Assessing the effectiveness of data-driven time-domain condition indicators in predicting the progression of surface distress under rolling contact

Abstract

Condition monitoring of machine health via analysis of vibration, acoustic and other signals offers an important tool for reducing the machine downtime and maintenance costs. The key aspect in this process is the ability to relate features derived from the recorded sensor signals with the physical condition of the monitored asset in real time. This paper uses simple machine learning techniques to examine the ability of specific time-domain features obtained from vibration signals to predict the progression of surface distress in lubricated, rolling-sliding contacts, such as those found in rolling bearings and gears. Controlled experiments were performed on a triple-disc rolling contact fatigue rig using seeded-fault roller specimens where micropitting damage was generated and its progression directly observed over millions of contact cycles. Vibration signals were recorded throughout the experiments. Features known as condition indicators were then extracted from the recorded time-domain signals and their evolution related to the observed physical state of the associated specimens using simple machine learning techniques. Five time-domain condition indicators were examined, peak-to-peak, root-mean-square, kurtosis, crest factor and skewness, three of which were found not to be redundant. First, a classification model using KNN nearest neighbor was built with the three informative condition indicators as training data. The cross-validation results indicated that this classifier was able to predict the presence of micropitting damage with a relatively high precision and a low rate of false positives. Secondly, a k-means clustering analysis was performed to measure the significance of each condition indicator by leveraging patterns. The peak-to-peak condition indicator was found to be a good predictor for progression of micropitting damage. In addition, this indicator was able to distinguish between micropitting and pitting failure modes with a high success rate. Finally, the condition indicator response was correlated with the predicted damage state of the test specimen obtained through an existing physics-based surface distress model in order to illustrate the potential of hybrid models for improved prognostics of damage progression in rolling-sliding tribological contacts.

Fault Diagnosis for Abnormal Wear of Rolling Element Bearing Fusing Oil Debris Monitoring

“…Based on resonance theory, SKF proposed the shock pulse method (SPM) for the fault diagnosis of bearings [ 23 , 24 ]. However, for early bearing wear, the vibration signal of the bearing lacks obvious fault characteristics [ 25 , 26 ]. The bearing wear cannot be detected effectively only by the vibration signal [ 27 ].…”

Algorithmic performance constraints for wind turbine condition monitoring via convolutional sparse coding with dictionary learning

“…Kurtosis is a measurement of the peakedness of a distribution and this value tends to increase with defect severity. 15 These indicators are significant given the hints they can provide concerning a possible high-force related issue present in the bearing. Frequency domain analysis is the most widely used approach in the detection of faults from vibration signals.…”

Research on Key Control Technology of Intelligent Rolling Contact Fatigue Test Facility

“…In recent years, related research on fatigue testing technology has attracted more and more attention from scholars around the world [2][3][4][5][6]. Hua et al developed a special test device to test and evaluate the fatigue of artificial hip joint effectively [7]; Tan et al designed energy-saving transmission fatigue tester system using common DC bus energy recovery technology that can adapt to high-power, high-speed test environment, and energy-saving effect up to 90% [8]; Bai et al proposed a new method of resonant fatigue test using 2D high-frequency rotary valve control hydraulic cylinder, which can improve the frequency of electrohydraulic fatigue test effectively [9]; Li et al introduced a new type of testing machine that can perform rolling fatigue test on the bearing ball under large loads and high speed, and the test system monitors various signals at the same time, which can analyze and determine the fatigue status [10]; and Schneider et al developed a large-scale beam resonant contact fatigue testing machine.…”