Condition-based maintenance of rotating machines has become a subject of growing interest in 4.0 industry. Significant failures in industrial equipment are directly related to bearing degradation. Many techniques have been used successfully for diagnosing and forecasting bearing failures. However, accurately estimating the remaining useful life (RUL) of a bearing in operation remains a challenge. In industrial applications, the difficulty is usually related to the lack of available historical degradation signals. Therefore, in this chapter, a prognostic approach addressing the shortcomings of historical degradation data is presented. The latter bases on real-time acquired signals to build an adaptive predictive model for bearing degradation. Initially, diagnostic features associated to bearings are extracted from the available vibration signals. Then, an unsupervised DBSCAN classifier is used to detect degradation. As new degradation data become available, the extracted features are interactively ranked according to a defined selection criterion. The relevant feature is chosen as health indicator (HI). The degradation evolution and the RUL are estimated by an applied adaptive exponential degradation model, dynamically updated with each acquired sample. The applied strategy has been validated against vibration data acquired from a real wind turbine’s shaft bearings system.