Electrostatic Monitoring Feature Extraction And Fault Severity Assessment Methods For Rolling Bearing Based On Variable Operating Condition

With the rapid development of modern industry and science and technology, machinery andequipment are also developed to large-scale, complex, high-speed, automation and intelligent. Notonly different parts of each piece of equipment should work closely with each other, but also thecooperation between different equipments should be good and stable. As an important component ofrotating machinery, the performance degradation and failure of rolling bearing affects the overallperformance and even leads to unplanned equipment downtime, resulting in economic losses and evencasualties. How to maintenance in time and to avoid “lack of maintenance” and “excess ofmaintenance” is an important issue to be addressed at this stage. Therefore, fault severity assessmentof bearing becomes particularly important, we need to identify fault severity and take appropriatemeasures in time when equipment failure.Wear site electrostatic monitoring is still in the preliminary study stage, and bearing will generatewear particles with static electricity in the process of wear. We design an electrostatic sensor, and usethis sensor to monitor the electrostatic signal of rolling bearing with different fault severity forexperiment. Then we extract time domain feature, frequency domain feature, EMD energy entropy,wavelet energy, singular spectrum entropy of electrostatic signal. At last, the paper compare threedimensionality reduction methods, PCA, LPP, OLPP, and then a multi-parameter fusion method onfault severity assessment is proposed in this paper.The main contents of this paper are as follows:(1) Described in this article the background and significance of the topic, analyze the domesticand foreign research about performance degradation assessment, signal feature extraction,dimensionality reduction, electrostatic monitoring and bearing fault severity assessment with variableoperating condition. The content is established in this paper.(2) The wear site charging mechanism is revealed in this paper, and we design the electrostaticsensor based on the electrostatic induction principle. Bearings are injected fault and rolling bearinglife experiments are carried out with different operating conditions.(3) This paper research methods of the wear site electrostatic monitoring signal featureextraction. Through experimental results, time domain feature, frequency domain feature, EMDenergy entropy, wavelet energy, singular spectrum entropy methods are compared to verify theelectrostatic sensor sensitivity on fault severity assessment. (4) For the multi-parameter features of monitoring signal are redundant in the calculation, thepaper proposes to accurately extract the inherent structure of multi-parameter feature and to reduceredundancy and inconsistency in high-dimensional data. In view of these, this paper introduces PCA,LPP and OLPP algorithms, compares the advantages and disadvantages of these algorithms structure.We respectively use PCA, LPP and OLPP to reduce dimensionality of extracted experimental datawith different fault severity. We take the first and second dimensional data mapping, compare andanalyze data. The results reveal that OLPP have better partial retention and more accurate judgmentthan PCA and LPP.(5) In view of single feature is insufficient on the ability of fault severity assessment of thebearing, a multi-parameter fusion method based on orthogonal locality preserving projection-gaussian mixture models (OLPP-GMM) is proposed in this paper. A new index, negative loglikelihood probability (NLLP), is used to characterize the degree of deviation of the test data to theGMM model, which is used as a quantitative indicator to reflect the bearing fault severity degree.Combined with experimental electrostatic monitoring data of rolling bearings in variable operatingcondition, NLLP can quantitatively distinguish fault severity of rolling bearing in variable operatingcondition.