Multimodal Bearing Fault Classification Under Variable Conditions: A 1D CNN with Transfer Learning

TL;DR

The paper tackles bearing fault classification under variable operating conditions by fusing vibration signals with two motor phase currents in a 1D CNN using late fusion. It demonstrates a strong baseline performance, achieving $0.96$ accuracy with $L2$ regularization, and shows that transfer learning can adapt the model to new operating conditions, with a parameter-freezing strategy that preserves layers up to the first max-pool yielding the best results at higher computational cost. The study also quantifies trade-offs between accuracy and training efficiency across three TL models, highlighting Model 2 as the best balance between knowledge retention and adaptation. Practically, the approach enables accurate, adaptable fault monitoring in industrial settings with varying speeds, torques, and loads, potentially reducing downtime and maintenance costs. The work further compares favorably to prior single-modal approaches and discusses avenues for explainability, attention-based architectures, and dataset diversification for broader applicability.

Abstract

Bearings play an integral role in ensuring the reliability and efficiency of rotating machinery - reducing friction and handling critical loads. Bearing failures that constitute up to 90% of mechanical faults highlight the imperative need for reliable condition monitoring and fault detection. This study proposes a multimodal bearing fault classification approach that relies on vibration and motor phase current signals within a one-dimensional convolutional neural network (1D CNN) framework. The method fuses features from multiple signals to enhance the accuracy of fault detection. Under the baseline condition (1,500 rpm, 0.7 Nm load torque, and 1,000 N radial force), the model reaches an accuracy of 96% with addition of L2 regularization. This represents a notable improvement of 2% compared to the non-regularized model. In addition, the model demonstrates robust performance across three distinct operating conditions by employing transfer learning (TL) strategies. Among the tested TL variants, the approach that preserves parameters up to the first max-pool layer and then adjusts subsequent layers achieves the highest performance. While this approach attains excellent accuracy across varied conditions, it requires more computational time due to its greater number of trainable parameters. To address resource constraints, less computationally intensive models offer feasible trade-offs, albeit at a slight accuracy cost. Overall, this multimodal 1D CNN framework with late fusion and TL strategies lays a foundation for more accurate, adaptable, and efficient bearing fault classification in industrial environments with variable operating conditions.

Figures (8)

• Figure 1: Testbed for Paderborn dataset.
• Figure 2: Overlap Sliding Window Creation.
• Figure 3: The Proposed Multimodal 1D CNN with Late Fusion.
• Figure 5: Training accuracy and loss with number of epochs.
• Figure 6: The Confusion matrix of (a) 80/20 Split and (b) with L2 Regularization.