IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 49, NO. 5, OCTOBER2000 1083 Experimenting with Pixel-Level NDT Data Fusion Techniques Xavier E. Gros, Member, IEEE, Zheng Liu, K. Tsukada, and K. Hanasaki Abstract—Results from the fusion of images from multiple non- destructive testing (NDT) sources gathered during the inspection of composite material damaged by impact are presented and dis- cussed. The fusion of multiple images was performed at pixel level using several data fusion techniques based on Bayesian analysis, wavelet theories (Daubechie wavelet and steerable pyramid trans- form) and multi-resolution mosaic technique. Based on assessment of the resulting fused images, it appears that the multi-resolution mosaic approach is best suited to offer an estimate of the actual damaged area. Index Terms—Composite materials, data fusion, image analysis, nondestructive testing, signal processing. I. INTRODUCTION I T IS NOW common knowledge that no single nondestructive testing (NDT) method can accurately detect and quantify im- pact damage in composite materials. In the case of low-energy impact damage, the detection of such defects remains difficult, even using state-of-the-art ultrasonic testing systems. In addi- tion to more stringent requirements for the inspection of aging aerospace structures, advanced composite materials are being developed that are difficult to inspect [1]. Moreover, an increase in the use of composite materials for primary structures will lead to the development of more accurate and efficient nondestruc- tive inspection methods. Despite their limitations, conventional NDT techniques are still used. One way to compensate for the limitations of the techniques applied is to process NDT signals using data fusion techniques. NDT data fusion has already been successfully applied to combine information from multiple sen- sors gathered during the examination of welds [2], [3], and in- formation from eddy current and ultrasonic data [4]. The use of data fusion in the field of NDT is to improve defect detection and provide a more accurate measurement of defect dimensionality. In this article the problem of combining images resulting from the nondestructive examination of an impacted carbon fiber re- inforced plastic (CFRP) composite panel is addressed. In order to provide comprehensive information of the damage extent, im- ages from eddy current and infrared thermographic inspections were fused to generate a single image. The fusion operation was performed using multiple techniques such as Bayesian anal- ysis, Daubechie wavelet, steerable pyramid transform wavelet Manuscript received September 15, 1999; revised June 16, 2000. X. E. Gros is with the European Commission Joint Research Centre, Institute for Advanced Materials, 1755 ZG Petten, The Netherlands (e-mail: xgros@ieee.org). Z. Liu, K. Tsukada, and K. Hanasaki are with the Earth Resources Depart- ment, Graduate School of Engineering, Kyoto University, Kyoto 606-8501, Japan. Publisher Item Identifier S 0018-9456(00)07574-4. Fig. 1. Result of the eddy current inspection of the impacted CFRP sample and contour plot Fig. 2. Thermograph of the impacted CFRP panel and contour plot theory as well as a multi-resolution mosaic technique. The ex- perimental results are presented next, and the efficiency of each fusion technique discussed. II. NONDESTRUCTIVE TESTING OF COMPOSITE MATERIALS As previously mentioned, NDT of composite materials re- mains difficult, and more than one method is usually required to provide a more complete assessment of the structural integrity of a part [5]. In our experiments, electromagnetic and thermal techniques were used to inspect an impacted CFRP composite panel. The material tested was a square ( cm) carbon fiber hybrid fabric panel with a 2 Joules impact. A conventional eddy current system, the Hocking Phasec, was used to inspect the sample. The eddy current inspection procedure consisted of scanning the surface of the composite panel with a 2 MHz probe. The output signal was fed into a computer, and an ar- tificial map of the area inspected was generated. The result of the eddy current inspection is shown in Fig. 1. Infrared ther- mographic testing was also performed using a Nikon Laird 3A infrared camera. The sample was pre-heated up to 100 C prior to inspection. The variation of temperatures over the surface of the material as it cools down was monitored using the infrared camera. The output signal is a digital image as shown in Fig. 2. There was no registration problem as great care was taken to 0018–9456/00$10.00 © 2000 IEEE