Quantitative performance evaluation of a knock sensor for vibration-based fault diagnosis in a planetary gearbox

Abstract

A gearbox is one of the critical components in rotating machinery. Timely prediction of gearbox faults become of great importance to minimizing unscheduled machine downtime. Most of gearbox diagnosis studies are focused on the development of gearbox diagnosis algorithms using costly vibration sensors. However, vibration sensor cost matters in some applications, thus pushing to the use of a low-cost accelerometer, such as a knock sensor. This study uses a planetary gearbox with a knock sensor, known to be cheap and good for application (i.e., diesel engines) to detect high frequency. This study develops a quantitative sensor evaluation process for fault diagnosis. The performance of the sensor is evaluated in terms of vibration performance and fault diagnosis performance. Vibration performance evaluates the sensor's noise level through an experiment using an electro-dynamic transducer. The fault diagnosis performance is evaluated by analyzing the vibration signal obtained from the faulty gear through gearbox feature engineering to check whether the fault has been separated or not. In this study, two ideas will be proposed. First, some quantitative metrics will be used to evaluate the performance of the sensor. The vibration performance of the knock sensor is comparatively evaluated with the signal-to-noise ratio (SNR). The stable frequency range of the knock sensor for the fault diagnosis is defined based on the SNR value. From the point of view of the fault diagnosis evaluation, fault separation capability is carried out by probability of separation (PoS) and Fisher discriminant ratio (FDR). Second, a new application of the base signal of feature used for the fault diagnosis will be proposed. The filtered signal obtained by the alternative signal processing method(i.e., autoregressive-minimum entropy deconvolution (AR-MED) and spectral kurtosis (SK)) in the case of no tachometer is used as a difference signal. Analyze the frequency range and periodicity of the fault signal to determine the availability of the feature calculated with the newly applied differential signal. Two case studies are presented to demonstrate the effectiveness of the proposed sensor evaluation process and metric: 1) one-stage planetary gearbox in a wind-turbine rig tester and 2) a swing reduction gear (two-stage planetary gearbox) in an excavator. It is concluded that a knock sensor can be used for the fault diagnosis of a gearbox.