Diagnosis of gradual faults in high-speed gear pairs using machine learning

Abstract

A high-speed gearbox is special equipment used in industrial and aerospace application. The rotational speeds of these gearboxes can exceed 50,000 rpm, and any defect in the gears could result in catastrophic failure under such high speed. Hence, it is imperative to detect faults as early as possible. There are number of different methods applied in fault detection with various levels of success. In this paper, a condition monitoring for a high-speed gearbox based on the machine learning method is applied to classify evolutionary multi-faults of high-speed gearbox. The types of faults considered in this paper include pitting, pitting on alternate teeth, and tooth breakage. The rotational speed of a high-speed gearbox in this paper is in excess of 10,000 rpm. Face width of the pinion is about 15.28 mm, and the pitting is about 1.5 mm. Although it is easy to recognize serious faults such as tooth breakage, it is difficult to identify indicators for early fault initiation and growth. In this paper, feature extraction of combined time and frequency domains is proposed, including time-domain amplitudes and frequency-domain peaks from the 1st to the 8th harmonic, as well as the individual harmonic sideband values based on the signal’s power spectral density (PSD). Data normalization is applied to make data uniform, and backpropagation algorithm is used to reduce errors. The proposed machine learning method resulted in the training accuracy and testing accuracy of 100% and 98.75%, respectively. The method presented in this paper can provide early detection and recognition of the tooth health status in a high-speed gearbox.