 This paper proposes a novel deep learning architecture for estimating the performance degradation of multiple components of a high-speed train bogey. It first considers the multi-scale characteristics of the vibration signal and then employs a soft parameter-sharing mechanism to share parameters between different tasks. This allows for improved feature extraction from the signal and thus enhances the performance of the model. Experimental results show that the proposed architecture outperforms other existing models and can be used to accurately estimate the performance degradation of multiple components of a high-speed train bogey. This article was authored by Jin Xiaoren, Wei Dongjin, Yun Puiwu, and others.