Development of an automatic damage detection methodology using ultrasonic piezoelectric sensors under varying temperature conditions.

This article presents an experimental study of the effects of temperature variations on ultrasonic waves and proposes a methodology to improve the robustness of damage-detection indicators under such environmental conditions. The investigation is based on laboratory tests carried out in air on specimens instrumented with embedded piezoelectric sensors and subjected to controlled thermal cycles. As a first step, the well-known time-stretching technique is applied to correct propagation delays induced by thermal expansion and the temperature dependence of wave speed. Interestingly, while this method remains effective for moderate excursions, its performance degrades at higher temperatures due to strong waveform distortion. Under such conditions, classical indicators relative velocity variation and correlation coefficient-lose reliability. To overcome this limitation, we evaluate three processing chains that combine time-stretching with (i) an autoassociative neural network (autoencoder), (ii) principal component analysis (PCA), and (iii) a support vector machine (SVM). The first two approaches extract features that are more resilient to thermal effects and provide better stability when temperature fluctuates. In addition, the squared Euclidean distance between the input and its reconstruction is used as a damage indicator, while extreme value statistics (EVS) are employed to define adaptive alarm thresholds; among the candidate tail models, the Fréchet distribution proves particularly suitable for representing the extremes of the indicator. By contrast, in our protocol, the SVM approach does not yield a significant gain. Overall, the results show that coupling time-stretching with dimensionality reduction (linear or nonlinear) and EVS-based thresholding markedly improves monitoring reliability, distinguishing healthy from damaged states with an acceptable false-alarm rate under variable environmental conditions.