Structural health monitoring (SHM) of materials is a fundamental measure to master thedurability and the reliability of structures in service. Beyond the industrial and human issuesever increasing in terms of safety and reliability, health monitoring must cope with demandsincreasingly sophisticated. New health monitoring strategies must not only detect and identifydamage but also quantify the various phenomena involved in it. To achieve this objective, itis necessary to reach a better understanding of the damage process. Moreover, they frequentlyoccur as a result of mechanical and environmental stresses. Thus, it is essential, first, to developsignal processing methods for estimating the effects of environmental and operational conditions,in the context of the analysis of precursor events of damage mechanisms, and on theother hand, to define the damage descriptors that are the most suitable to this analysis. Thisstudy proposes signal processing methods to achieve this goal. At first, to the estimation ofexternal effects on the scattered waves in an active health control context, in a second step, tothe extraction of a damage indicator from the signals analysis of acoustic emission in a passivehealth monitoring context.In the first part of this work, four signal processing methods are proposed. These allow takinginto account the variation of environmental conditions in the structure, which in this thesis,were limited to the particular case of temperature change. Indeed, temperature changes have theeffect of altering the mechanical properties of the material and therefore the propagation velocityof ultrasonic waves. This phenomenon then causes a dilation of the acoustic signals that shouldbe estimated in order to monitor changes in temperature. Four estimators of dilation coefficientsare then studied: the intercorrelation sliding window, used as reference method, the stretchingmethod, the minimum variance estimator and the exponential transform. The first two methodshave already been validated in the literature while the latter two were developed specificallyin the context of this study. Thereafter, a statistical evaluation of the quality of estimates isconducted through Monte Carlo simulations using synthetic signals. These signals are basedon a scattered signal model taking into account the influence of temperature. A raw estimateof the computational complexity of signal processing methods also completes this evaluationphase. Finally, the experimental validation of estimation methods is performed on two types ofmaterial: First, in an aluminum plate, homogeneous medium whose characteristics are known,then, in a second step in a highly heterogeneous environment in the form of a compositeglass/epoxy plate. In these experiments, the plates are subjected to different temperatures in acontrolled thermal environment. The temperature estimates are then faced with an analyticalmodel describing the material behavior.The second part of this work concerns in situ characterization of damage mechanisms byacoustic emission in heterogeneous materials. Acoustic emission sources generate non-stationarysignals. The Hilbert-Huang transform is thus proposed for the discrimination of signals representativeof four typical sources of acoustic emission in composites: matrix cracking, debondingfiber/matrix, fiber breakage and delamination. A new time-frequency descriptor is then definedfrom the Hilbert-Huang transform and is introduced into an online classification algorithm. Amethod of unsupervised classification, based on the k-means method, is then used to discriminatethe sources of acoustic emission and the data segmentation quality is evaluated. Thesignals are recorded from blank samples, using piezoelectric sensors stuck to the surface of thematerial and sensitive samples (sensors integrated within the material)...
