Rolling Bearing Performance Degradation Assessment Based On Support Vector Data Description

Rolling bearings are common components in machinary, and their running states can partly affect the performance of equipments. Rolling bearings often experience a series of different degradation process from normal to complete failure. If the degradation degree of rolling bearings can be quantitatively assessed, the formulation of equipment maintenance strategy can be targeted. In this paper, the rolling bearings are taken as the object. The extraction of degenerated features, the establishment of assessment model and the validation of assessment results were systematically studied using the method of theory and experiment.To extract the degenerated features, the validity of lifting wavelet packet transform in fault analysis with different severity was analyzed with the inner race fault as an example. On this basis, lifting wavelet packet was combined with singular spectral entropy and symbolic entropy respectively, then the extraction of degenerated features based on lifting wavelet packet singular spectral entropy(LWPSSE) was got, and the extraction of degenerated features based on lifting wavelet packet symbolic entropy(LWPSCE) was proposed. Analytical results of experimental data indicate that LWPSSE and LWPSCE are both sensive to the inner race fault with different severity, then they can reflect the degradation state of a bearing.To establish the assessment model, support vector data description(SVDD) was introduced into bearing performance degradation assessment, and the assessment model based on SVDD was established. On this basis, performance degradation assessment method based on LWPSSE-SVDD and performance degradation assessment method based on LWPSCESVDD were proposed. Moreover, the failure threshold setting method for actual online assessment was given. Data acquisition is the basis of bearing performance degradation assessment. Therefore, the life-cycle dynamic signal acquisition system based on virtual instrument was developed using the producer-consumer structure and state machine structure in LabVIEW. And the system was used to the acquisition of bearing fault severity testing data. Then the validity of the proposed assessment methods were validated through the acquired fault severity testing data.To validate the practical application effect of the proposed assessment methods in bearing performance degradation assessment, the life-cycle testing data from University of Cincinnati was used to assess the performance degradation of a bearing, and the selection of SVDD parameters, symbol entropy were discussed. Results show that compared with other assessment methods, the two proposed assessment method can both detect the incipient fault more accurately, and they can better describe the complex deterioration trend of bearing fault, thus providing two ways for bearing performance degradation assessment. Moreover, considering the problem that the hypersphere boundary cannot update in SVDD algorithm, an adaptive SVDD algorithm was introduced into bearing performance degradation assessment. Analytical results of life-cycle data show that the proposed method with only a few offline normal samples can get the assessment result gotten from a large number of training normal samples, and the adaptive SVDD model established has more applicability.To validate the assessment results, the fault diagnosis method based on Empirical Mode Decomposition(EMD) and Hilbert envelope demodulation was introduced. Through correlation coefficient criterion and kurtosis coefficient criterion, the intrinsic mode functions(IMFs) correlated to the original signal and retaining more impulses were selected to demodulate. Results demonstrate that the used fault diagnosis method can effectively diagnose incipient fault, and the assessment results validated via the fault diagnosis method are approximately right.