Data-driven Modelling for Guided Wave Structural Health Monitoring and Prognostic Health Management on Aeronautical Structures

Health Monitoring (GWSHM) and Prognostic Health Management (PHM) on aeronautical structures. Main attention is paid to damage detection and localisation on Carbon Fiber Reinforced Polymer (CFRP) panels together with the Remaining Useful Life (RUL) prediction of aircraft turbofans. A baseline-dependent GWSHM approach on CFRP plates is proposed integrating the statistical autoregressive (AR) modelling and signal compression techniques. An adaptive block-wise down-sampling method is designed to compress guided wave signals and compared to other signal compression techniques, and the compressed measurements are processed by AR models to extract damage-sensitive features. Further validations are implemented on both quasi-isotropic and anisotropic CFRP coupons. Various variables, including the AR model parameters, number of tone burst cycles and signal compression ratios, are investigated regarding their effects on the damage detection and localisation. Baseline-free GWSHM is investigated within the aeronautical composite cases. Existing baseline-free techniques are reviewed, with several of them experimentally compared considering material anisotropy, temperature variations, multiple damages, boundary reflections and structural complexities, e.g., stiffeners. Furthermore, a novel data-driven baseline-free technique is presented based on unsupervised shapelets and shift-invariant dictionary learning. The proposed method can detect and localise potential damages in complex composite structures, e.g., stiffened panels, when integrated with the building block philosophy. The effectiveness and robustness of the proposed method is validated through experimental investigations and compared with several existing baseline-free techniques for damage detection and localisation under varying temperatures. During the transition from GWSHM to PHM, the RUL prediction of aircraft engines is implemented by a proposed hybrid data-driven approach which is able to model degradation behaviour of turbofans under varying environmental and operational conditions (EOCs) while yielding satisfactory RUL estimates simultaneously.