Data-driven predictive maintenance of technical systems has the potential to increase uptime and reduce maintenance costs. However, the collection of run-to-failure datasets, required for testing and evaluating these methods, is often expensive. To address this challenge, a benchmark model for a Li-ion battery is proposed for generating synthetic run-to-failure data inspired by real-world industrial data. Streamed and aggregated data including various types of noise and censoring can be generated. Initial survival analysis results, using Cox proportional hazards and advanced neural network models show that the generated data is complex enough for developing and testing advanced survival prediction models, and making the data publicly available the goal is also that it can contribute to the research community.