Abstract—The conventional stress wave signal interpretation in heat exchanger tube inspection is human dependent. The difficulties associated with accurate defect interpretations are skills and experiences of the inspector. Hence, in present study, alternative pattern recognition approach was proposed to interpret the presence of defect in carbon steel heat exchanger tubes SA179. Several high frequency stress wave signals propagated in the tubes due to impact are captured using Acoustic Emission method. In particular, one reference tube and two defective tubes were adopted. The signals were then clustered using the feature extraction algorithms. This paper tested two feature extraction algorithms namely Principal Component Analysis (PCA) and Auto-Regressive (AR). The pattern recognition results showed that the AR algorithm is more effective in defect identification. Good comparisons with the commonly global statistical analysis demonstrate the effective application of the present approach for defect detection. Index Terms—Auto-regressive, pattern recognition, principal component analysis, stress wave. I. INTRODUCTION Heat exchanger tube leakage is expensive for plant operation. It has high potential of catastrophic failure that could results in property damage and loss of life. Hence, tubes inspection is carried out at specified periodical interval as a proactive measure in monitoring and assessing the tube health and integrity. Several different approach have been developed to gain information on the tube condition. One of successful approach is interactions of stress wave propagation with the irregularities in the tube structure. Since this approach is indirect, the evaluation of recorded signals requires basic understanding involved in the physical process to infer the tube health condition [1]. Moreover, in conventional tube inspection methods, defect is typically visually interpreted by certified personnel by comparing the captured signals from inspected tubes with the signals from calibrated tube according to either time of arrival or single parameter such as amplitude and frequency level [2]. Prior knowledge of defect signal pattern and experience are essential for a reliable interpretation. This Manuscript received March 17, 2014; revised May 10, 2014. The materials for this work was supported by Tenaga Tiub Sdn. Bhd. A.H. Zakiah thanks the Ministry of Education (Higher Education) Malaysia and Universiti Teknikal Malaysia Melaka (UTeM) for financial support of her study. A. H. Zakiah is with the Universiti Teknikal Malaysia Melaka, 76100 Melaka, Malaysia (e-mail: zakiahh@utem.edu.my). N. Jamaludin and J. Syarif are with the Universiti Kebangsaan Malaysia, 43600 Bangi, Malaysia (e-mail: nordin@eng.ukm.my, syarif@eng.ukm.my). S. Y. S. Yahya is with the Universiti Teknologi MARA, 40450 Shah Alam, Malaysia (e-mail: syedy237@salam.uitm.edu.my). requires adequate training hours to develop the expertise and competency. A large sum of resources is required for proper training, but more often being neglected because of limited fund [3]. Driven by the demand for higher performance and faster industrial production, advancing trends in automated signal interpretation are expected. One suitable approach to autonomous interpretation is the application of feature extraction based pattern recognition techniques. Previous studies are found to be successfully employed pattern recognition techniques to interpret the hidden trends of the complex stress wave signals [4]–[7]. Additionally, pattern recognition has successfully facilitated autonomous defect identification process in many structural application such as bearing, composite beam and pressure vessel [8]–[10]. Hence, the aim of this paper is to explore an alternative to signal interpretation using pattern recognition approach, which was not considered in these earlier studies. The present work compares two feature extraction algorithms, namely Principal Component Analysis (PCA) and Auto-Regressive (AR) to distinguish high frequency stress wave signal data from Vibration Impact Acoustic Emission (VIAE). VIAE was carried out on three similar dimension and material, with exceptional of presence of artificial defect in each of the tube. The findings of this study demonstrated the potential of pattern recognition approach and its suitability for signal interpretation is investigated. The current study will help the inspector to gain better insight on the stress wave signal pattern in the presence of defect and thus, improve defect assessment. II. PATTERN RECOGNITION APPROACH The objective of pattern recognition is to obtain the ideal patterns based on specified factor and thus split them into distinct groups. Generally, there are three major stages involves in pattern recognition, which are: i) preprocess of the stress wave signals, ii) clustering and iii) validation of formed clusters. The primary step in pattern recognition is preprocessing includes filtering and feature selection based on selected preprocessing algorithm. In current study, the feature extraction algorithms undertaken are PCA and AR in the attempt to speedy the signal interpretation. PCA has been successfully used in classifying the signals from different sources [11]–[13]. This algorithm determined the covariance value that represent relationship between each signal. Covariance is a measure between 2 dimensions (X and Y) is calculated using (1). If the data has more than 2 dimensions, there will be more than one covariance to be calculated. For an n-dimensional data set, the number of covariance, n cv Pattern Recognition Approach of Stress Wave Propagation in Carbon Steel Tubes for Defect Detection Zakiah A. Halim, Nordin Jamaludin, Syarif Junaidi, and Syed Yusainee Syed Yahya International Journal of Computer Theory and Engineering, Vol. 7, No. 2, April 2015 139 DOI: 10.7763/IJCTE.2015.V7.945