GASF-STC: A Gramian Angular Summation Field–Based Spatio-Temporal Contrastive Network for Remaining Useful Life Prediction

Remaining useful life (RUL) prediction is a key task in predictive maintenance, providing crucial insights into the health status and reliability of equipment. This paper proposes a novel RUL prediction method, GASF-STC, which integrates Gramian angular summation field (GASF), convolutional neural networks (CNNs), recurrent neural networks (RNNs), and contrastive learning techniques. The sensor data is transformed into both temporal features and two-dimensional images to capture global temporal characteristics and local spatial correlations, followed by feature processing using CNN and RNN. Finally, contrastive learning is employed for feature fusion, ensuring a balanced contribution from both features. The proposed method is evaluated on the C-MAPSS turbofan engine benchmark, and its generalization capability is further validated on the MIT Battery Aging Dataset. Experimental results demonstrate that GASF-STC not only outperforms traditional models on the engine dataset but also maintains robust prediction performance under a completely different degradation mechanism. The results highlight the effectiveness of spatiotemporal feature fusion, significantly improving prediction accuracy and stability, making GASF-STC a promising RUL prediction method for predictive maintenance applications.