Rolling Element Bearing Fault Detection and Diagnosis with One-Dimensional Convolutional Neural Network

Rolling element bearings are critical components in rotating machinery, and their condition significantly influences system performance, reliability, and operational lifespan. Timely and accurate fault detection is essential to prevent unexpected failures and reduce maintenance costs. Traditional diagnostic methods often rely on manual feature extraction and shallow classifiers, which may be inadequate for capturing the complex patterns embedded in raw vibration signals. In this study, a compact one-dimensional convolutional neural network (1D CNN) is developed for automated bearing fault diagnosis using raw time-domain vibration data, eliminating the need for manual feature engineering. The model is trained and evaluated on two established benchmark datasets: the Case Western Reserve University (CWRU) dataset and the Paderborn University (PU) dataset. The CWRU data were segmented based on four distinct motor load conditions (0 HP to 3 HP), with each load scenario trained and tested independently to ensure strict separation and prevent data leakage. The CNN achieved high average test accuracies of 99.14%, 98.85%, 97.42%, and 95.14% for 0 HP, 1 HP, 2 HP, and 3 HP, respectively. On the PU dataset, known for its naturally induced faults and greater operational variability the model achieved a robust average testing accuracy of 95.63%. These results affirm the model ability to generalize across datasets and varying operating conditions. Further improvements were observed through hyperparameter tuning, particularly window length and training epochs, underscoring the importance of tailored configurations for specific datasets and load conditions. Overall, the proposed method demonstrates the effectiveness and scalability of 1D CNNs for real-time, data-driven bearing fault diagnosis, offering a reliable foundation for condition monitoring in industrial applications.

💡 Research Summary

This paper presents a compact one‑dimensional convolutional neural network (1D‑CNN) for automated bearing fault detection and diagnosis directly from raw time‑domain vibration signals, eliminating the need for manual feature extraction. Two widely used benchmark datasets were employed: the Case Western Reserve University (CWRU) dataset and the Paderborn University (PU) dataset. The CWRU data were divided into four motor‑load conditions (0 HP, 1 HP, 2 HP, 3 HP), and each load condition was trained and tested independently to avoid data leakage. For preprocessing, the continuous vibration recordings were segmented into fixed‑length overlapping windows—500 samples with a stride of 300 for CWRU and 1 200 samples with a stride of 200 for PU—producing uniformly sized samples while preserving transient fault information.

The proposed CNN architecture consists of two convolutional layers followed by a fully‑connected layer and a Softmax output. The first convolutional layer employs 64 filters with a kernel size of 100 samples, providing a wide receptive field to capture low‑frequency envelope modulations characteristic of bearing defects. The second layer uses 32 filters with a kernel size of 50 samples, focusing on higher‑frequency harmonics. Both layers use ReLU activation and max‑pooling (pool size = 4) to reduce dimensionality and computational load. The flattened feature map (2 816 dimensions) is fed into a dense layer of 100 neurons before the final Softmax classifier. The entire network contains fewer than 0.5 million trainable parameters, making it suitable for edge‑device deployment.

Training was performed with the Adam optimizer, categorical cross‑entropy loss, a batch size of 32, and 50 epochs. Hyper‑parameter tuning (learning rate, dropout rate, batch size) was conducted separately for each load condition to ensure optimal convergence and to mitigate over‑fitting.

Experimental results show that the model achieves very high accuracies on the CWRU dataset: 99.14 % (0 HP), 98.85 % (1 HP), 97.42 % (2 HP), and 95.14 % (3 HP). On the more challenging PU dataset, which includes naturally induced faults and a broader range of operating conditions, the average test accuracy reaches 95.63 %. t‑SNE visualizations confirm that the two‑layer CNN learns discriminative feature representations that clearly separate the fault classes. Comparisons with shallower models reveal reduced separability and lower accuracy, while deeper models provide marginal gains at the cost of increased training time and memory consumption.

The study demonstrates that a lightweight 1D‑CNN can reliably diagnose bearing faults across different datasets and load conditions without any handcrafted features. Its compact size and high accuracy make it a promising candidate for real‑time condition monitoring in industrial environments, especially where computational resources are limited. Future work is suggested to explore multi‑channel sensor fusion, transfer learning across different machinery, and online adaptation mechanisms to further enhance robustness in real‑world deployments.