Experimental evaluation for detecting bevel gear failure using univariate statistical control charts

Abstract

This study investigates the detection of crack faults in bevel gears with a new method based on univariate statistical control charts (USCCs). The focus was on determining four types of operating states for the gears: healthy operation, faulty operation with a crack length of 0.25�mm, a crack length of 0.5�mm, and a crack length of 0.75�mm. Laboratory vibration signals of the single-stage bevel gearbox system are obtained as a first step to implement the proposed approach. Next, the time-domain features each of the standard deviation (STD), root mean square (RMS), and kurtosis extracted from the vibration signal segments measured in a number of gear revolutions are used. Then, based on the normal distribution of the STD feature, it is selected among the other features as the most fault-sensitive univariate indicator to construct the USCC. Subsequently, the X-bar control chart and the exponentially weighted moving average (EWMA) chart are designed and tested. Finally, a comparison was made between the performance of the control schemes, and EWMA obtained the best performance, as it detected all crack lengths quickly and at an early stage, as a result of the specificity of EWMA in giving a weighted average of the observed samples by combining the previous and current data of the samples throughout the control process. � The Author(s), under exclusive licence to The Brazilian Society of Mechanical Sciences and Engineering 2024.