The quantification of impact damage distribution in composite laminates by analysis of X-ray computed tomograms

11 One of the great strengths of X-ray computed tomogr aphy over conventional inspection 12 methods (ultrasound, thermography, radiography) is that it can image damage in 3D. 13 However for curved or deformed composite panels it can be difficult to automatically ascribe 14 the damage to specific plies or inter-ply interface s. An X-ray computed tomography (CT) 15 data processing methodology is developed to extract he through-thickness distribution of 16 damage in curved or deformed composite panels. The method is applied to [(0°/90°) 2]s carbon 17 fibre reinforced polymer (CFRP) panels subjected lo w velocity impact damage (5 J up to 18 20 J) providing 3D ply-by-ply damage visualisation a d analysis. Our distance transform 19 approach allows slices to be taken that approximate ly follow the composite curvature 20 allowing the impact damage to be separated, visuali sed and quantified in 3D on a ply-by-ply 21 basis. In this way the interply delaminations have been mapped, showing characteristic 22 peanut shaped delaminations with the major axis ori ented with the fibres in the ply below the 23 interface. This registry to the profile of the pane l constitutes a significant improvement in our 24 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT ability to characterise impact damage in composite laminates and extract relevant 25 measurements from X-ray CT datasets. 26

can lead to loss of strength and stiffness. Ultimately, the load-bearing capability can be 51 significantly reduced in both tension and compression, and catastrophic failure can occur 52 under relatively low applied loads. As a result there is a concerted research effort to improve 53 the damage resistance and tolerance of these materials. 54 A wide range of characterisation techniques, both destructive and non-destructive, can be 55 employed to improve our understanding of the damage mechanisms occurring in CFRP 56 panels. As regards destructive methods, thermal de-plying [9] or more commonly sectioning 57 followed by optical and/or scanning electron microscopy [10,11] are employed for 58 determining the cause of failure, as well as establishing the area of crack initiation [12]. The increasing number of studies have been performed using CT to characterise composite 75 materials, mainly for the assessment of porosity and defects [20][21][22][23][24]. The use of CT for the 76 study of impact damage in composite structures has received less attention [18,[25][26][27]. 77 Accurate quantification of damage in carbon fibre composites to date has been limited partly 78 by the difficulty of obtaining sufficient defect contrast (related to low contrast and 79 insufficient spatial resolution) [26,28] and partly by the lack of sufficiently sophisticated 80 image analysis procedures to segment and quantify the resulting low contrast images and 81 complex geometry of the damage. 82 The study by McCombe et al. [18] was the first to report on the through-thickness 83 distribution of the damage, i.e. the mapping of the damage as a function of depth; in their 84 case through a self-healing laminate. While damage was quantified for every slice in the CT 85 volume a significant shortcoming of this methodology is that it neglects the curvature of the 86 panel. This means that a slice in the CT volume may contain information from more than one 87 ply. Since composite panels can be non-planar by design, or as a result of loading, a method 88 that can take account of this is needed so that the damage can be located with respect to the and c) an XY plane. (Air appears black, matrix and carbon fibers grey, and glass filament yarns white) The paper describes a novel data analysis methodology able to quantify damage in non-planar 99 as well as flat composite laminates. The objective is to obtain the through-thickness damage 100 distribution in three dimensions, thereby allowing inter-ply and intra-ply cracking to be segmented and assessed qualitatively and quantitatively at the ply-by-ply level. This is 102 needed to improve our understanding of the type (failure mode) and extent of the impact All test specimens used in this study were manufactured using vacuum assisted resin film

Impact testing 131
In order to generate damage in the composite coupons, a drop weight impact testing machine 132 was used to deliver a low velocity impact. Impact testing was performed using an Instron  The impacted specimens were scanned on a Nikon Metrology 225/320 kV Custom Bay [33].

161
The system was equipped with a 225 kV static multi-metal anode source (Cu, Mo, Ag, and   The distance of each voxel beneath the reference surface was then determined. Rather than 220 fit a curve to the reference surface and measure the distance of each voxel from this curve, 221 the distance transform defined by Gustavson and Strand [36] was used. This is defined as 222 "the mapping from each foreground picture element (i.e. each composite/damage pixel) to 223 its distance from the closest background pixel (i.e. the air above the reference surface)", was 224 employed to calculate the Euclidean distance from each voxel to the impacted face (Fig. 7c).

225
The output from this operation (Fig. 7c) is then combined with the damage label (Fig. 7b) in 226 order to create the damage distance transform ( fig. 7d), relative to the impact face. In this 227 manner each damage voxel is assigned a distance from the reference surface.   obtained for the specimen tested at 5 J are plotted in Figure 9.

254
The binder yarn profile comprises 9 peaks. The first (corresponding to the top surface) and 255 last (corresponding to the bottom surface) peak in Figure 9 are of much lower intensity than 256 those corresponding to the 7 internal interfaces.

257
The corresponding damage profile in Figure 9 comprises six peaks of varying intensity. The

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
A comparison between the damage profiles obtained by the standard methodology, i.e. 268 orthoslice-by-orthoslice, and using the distance transform, is presented in Figure 10. In both 269 cases 6 peaks with increasing maxima, except that of peak 5, are obtained. The main 270 differences are the range of distances from the impact face and the peak shapes and maxima.

271
The mean panel thickness measured after impact by CT was 2.97 mm ± 0.08 mm with a

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
The through thickness damage distribution, i.e. the evolution of the damage volume as a 281 function of the distance from the impact face, has been obtained for all four impact energies 282 tested (5 J to 20 J) using the distance transform approach presented in section 2.6. The results 283 presented in Figure 11 demonstrate that for all energies, the damage profiles comprise six 284 peaks corresponding to the six ply interfaces (alternating 0°/90° and 90°/0°). Larger length 285 splits and hence induced delaminations develop in the lower part of the plate since it is loaded 286 in tension. As the impact energy increases, the peak intensity increases and the peaks broaden 287 but the evolution is different for each peak. Figure 11 demonstrates that the increase in  rendering (with a quadrant removed to aid viewing).
In order to obtain a full 3D analysis, the damage corresponding to each peak needs to be 307 separated. The peak separation and analysis are described in the next section.  309 The distance transform allows the damage between each ply to be separated and visualized.

310
The voxels corresponding to each peak are assigned different labels in Avizo, so that each 311 damage layer can be visualised independently. The results for the specimens tested at 5 J and 312 20 J are shown in Figures 13 and 14, respectively.