Ultrasonic NDE of composite material structures using wavelet coef®cients S. Legendre * , J. Goyette, D. Massicotte Hydrogen Research Institute, Universite  du Que Âbec a Á Trois-Rivie Áres, C.P. 500, Trois-Rivie Áres, Que Âbec, Canada G9A 5H7 Received 20 November 1999; received in revised form 2 May 2000; accepted 16 May 2000 Abstract A wavelet-based method is proposed to perform the analysis of NDE ultrasonic signals received during the inspection of reinforced composite materials. The non-homogenous nature of such materials induces a very high level of structural noise which greatly complicates the interpretation of the NDE signals. By combining the time domain and the classical Fourier analysis, the wavelet transform provides simultaneously spectral representation and temporal order of the signal decomposition components. To construct a C-scan image from the wavelet transform of the A-scan signals, we propose a selection process of the wavelet coef®cients, followed by an interpretation procedure based on a windowing process in the time±frequency domain. The proposed NDE method is tested on cryogenic glass/epoxy hydrogen reservoir samples. q 2001 Elsevier Science Ltd. All rights reserved. Keywords: Composite material; Signal processing; Ultrasonic evaluation; Wavelet transform 1. Introduction In its liquid form, hydrogen is often stored in cryogenic composite material reservoirs. These reservoirs can be subjected to high mechanical and thermal constraints that could create ¯aws having a potential of weakening the reser- voir structure [1]. Non-destructive evaluation (NDE) of conventional reservoirs is often made through a C-scan analysis using ultrasonic bulk-waves as a probe; the signal coupling between the ultrasonic transducer and the piece under inspection is made by immersing it into a water tank. Due to the non-homogenous nature of reinforced composite materials, new inspection techniques need to be developed in order to be able to perform the analysis of NDE ultrasonic signals that can be highly complex. For example, some peak overlaps are created by scattering phenomenon and the resulting background noise, also called structural noise, is high enough to bury the meaningful re¯ection echoes in a signal having a large number of frequency components. Consequently, the process of ¯aw detection is not simple and a well-adapted kind of signal analysis is required to interpret the ultrasonic inspection signals. Since, in this case a meaningful Fourier transform analysis could be dif®cult to perform due to a possible shift- ing of the peak frequency, some processing methods based on the principle of quasi-frequency diversity (QFD), such as the well-known split spectrum processing (SSP) and its de- rivatives have been proposed [2±4]. These ®ltering methods allow a signi®cant reduction of the structural noise in signals by applying a process of noise decorrelation that permits a coherent information detection in the time domain. Some new kinds of processing methods based on a time± frequency analysis, such as the wavelet transform, have been proposed recently to treat similar problem [5±8]. By combining the time domain and the classical Fourier analy- sis, these methods provide simultaneous spectral represen- tation and temporal order of the signal decomposition components [9]. The main advantage of the wavelet-trans- form-based analysis method is that they are suitable for peak detection problems in highly noisy environment [8]. The wavelet transform produces different representations Ð wavelet coef®cient sets or wavelet levels Ð of an analyzed signal, separating the different frequency components while keeping its temporal representation. In this paper, we propose a method of A-scan ultrasonic signal analysis based on a selection process of coef®cients, also called feature extraction, provided by the wavelet transform. Once selected and interpreted these features allow us to construct a C-scan image of the inspected structure. The proposed NDE method is tested on cryogenic glass/epoxy NDT&E International 34 (2001) 31±37 0963-8695/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved. PII: S0963-8695(00)00029-3 www.elsevier.com/locate/ndteint * Corresponding author. Tel.: 11-819-376-5071; fax: 11-819-376-5219. E-mail addresses: sylvie_legendre@uqtr.uquebec.ca (S. Legendre), jacques_goyette@uqtr.uquebec.ca (J. Goyette), daniel_massicotte@uqtr. uquebec.ca (D. Massicotte).