A Digital Twin Framework for Structural Health Monitoring Enhanced by Physics-Based Reduced Order Modeling and Stochastic Time Series Models

Abstract

In this work, a hybrid digital twin (DT) framework for structural health monitoring (SHM) of rotorcraft structures is proposed. This framework integrates both ultrasonic-guided wave-based and vibration-based SHM schemes for tackling damage detection, identification and quantification under uncertainty. To achieve that, two novel data-driven reduced-order models (ROMs) are introduced to approximate high-frequency (ultrasonic) guided-wave dynamics and low-frequency vibration dynamics, respectively. For the high-frequency DT (HFDT), a novel scheme taking direct guided wave signals as inputs is proposed for damage detection and quantification task. In the training phase, a Convolutional AutoEncoder (CAE) is trained to achieve accurate ultrasonic wave signal reconstruction. In the online monitoring phase, the latent space representations of inputs can be extracted in an automated manner through the encoder. The extracted features are then combined with a feed forward neural network (FFNN) and Gaussian process (GP) models to provide damage level estimates in a deterministic and probabilistic manner, respectively. For the low-frequency DT (LFDT), the cornerstone of the proposed approach is the stochastic functionally pooled (FP) time series family of models. The FP model structure makes use of functional data pooling techniques for combining and optimally treating as one entity the data obtained from various structural states, and statistical techniques for model estimation. Damage detection and quantification are treated as an inverse problem and appropriate optimization techniques. To address the LFDT requirement for training data under various structural states, a Bayesian inversion physics-based finite element model (FEM) framework is postulated that enables the FEM updating and calibration via the use of experimental data. To demonstrate the applicability and evaluation of the proposed DT platform, two case studies are performed: (i) HFDT/ROM of guided-wave in Airbus H125 main rotor blade for SHM purposes and a (ii) LFDT/ROM of an Airbus H125 main rotor blade updated via vibrational data.