Abstract: With the development of sensing and information technology, massive data have been collected from machinery by various sensors for remaining useful life (RUL) prediction, which has promoted mechanical RUL prediction entering the big data era. Meanwhile, it is still a big challenge to quickly mine degradation information from massive multi-sensor data to guide RUL prediction. Most model-based methods only focus on a one-dimensional degradation data, thus cannot effectively utilize the rich data resources. Data-driven methods are highly dependent on monitoring data. They do not rely on any experimental knowledge. Thus, they suffer from the problem of poor interpretability. To fuse multi-sensor signals for RUL prediction, a model-data-fusion RUL prediction method with multi-sensor fusion is established based on the basic idea of model and data fusion. In this method, the degradation process is expressed by a generalized Wiener process model. The relationship between states and data are interpreted from the aspects of causality and correlation using multivariate measurement function and multivariate mapping function. Multi-sensor data are fused to update the model in real time using the particle filtering algorithm for RUL prediction. Under the basic framework of the proposed method, three specific models are selected to predict the RUL of milling cutters. The prediction results verify the effectiveness of the proposed method.