Real-Time Vibration-Based Bearing Fault Diagnosis Under Time-Varying Speed Conditions

TL;DR

The authors tackle real-time bearing fault diagnosis under time-varying speeds and noisy conditions by proposing a lightweight CNN and a Fisher-based spectral separability analysis (SSA). Using KAIST data with four fault classes, they demonstrate that the CNN achieves up to 15.8 percentage points higher accuracy than the prior state-of-the-art while maintaining real-time inference (~$20.2$ ms). SSA provides spectral interpretability, revealing discriminative bands in clean data and explaining performance degradation under noise, motivating joint time-frequency representations. The work highlights practical robustness and deployability for edge-enabled bearing diagnostics and offers a framework for explaining deep learning performance in non-stationary, noisy environments.

Abstract

Detection of rolling-element bearing faults is crucial for implementing proactive maintenance strategies and for minimizing the economic and operational consequences of unexpected failures. However, many existing techniques are developed and tested under strictly controlled conditions, limiting their adaptability to the diverse and dynamic settings encountered in practical applications. This paper presents an efficient real-time convolutional neural network (CNN) for diagnosing multiple bearing faults under various noise levels and time-varying rotational speeds. Additionally, we propose a novel Fisher-based spectral separability analysis (SSA) method to elucidate the effectiveness of the designed CNN model. We conducted experiments on both healthy bearings and bearings afflicted with inner race, outer race, and roller ball faults. The experimental results show the superiority of our model over the current state-of-the-art approach in three folds: it achieves substantial accuracy gains of up to 15.8%, it is robust to noise with high performance across various signal-to-noise ratios, and it runs in real-time with processing durations five times less than acquisition. Additionally, by using the proposed SSA technique, we offer insights into the model's performance and underscore its effectiveness in tackling real-world challenges.

Figures (7)

• Figure 1: A rolling element bearing example. The main parts outer race, inner race, balls, and cage are illustrated. Redrawn from https://upload.wikimedia.org/wikipedia/commons/f/f6/HARP_bearing.jpg.
• Figure 2: Frequency analysis of the motor's rotational speed. The power spectral density was estimated by the nonuniform Fourier transform (NUFT) because of the speed's inconsistent acquisition rate. Frequency scaling for the NUFT was performed using a sampling frequency of 12.5 Hz; the reciprocal of the shortest acquisition time. The results were averaged across the different classes and dataset files and plotted with their 68.3% confidence intervals.
• Figure 3: Comparing the testing macro-averaged accuracy of the CNN network to the current state-of-the-art technique, the PIResNet Ni2023, using both clean and noisy vibration signals. The accuracy curves are plotted at varying SNR levels along with their 95% confidence intervals ($\mu\pm2\sigma$).
• Figure 4: The testing confusion matrices averaged over the five data splits. The results depict the performance when using (a) clean, and (b)-(d) noisy vibration signals at the following SNR levels: (b) 5 dB, (c) 0 dB, and (d) -5 dB.
• Figure 5: The model's testing t-SNE from the first data split. The results depict the t-SNE when using (a) clean, and (b)-(d) noisy data at the following SNR levels: (b) 5 dB, (c) 0 dB, and (d) -5 dB. We used the model's last layer features as input to the t-SNE algorithm.