A dynamic data driven reliability prognosis method for structural digital twin and experimental validation

Accurate life and reliability prognosis are critical goals pursued by structural digital twin modeling. However, prognosis of in-service structures subject to uncertainties from both service load and structural characteristics. In this paper, dynamic sensor data and physical model are merged into a structural digital twin framework to cope with multiple source uncertainties and reduce errors in structural reliability prognosis, instead of pure physical model or data driven prognosis methods. Structural characteristics are represented by structural vibration modes and prognosis model parameters. Structural mode changes induced by crack growth are sensed by cross validation of strain reconstructions and employed for model form correction. Vibration loads are sensed through strain reconstruction based on the sensor data and corrected model, thus reducing load uncertainties. A dynamic Bayesian network containing uncertain model parameters is adapted to the physical system via Bayesian inference from observed crack length data, thus reducing model uncertainties. The proposed framework is validated by random vibration fatigue experiments of metallic structures. Results showed that the whole-life-fatigue crack growth prognosis agreed very well with experiments when both load and model uncertainties considered. It captured the accelerated crack growth and rapid degradation of structural reliability at the near-fracture stage, which cannot be achieved by the traditional prognosis methods considering model uncertainties only. The proposed method can drive the progress of digital twin-based structural health monitoring for safety management and risk reduction of various structures including aircraft, reusable spacecraft to optimize missions and save costs.