A new fault feature extraction method of rolling bearings based on the improved self-selection ICEEMDAN-permutation entropy.

The vibration signals of rolling bearings are complex and changeable, and extracting meaningful features is difficult. Currently, the commonly used empirical mode decomposition (EMD) algorithms have the problem of mode aliasing. In this paper, a new feature extraction method based on the improved complete ensemble empirical mode decomposition with adapted noise (ICEEMDAN) and permutation entropy is proposed. In this method, the ICEEMDAN algorithm is first improved and optimized to enable a self-selection function The vibration signal is then decomposed into several intrinsic modal functions using this algorithm, and the permutation entropy is extracted as the fault feature of rolling bearings, which improves the accuracy of fault classification and realizes the intelligent feature extraction of different fault states. Then, the Case Western Reserve University dataset is used for verification, and the results show that this scheme can effectively separate the vibration signal characteristics of bearings in different states, and can be used to characterize the characteristics of different bearing signals. Finally, based on the mechanical transmission system bearing experimental platform independently developed by our school, the experimental results show that compared with the unimproved ICEEMDAN algorithm, the diagnostic accuracy rate of the proposed method is 99.5%, which is increased by 6.4%, and it can be effectively used for feature extraction of rolling bearings.

Bayesian-Optimized Hybrid Kernel SVM for Rolling Bearing Fault Diagnosis.

We propose a new fault diagnosis model for rolling bearings based on a hybrid kernel support vector machine (SVM) and Bayesian optimization (BO). The model uses discrete Fourier transform (DFT) to extract fifteen features from vibration signals in the time and frequency domains of four bearing failure forms, which addresses the issue of ambiguous fault identification caused by their nonlinearity and nonstationarity. The extracted feature vectors are then divided into training and test sets as SVM inputs for fault diagnosis. To optimize the SVM, we construct a hybrid kernel SVM using a polynomial kernel function and radial basis kernel function. BO is used to optimize the extreme values of the objective function and determine their weight coefficients. We create an objective function for the Gaussian regression process of BO using training and test data as inputs, respectively. The optimized parameters are used to rebuild the SVM, which is then trained for network classification prediction. We tested the proposed diagnostic model using the bearing dataset of the Case Western Reserve University. The verification results show that the fault diagnosis accuracy is improved from 85% to 100% compared with the direct input of vibration signal into the SVM, and the effect is significant. Compared with other diagnostic models, our Bayesian-optimized hybrid kernel SVM model has the highest accuracy. In laboratory verification, we took sixty sets of sample values for each of the four failure forms measured in the experiment, and the verification process was repeated. The experimental results showed that the accuracy of the Bayesian-optimized hybrid kernel SVM reached 100%, and the accuracy of five replicates reached 96.7%. These results demonstrate the feasibility and superiority of our proposed method for fault diagnosis in rolling bearings.