A Method for Detecting the Reducers’ Grease Failure Based on Vibration Signal Analysis

Real time detection of the lubricating grease’s effectiveness is the best method to solve wear faults, which can achieve the purpose of predictive maintenance and improve the service life of mechanical equipment. A new online detection method for gearbox lubricating grease failure is proposed to address the issues about long offline detection cycles and high labor costs associated with phased sampling analysis. Compared with online detection methods such as inductive and photoelectric, our method is difficult to detect the specific physicochemical parameters of the grease. However, it has the advantage of simple operation without touching the grease or changing the structure of the tested object. It only needs to attach the vibration sensor table to the mechanical arm reducer to measure and analyze the vibration signal. This article briefly analyzes the impact of grease failure on vibration signals, and then we introduce a novel approach that uses support vector machines along with feature extraction through wavelet packet decomposition for our detection methodology. Next transfer the commonly used spectral image recognition algorithms for temporal signal classification to achieve failed grease detection as a baseline and compare with our traditional machine learning methods. The experimental findings indicate that the recognition accuracy of both methods are above 90%, with traditional machine learning performing better on this task.