Comparison of Two ICA Algorithms in BSS Applied to Non-destructive Vibratory Tests Juan-Jos´ e Gonz´alez de-la-Rosa 1 , Carlos G. Puntonet 2 , Rosa Piotrkowski, Antonio Moreno, and Juan-Manuel G´ orriz 1 University of C´adiz, Research Group TIC168 - Computational Electronics Instrumentation and Physics Engineering, EPSA, Av. Ram´on Puyol S/N. 11202, Algeciras-C´adiz, Spain juanjose.delarosa@uca.es 2 University of Granada, Department of Architecture and Computers Technology, ESII, C/Periodista Daniel Saucedo. 18071, Granada, Spain carlos@atc.ugr.es Abstract. Two independent component analysis (ICA) algorithms are applied for blind source separation (BSS) in a synthetic, multi-sensor situation, within a non-destructive pipeline test. CumICA is based in the computation of the cross-cumulants of the mixtures and needs the aid of a digital high-pass filter to achieve the same SNR (up to -40 dB) as Fast-ICA. Acoustic Emission (AE) sequences were acquired by a wide frequency range transducer (100-800 kHz) and digitalized by a 2.5 MHz, 8-bit ADC. Four common sources in AE testing are linearly mixed, involving real AE sequences, impulses and parasitic signals modelling human activity. 1 Introduction AE and vibratory signal processing usually deals with separation of multiple events which sequentially occur in several measurement points during a non- destructive test. In most situations, the test involves the study of the behavior of secondary events, or reflections, resulting from an excitation (the main event). These echoes carry information related with the medium through which they propagate, as well as surfaces where they reflect [1]. But, in almost every measurement scenario, an acquired sequence contains information regarding not only the AE under study, but also additive noise processes (mainly from the measurement equipment) and other parasitic signals, e.g. originated by human activity or machinery vibrations. As a consequence, in non-favorable SNR cases, BSS should be accomplished before characterization [2], in order to obtain the most reliable fingerprint of the AE event. The purpose of this paper is twofold. First we show how two ICA algorithms separate the true signal from the parasitic ones taking a multi-sensor array of inputs (SNR=−40). Secondly, we compare performances of Cum-ICA and Fast- ICA, resulting that Cum-ICA needs the aid of a post high-pass filter to achieve the same SNR as Fast-ICA. This comparison could be interesting for a future implementation of the code in an automatic analysis system. T.-D. Wang et al. (Eds.): SEAL 2006, LNCS 4247, pp. 750–755, 2006. c  Springer-Verlag Berlin Heidelberg 2006