Self-supervised learning for remaining useful life prediction using simple triplet networks

Remaining Useful Life (RUL) prediction is critical in optimizing predictive maintenance and resource management in industrial machinery. However, existing approaches often struggle in scenarios with limited labeled data and complex multivariate time series. To address these limitations, we propose a novel self-supervised deep learning framework utilizing a triplet network architecture, which leverages unlabeled data for robust representation learning. Our model effectively captures data similarities and parameter sharing, enabling reliable training even with scarce labeled data. Comprehensive experiments conducted on the widely used NASA-CMAPSS dataset demonstrate that our method significantly outperforms state-of-the-art models across various metrics, including RUL-Score, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), particularly in data-limited scenarios. Furthermore, a detailed component- wise analysis verifies the individual contributions of each element of the proposed method. These findings have important implications for industrial stakeholders, facilitating more accurate predictive maintenance strategies, improved cost management, and sustainable operations.