Keywords: wavelet packet transform, deep learning, vibration images, condition monitoring.

In the era of industry 4.0, railways systems are becoming more advanced through employing modern methods like Deep Learning (DL) algorithms. DL algorithms have been able to accomplish excellent outcomes in condition monitoring of railway systems. Thus, railway industry has been adopting it for various processes for safe and uninterrupted operation. As the safe operation of traction motors rely on normal operation of bearing, thus they require timely detection and identification of various faults. In this paper, a comparative study on bearing fault identification using UNET method based on time domain vibration images and wavelet transformed vibration images is presented. The three-step method involves implementation of UNET model, extracting Wavelet Packet Transform (WPT) features from raw vibration data, and transforming the WPT data to gray-scale vibration images. The time-series vibration data is transformed into 32 × 32 × 1 WPT vibration images. The comparative analysis of UNET with time domain vibration images (TVI-UNET) and WPT based vibration images depicted best performance on later one. The UNET model with WPT based vibration images (WPT-UNET) achieved F1-score and MIoU of 99.3% and 50.26%, respectively. The proposed method demonstrated robust response and superior performance than the UNET with time domain vibration images.

download the full-text of this paper (PDF, 7 pages, 535 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description