Fingerprinting based data abstraction technique for remaining useful life estimation in a multi-stage gearbox

Abstract Remaining Useful Life (RUL) studies generate large volumes of data which demand complex computational resources to process. Removal of anomalies from the acquired raw data is a significant paradigm which makes the condition monitoring process robust. In this investigation, a fingerprinting based data abstraction technique is proposed to identify the prominent data points from the acquired data. Run-to-failure experiments are performed on a scaled gearbox to acquire vibration signatures. Continuous Wavelet Transform (CWT) is performed and the most prominent data points (fingerprints) are abstracted from the CWT coefficients. Descriptive statistics are computed for these fingerprints. Cumulative energy is computed from the fingerprint to build health index for predicting the RUL for different speed stages of gearbox. Zone demarcation points are estimated for gearbox stages to determine individual stage’s health. A comparison between different classification algorithms yielded RNNs (long short-term memory networks) as the best in conjunction with the proposed algorithm.