Verification of the Cycloidal Gear Train Fault Diagnosis Methods Based on Measured Data

This article presents an experimental verification of various fault diagnosis methods for a cycloidal gear train. Measurements were performed on a laboratory test rig equipped with a cycloidal gearbox. Seven cycloidal discs with wear defects in successive lobes were manufactured using ABS material and mounted in a steel gearbox. For each cycloidal disc, the output torque was measured and analyzed with MATLAB algorithms to identify the number of defects. Previous research employed a multibody dynamics model of a cycloidal gear train, programmed in Fortran, to propose a new fault diagnosis method. However, experimental verification did not confirm the effectiveness of using only Morris Minimum Bandwidth Wavelets for diagnostic purposes. Therefore, several alternative methods were investigated: 1) adjustment of Morris wavelet parameters to match the general frequency characteristics of the analyzed gearbox, 2) Power Spectral Density (PSD) analysis, 3) methods based on PSD roll off and centroid, 4) analysis of narrowband-to-wideband energy ratios 5) spectral flatness analysis, 6) application of spectral entropy. Among the tested approaches, only spectral entropy enabled successful classification of the number of defects. Nevertheless, the entropy values were generally low, and the differences between certain defect cases were sensitive to noise interference. The successful application of spectral entropy for defect detection demonstrates that, in cycloidal gear trains – unlike in conventional cylindrical gears - defects in cycloidal discs lead to spectral scattering. This phenomenon represents a promising feature for future fault diagnosis methods.