The Largest Open Access Portal of Nondestructive Testing (NDT)

Damage of composite materials is a key factor in the durability in service. It is therefore essential to define the most suitable damage indicators and to develop models to estimate the Remaining Useful Lifetime (RUL) from analysis of precursor events resulting from damage. Acoustic emission is relevant to the development of the PHM (Prognostic and  Health Management) because it allows knowing the state of damage of a composite structure in real time. This work is dedicated to lifetime prediction using AE for long-term tests on CMC during static and cyclic fatigue tests. New indicators of damage have been defined, based mainly on acoustic energy analyses. In this context, an equivalent energy of AE sources is defined in order to eliminate effects of attenuation due to propagation distance. These indicators highlight critical times (around 50 % of the composite lifetime) allowing an evaluation of the remaining lifetime. A linear correlation is observed between these critical times and the lifetime duration. For a prognostic phase, the results obtained with this empirical law are compared to those obtained with a power law such as a Benioff law. Moreover, the clustering of acoustic emission, using a random forest approach, makes possible to identify the mechanism responsible for this critical time. For the static fatigue test, the critical time around 50 % correspond to the delayed failure of fibres. Moreover, the determination of the acoustic signature and characteristic time is linked to testing conditions and specimen geometries. Therefore, sizes effects are investigated with modelling works.

The objective of this work is to build a quantitative relationship between the fiber break as source of Acoustic Emission (AE) and the detected signal by unravelling the effect of each stage of the AE acquisition chain. For this purpose, an AE modelling is carried out using the Finite Element Method and then the simulation is compared to experimental results of Single Fiber Fragmentation Test (SFFT). The SFFT is used in order to produce preferential fiber break. It is carried out on dogbone shape specimens made from epoxy/amine matrix and a long carbon fiber T700 embedded in this resin. Two different types of transducer are used in order to gather information on a wider frequency bandwidth. The analysis of detected signals shows an important dependency of distance between transducer and source on the frequency content of signals. In this case, high frequency content for the signal associated to fiber breakage is not validated for all signals. For the modeling part, the entire geometry of the specimen is modelled. The geometry is subjected to a tensile loading. The fiber breakage occurs by separating the nodes forming fracture faces. The numerical out-of-plane velocities are collected on the specimen surface. The sensor is taken into account by its transfer function, which is experimentally determined by the reciprocity method. After being validated, the FE model is used to study the effects of different parameters on the signal, such as specimen geometry, the propagation medium and the location of the failure

The Acoustic Emission (AE) technique is widely used for material characterization and structural health monitoring. Once collected, the standard approach to process and interpret the potentially large amount of AE streaming data is generally based on four main phases: 1) wave picking, 2) feature extraction, 3) clustering, 4) evaluation of clusters with, if possible, correlation with AE sources. The first phase, wave picking, isolates relevant AE signals from noise within the streaming. It is based on a set of thresholds to ensure that AE signals have a sufficiently high amplitude during a certain amount of time. In the second phase, features are extracted from AE signals, where the goal is mainly to represent AE signals in the same dimensional space. Indeed, AE signals have generally different lengths which cannot be managed by standard clustering methods. Clustering aims at finding out the data structure to infer the behavior of the material during solicitations. Evaluation is finally focused on the optimization of the parameters (best subset of features and optimal number of clusters) according to some criteria. In this paper, we propose to work at the signal (AE waveform) level, in an unsupervised manner. While the supervised case has already been proposed in the past, to the author’s best of knowledge, unsupervised learning at the signal (or streaming) level for AE interpretation is a novel approach. The proposed method has three main advantages: 1) it can be used for supervised or unsupervised learning; 2) it can work directly on streaming without wave picking. This particularly simplifies the processing by reducing the number of algorithms and parameters to be used; 3) if wave picking is used to extract AE hits, then the method does not require feature extraction from AE signals. It simplifies the processing by avoiding the heavy task of feature selection. The proposed method relies on a statistical model of the temporal evolution of the streaming data. This model is preliminary built using previous data or physics-based knowledge. The model is then applied online and generates a set of similarity measures which can then be used in clustering to get the data structure. The influence of the choice of the model is alleviated by bootstrapped ensembles and clustering fusion. The method has connections with a well-known approach in SHM called anomaly detection, which will be discussed. Illustrations are provided for bio-sourced composite materials.

This paper is focused on the calibration of acoustic emission sensors by means of the reciprocity method. Therefore, the problem of the reciprocity criterion in the case of ultrasonic waves in solids is addressed. This latter criterion, upon which the reciprocity method is based, is well established and widely used in the case of waves in fluid medium (air, water etc.). However, and despite that the first papers have been published forty years ago, its application in the case of ultrasonic waves in solids (Rayleigh waves, longitudinal and transversal waves and Lamb waves) now raises a number of discussions in the community of acoustic emission. Therefore, calibration of acoustic emission sensors by means of reciprocity method starts anew to be controversial. After a review of the concept of reciprocity since its origin in electromagnetism to its application to electro-acoustic transducers, a critical analysis of previous theoretical work on the proof of the validity of the reciprocity theorem in the case of ultrasonic waves in solid will be done. An approach for calculating the coefficient of reciprocity by finite element method and an attempt to experimentally check its validity by means of impedance measurement of acoustic emission sensors will be presented. Finally, a presentation will be given on our recent works to extract an aperture function which is closer to the real behavior of sensor and to take it into account in the calibration process. This work is financially supported by the ANR, contract n° ANR-06-CIS6-01.