For the realization of autonomous production machines, it is necessary to enable them to independently assess the condition of their components and, in a medium-term step, inform the maintenance department about it. This requires methods that allow the reliable estimation of the condition of machine components. For the realization of a predictive maintenance strategy, it is also necessary to analyze and understand the wear progression of machine tool components in order to develop prognosis models on that basis. As a commonly used and highly stressed machine tool component, the ball screw drive (BSD) plays a central role for the reliable operation of autonomous production machines. Previous work on condition monitoring of BSDs has been based on the indirect interpretation of the wear characteristics. In addition, the development of surface defects on the BSD has not been documented and investigated in image data yet, which has prevented the reliable implementation of a prognosis model based on direct wear signals. The approach presented in this work documents the entire wear development on the surface of the BSD up to the mechanical failure of the component for the first time in image data. On this basis, the classification of the damage is investigated using machine learning methods for the first time, thus creating the basis for a reliable condition monitoring system. Based on this, a wear forecasting system is investigated. The functionality of the forecasting system could be demonstrated for the first time on image data of the BSD. The investigations are extended by experiments on data efficient classification by investigating a novel approach for data efficient classification of image data. The so called SBF-Net approach is validated on images of ball screw drives as well as on image data of other technical and non-technical domains. The superiority of the approach compared to state of the art models could be demonstrated.