The SmartGear research project is working on a predictive maintenance concept for the sustainable use of lubricant oil. The core idea of this concept is to estimate laboratory quality measurements from real-time sensor data with a machine learning regression model. This work focuses on the estimation of the water content from sensor data. The federated learning approach is chosen to increase data security and to reduce network traffic, given the nature of the data, which is distributed across multiple sources, such as multiple machines within a factory or across different companies. This article presents the architecture of the federated learning environment. And to test the feasibility of the architecture, a dataset recorded on a laboratory test rig is split by experiments so that each client in the simulation contains a unique feature and target distribution. Exemplary the results of 4 different federated learning strategies are compared with a model trained on the same data in a centralised fashion. The centrally trained model achieves a coefficient of determination of 0.9 on the test set, while the best federated server model achieves a coefficient of determination of 0.79. The beast mean coefficient of determination of all clients on the validation set is 0.80. The investigation of a federated learning environment with real-world time series data shows accurate results for real-time condition monitoring while respecting data privacy and provides a reliable basis for predicting the remaining life of lubricant oil.