Guided Lamb-wave monitoring for structural ice detection: Gaussian-process-based detection under environmental...

Ice accretion on engineered structures such as aircraft and wind turbines compromises safety and operational performance, while conventional de-icing systems often entail significant energy penalties. Piezoelectric transducers offer a promising basis for low-power, dual-function ice management systems combining detection and removal in a single network. This work addresses the detection component of that framework, proposing a probabilistic baseline-monitoring methodology for ice detection using guided Lamb waves and Gaussian Process Regression (GPR). The method operates on a signal representation based on the normalized autocorrelation and logarithmic envelope of sensor signals, analysed over extended time windows to capture cumulative wavefield distortions induced by ice accretion. A GPR model trained on ice-free measurements under varying thermal conditions characterizes the baseline structural response; ice presence is assessed through a scalar ice indicator defined as the root mean square of normalized residuals, weighted by the local predictive uncertainty of the model. The methodology is validated through numerical simulations and controlled experiments on an aluminium plate instrumented with piezoelectric actuators and sensors. The numerical study confirms detection capability under noisy conditions, while the experimental campaign demonstrates robustness against temperature-induced baseline variations over the range from +18°C to -18 °C and assesses ice-detection sensitivity at -10 °C for ice masses ranging from 2 g to 50 g, achieving 100% accuracy in three sensor-mode combinations and 96% in the fourth. The results highlight a trade-off between sensitivity and thermal robustness as the signal window length increases. Overall, the study demonstrates that combining guided-wave physics with probabilistic baseline modelling provides an effective strategy for ice detection under environmental variability, while indicating the need for future validation on more realistic structural components.