This paper proposes a novel data fusion method based on deep learning for degradation modeling and prognostics. The method constructs a composite health index (HI) through the fusion of multiple sensor signals, addressing the limitations of linear data-fusion models in capturing nonlinear relations in complex systems.