring is to be able to interpret vibration signals from nonstationary machinery in order to identify the type and severity of gear damage. To advance towards this goal, a lumped-parameter model that can be analyzed efficiently is developed which characterizes the gearbox vibratory response at the system level. The model parameters are identified through correlated numerical and experimental investigations. The model fidelity is validated first by spectrum analysis, using constant speed experimental data, and secondly by HWT analysis, using nonstationary experimental data. Model prediction and experimental data are compared for healthy gear operation and a seeded fault gear with a missing tooth. The comparison confirms that both the frequency content and the predicted, relative response magnitudes match with physical measurements. The research demonstrates that the modeling method in combination with the HWT data analysis has the potential for facilitating successful fault detection and diagnosis for gearbox systems.

Highlights

Gears, commonly used to transmit torque, are critical components of many energy and power, aircraft, automobile, and marine/ship systems
It is identified that the combination of the proposed dynamic gearbox models (DGMs) and the harmonic wavelet transform (HWT) analysis yields effective correlation of vibration features with and without faults, showing promise for feature extraction and fault identification in nonstationary equipment
A series of investigations were performed to identify the most accurate single value for bearing stiffness. These investigations included comparing the load and displacement data supplied by the manufacturer (Rexnord MB) against published estimation techniques, parametric sensitivity of the DGM to bearing stiffness and comparison to constant speed experimental data, and hammer test frequency response analysis comparison to FEA modal results where bearing stiffness was included between shaft and casing

Summary

Introduction

Commonly used to transmit torque, are critical components of many energy and power, aircraft, automobile, and marine/ship systems. Data-based condition monitoring approaches of gearboxes, process vibration signals which contain a complex mixture of harmonic and subharmonic components of mechanically induced motion from gear meshing, bearing rolling, and driver and load motion and random noise. Condition monitoring of rotating machinery has historically been performed on machines operating at constant speeds by processing vibration data with the Fast Fourier Transform (FFT) into the frequency domain. We use a benchmark two-stage gearbox test bed with variable speed controller to gather nonstationary experimental data and compare to DGM results. To facilitate effective data processing especially under nonstationary conditions, the harmonic wavelet transform is employed to analyze the features of the vibration responses. It is identified that the combination of the proposed DGM and the HWT analysis yields effective correlation of vibration features with and without faults, showing promise for feature extraction and fault identification in nonstationary equipment

Test Bed Structure and Experimental Setup

Correlation of Predictive Modeling and Experimental Investigation

Concluding Remarks