Aiming at the problem of limited accuracy in motor bearing fault prediction under low-frequency, low-dimensional, and imbalanced data scenarios, this paper proposes a motor bearing fault early warning model based on a temporal convolutional network that integrates multi-attention mechanisms. The model leverages these mechanisms to enhance the network's ability to model long-term dependencies and focus on critical fault features from sparse data. To validate its performance, ablation and comparative experiments were conducted on a dataset of 10,000 samples collected from a self-built test bench. The experiment results demonstrate the superiority of the proposed model. Specifically, the Multi-Attention Temporal Convolutional Network achieved a coefficient of determination of 0.9629 and a Mean Absolute Error of 0.0244, significantly outperforming the standard Temporal Convolutional Network baseline which only achieved a Mean Absolute Error of 0.1349. These results indicate the excellent performance of the proposed model in this specific data prediction scenario, providing an effective solution to practical engineering problems where high-precision sensing equipment is unavailable.
Liu et al. (2026) studied this question.