Signal processing-based identification of pathology using ultrasonics

Resumo

Nondestructive evaluation is an emerging technology that enables to raise the safety and lifespan of nowadays structures, as well as to characterize advanced materials and biomaterials in medical science. Ultrasound is currently one of the most frequently used nondestructive inspection techniques, since it has been proven to provide effective and reliable results at relatively low cost for the estimation of the quality and structural functionality of a material, and for the characterization of its mechanical properties. Indeed, ultrasonic nondestructive evaluation is a well-established method to obtain physically relevant parameters to characterize pathologies in isotropic homogeneous materials. Pathologies are here understood as material's defects or consistency change, which altered the linear and/or nonlinear mechanical properties of materials. However, ultrasonic signals obtained from multilayered materials (composites, tissue-engineered products, biomaterials, etc.) require special care in signal interpretation (i.e. multiple and overlapping ultrasonic echoes) due to their structural complexity. For competitive pathology assessment and quality control of stratified materials, quantitative non-destructive evaluation techniques based on the use of theoretical models of the ultrasonic wave propagation have been developed to extract additional information from experimental measurements. Despite the structural complexity of those materials, relative simple models are required for efficient and real-time monitoring of their structure health. Consequently, the complexity of the signals recorded by the transducers suggests to directly compare the experimental measurements with the theoretical results, with the purpose of extracting quantitative information from damage or consistency changes. A possible approach to solve this problem is provided by the model-based estimation procedure. However, conventional model-based estimation procedure developed in the mechanical engineering community are not attractive from a practical point of view (e.g. imperfections of the acquisition system, excessive computational resources, model uncertainties, etc.). In response to those problems, some procedures have been developed in the information technology community to enhance both the reliability and the quantitative pathology-informational content of ultrasonic signals obtained from conventional nondestructive evaluation systems. Therefore, in this thesis, we intent to unify the grounds implied in both areas by developing efficient and novel methods for practical applications on layered media, facing towards the optimization of the performance of such estimation procedure. In particular, we present a general framework that relies on an advanced model-based estimation procedure to nondestructively evaluate pathologies using ultrasonics, which incorporates and adapts classical signal processing and modeling techniques to extract relevant features from the ultrasonic signals and enhance the signal interpretation. The main contributions of this dissertation concern the modeling approaches developed within this procedure to cope with the wave propagation in multilayered media. We first revisit a conventional approach known as the Transfer Matrix formalism to review the theoretical grounding for our dissertation and obtain a formulation that offers us the possibility of extending this method to more complex problems. Alternatively, signal modeling has also been proven to be an useful tool to characterize damaged materials under ultrasonic non-destructive evaluation. Consequently, we introduce a novel digital signal model for ultrasonic nondestructive evaluation of multilayered materials. This model borrows concepts from lattice filter theory, and bridges them to the physics involved in the wave-material interactions. In addition, we demonstrate that this digital model has several advantages with respect to purely physics-based models or classical spectral estimation approaches. Finally, we propose an extension of these two models to deal with the classical nonlinear constitutive behavior of such layered materials. Indeed, nonlinear mechanical properties may vary several orders of magnitude with damage, opposed to the marginal variation of linear properties. For this reason, it is proposed as an ultrasonic signature that may be more sensitive to early damage. In a further part, the development of consistent optimization strategies and the obtaining of relevant experimental observations necessary to achieve a performant model-based estimation procedure are also contemplated. In particular, we introduce the context and motivation of the employed materials, describing their potential and the challenge that they offer from a structural viewpoint, and focusing on the requirement of efficient ultrasonic nondestructive evaluation techniques to identify their damage mechanisms. In addition, we present the specimens tested and the experimental configurations used to analyze them. Finally, we provide the theoretical background for the inverse problem and system identification approaches used for characterizing the pathologies of the introduced specimens. The developed models are finally compared and validated with experimental measurements obtained from multilayered media that consist of traditional materials. Once validated, those models are tested using several applications of practical interest, including: - The detection and identification of impact and fatigue damages in carbon fiber-reinforced polymers plates. In this case, both contact and immersion measurements are performed. - The monitoring of tissue-engineered materials using a embedded ultrasonic system. This novel system is first validated on a gelation process, and then used to characterize a fibrin-agarose based construct for artificial tissue development. From a theoretical point of view, the proposed digital signal model opens new perspectives in developing models for ultrasonic nondestructive evaluation, since it represents the material as a digital filter with sparse coefficients by merging concepts both from the mechanics and the signal theory. As a consequence, this model preserves both the strengths of purely physics-based models and (heuristic) parametric signals models. From a practical point of view, this model demonstrates its ability to simulate multilayered materials. In addition, it can be successfully inserted in a model-based estimation procedure to monitor the mechanical properties of relatively complex layered materials. The presented monitoring technique achieves for the first time the reconstruction of multiple damages in carbon fiber-reinforced polymer plates from a single measurement. In contrast to other studies, the pathologies are not identified by considering the time-of-flight or the broadband ultrasound attenuation, but by reconstructing the complete waveform. Moreover, the damage multiplicity does not only appear at several locations but simultaneously in different forms. The other encouraging results on tissue-engineered materials suggest that this methodology previously developed for structural applications could be further applied in the field of biomedical engineering.