Fusing Physics-based and Deep Learning Models for Prognostics

Data-driven and physics-based models for remaining useful lifetime (RUL) prediction typically suffer from two major challenges that limit their applicability to complex real-world domains: (a) high complexity or incompleteness of physics-based models and (b) limited representativeness of the training dataset for data-driven models. The work proposed in the paper describes a novel hybrid framework for fusing the information from physics-based performance models with deep learning algorithms for prognostics of complex safety critical systems under real-world scenarios. In this framework, we use physics-based performance models to infer unobservable model parameters related to the system's components health solving a calibration problem a.k.a. an inverse problem. These parameters are subsequently combined with sensor readings and used as input to a deep neural network to generate a data-driven prognostics model. The performance of the hybrid framework is evaluated on a synthetic dataset comprising run-to-failure degradation trajectories from a small fleet of nine turbofan engines under a wide range of real flight conditions. The dataset was generated with the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dynamical model developed at NASA. The performance of the approach is compared to an alternative data-driven approach where only sensor data are used as input to the deep neural network. A scenario of incomplete representation of the test degradation conditions in the training dataset is evaluated. The experimental results show that the hybrid framework outperforms the alternative pure data-driven approach by extending the prediction horizon by nearly 100% and decreasing the prediction error by 34% based on the applied metrics.