Non-linear Vibration Response Analysis of Rolling Bearing for Data Augmentation and Characterization

作     者：Rathore, Maan Singh Harsha, S. P. 

作者机构：Indian Inst Technol Roorkee Dept Mech & Ind Engn Adv Mech Vibrat Lab Roorkee 247667 Uttarakhand India 

出 版 物：《JOURNAL OF VIBRATION ENGINEERING & TECHNOLOGIES》 (J. Vib. Eng. Technol.)

年 卷 期：2023年第11卷第5期

页      面：2109-2131页

核心收录：

主　　题：Pattern classification Residual learning Generative adversarial networks Variational autoencoder Deep convolutional neural network 

摘      要：The bearing fault diagnosis is carried out using nonlinear vibration responses using a proposed framework of VAEGAN-RDCNN. The condition of imbalanced data augmentation is solved using variational auto-encoder generative adversarial network (VAEGAN). The characterization of 2D patterns (phase space trajectories) into conditions of chaos (strong and weak attractor), and quasiperiodic is implemented using residual deep convolutional neural network (RDCNN). Experimentation is carried out on a test rig for phase space reconstruction. Various metrics such as PSNR, SSIM, KLD, and histogram analysis are used to investigate the quality of generated samples. Performance assessment of the proposed method is demonstrated using VAE, and WGAN models based on metrics namely average accuracy, precision, recall, and F1-score. The comparison results obtained are 98.20%, 90.13%, and 94.56% corresponding to VAEGAN-RDCNN, VAE, and WGAN, respectively. The discriminative feature learning ability of RDCNN is also presented using t-SNE technique. Hence, the proposed data augmentation and characterization methodology works effectively and achieves superior results to conventional data synthesis-based approaches.