Benchmark: Remaining Useful Life Predictor for Aircraft Equipment

We propose a predictive maintenance application as a benchmark problem for verification of neural networks (VNN). It is a deep learning based estimator of remaining useful life (RUL) of aircraft mechanical components, such as bearings. We implement the estimator as a convolutional neural network. We then provide mathematical formalizations of its non-functional requirements, such as stability and monotonicity, as properties. These properties can be used to assess the applicability and the scalability of existing VNN tools. URL. Benchmark materials, such as trained models (.onnx), examples of properties (.vnnlib), test datasets, and property generation procedures, are available at https://github.com/loonwerks/vnncomp2022 .