Study On Feature Extraction And Assessment Method Of Rolling Element Bearing Performance Degradation

With the growth of technology and the development of industry requirement, equipments are being developed to the direction of hugeness, distribution, high speed, automation and complexity, and at the same time, their running condition is more and more rigorous. Once the key parts occur fault, the whole equipment may be destroyed even whole production efficiency will be affected and catastrophic accidents will occur. So, the key issue to be resolved is how to assess efficiently equipment's performance and make proper maintenance strategy. Generally speaking, equipment always experiences the process of from normal stateto failure, and this process is a continuous process. If the equipment performance degradation can be monitored, it would be possible to make proper maintenance strategy, so not only urgent broken can be prevented but also production efficiency can be maximized. Equipment performance degradation assessment is proposed based on the above idea. It emphasized particularly on the performance assessment through the whole lifetime, not on the fault classification at some time. So, it is different from the traditional fault diagnosis technology. Taking rolling element bearing as the research object, this paper has researched the feature extraction and assessment methods for the equipment performance degradation assessment. The main contents are as follows:(1) From the viewpoint of theoretical analysis and engineering application, this paper's background and significance of the present study are elucidated. A state of the art review is thoroughly completed, which consists of signal analysis and processing, pattern recognition and performance degradation assessment technologies. The issuses to be resoloved are summerized and the research contents of this paper are established.(2) Rolling element bearing accelerated life test, which has provided data for this research, is presented. On the on hand, eight general indices'reflection capabilities of rolling element bearing performance degradation are analyzed from the sensitivity and consistency, the results show that these indices are not sensitive to initial degradation. On the other hand, the data's credibility is proved.(3) Spectral correlation density slice power spectrum is proposed, which is based on the rolling element bearing's cyclostationary characteristic. Bearings'degradation is researched using this method, and the results show that it not only can intuitionistic reflect performance degradation, but also can reveal the dominant defect position.(4) Feature extraction methods for equipment performance degradation assessment are researched. The vibration signal of normal bearing is close to random distribution, while with the defect's deterioration, the proportion of random in vibration signal will decline. Complexity just right can reflect this change. In this paper, the complexity measurement is used as a means, multi-methods'capabilities of reflecting bearing's degradation are researched, and envelope approximate entropy and envelope sample entropy are proposed to resolve the problem of approximate entropy and sample entropy, which measure signal's complexity in the time-domain directly, can't reflect bearings'degradation well. These indices are compared through theoretical model, different defect degree experiment and whole life time experiment. The results show that complexity can reflect initial degradation more sensitive than general indices.(5) The performance degradation assessment methods respectively based on fuzzy c-means and support vector data description are proposed aiming for the problem of existing methods, and through analyzing these two methods, a new method combining them is proposed. It utilizes support vector data description to obtain the clustering center of normal state, combining failure data, the subjection to normal state is computed using fuzzy c-means, which is used as the degradation index. This hybrid method combines the merits of the two algorithms and resolves the main problems of relying on one algorithm solely. It holds some specialties such as low requirement for data's maturity, unsusceptible to parameters and results with excellent interpretability. The validity is shown by applying it to bearings'whole lifetime.