NDT Data and Mechanical Performance Loss in Composites

Relating NDT Data to Mechanical Performance Loss in Impact-Damaged Composites

Daniel Aylett and Ian Cooper
TWI Technology Centre (Wales) Port Talbot, SA13 1SB, UK

J C Arnold
College of Engineering, Swansea University Single Park, Swansea, SA2 8PP, UK

Paper presented at BINDT NDT 2014, 53rd Annual conference

Abstract

There is currently no accepted method of incorporating non-destructive test data, gained from the inspection of impact-damaged composites directly into finite element models capable of predicting remaining mechanical performance.

An improved ability to relate non-destructive testing data to performance is vital to allow damaged composites to be assessed rapidly and effectively prior to continued use, repair or replacement.

This paper presents finite element simulations of the compression after impact tests performed on polymer-matrix composite laminates reinforced by unidirectional carbon fibres. These tests are performed on laboratory coupons, which are monolithic, flat, rectangular composite plates with conventional stacking sequences. The tests are simulated using conventional failure criteria for fibre, matrix damage and interlaminar delamination.

Ultrasonic data acquired from the impact-damaged panels are incorporated automatically into the simulations to generate the initial condition of the laminates in terms of delaminations and matrix cracks. Good correspondence is achieved between simulation and experiment for aspects such as force-displacement responses, and buckling behaviour prior to final failure.

1. Introduction

This paper investigates the simulation of compression-after-impact (CAI) failure in uni-directional, carbon fibre-reinforced plastic (CFRP) composites. Compression failure of composite laminates previously damaged by an impact event is due to the propagation of impact-induced damage mechanisms such as interlaminar delaminations, matrix/fibre cracking as well as the interactions between them.

There is currently no accepted method of incorporating non-destructive test data, gained from inspection of an impact damaged composite directly into finite element models capable of predicting mechanical performance. Previous studies undertaken regarding the simulation of the CAI test utilise idealised delaminations^(1-3) and/or intra-ply damage⁽⁴⁾ that are manually generated, or more recently, the result of a prior impact event simulation^(5-6).

An improved ability to relate non-destructive testing data to remaining mechanical performance is vital to allow damaged composites to be assessed rapidly and effectively prior to continued use, repair or replacement.

This paper presents the experimental and simulation work carried out at TWI Technology Centre (Wales), Swansea University and Cardiff University as part of the IntACom project in order to develop a method of incorporating NDT data acquired from an impact damaged CFRP laminate directly into simulations capable of predicting remaining residual strengths.

2. Material

CFRP laminates were manufactured from M21/35%/UD268/T700 pre-preg and laid up in a quasi-isotropic, (0/-45/90/45)_2s stacking sequence. The mechanical properties of each ply are detailed in Table 1 . The panels were cured in an autoclave according to⁽⁷⁾ and then sectioned into 100 x 150mm coupons. The 0.26mm cured ply thickness resulted in a 4.16mm laminate thickness.

Table 1 M21/T700 material properties

3. Assessment of damage prior to impact loading

The panels were impact tested at four energy levels with an Instron Dynatup 9250HV Impact testing machine adhering to the D7136/D7136M standard⁽⁸⁾ where applicable. Impact energies of 14, 22, 30 and 38J all resulted in a dent depth less than barely visible impact damage (BVID) level (0.3mm). Each energy level was repeated four times on four separate panels. The resulting impact damage was assessed using ultrasonic testing (UT) for all panels and computed tomography (CT) was undertaken on four panels containing damage from each impact energy for validation purposes.

UT was undertaken in a USL immersion tank with a 5MHz probe and a scanning increment of 1mm in the XY plane. Raw ultrasonic data in the form of A-scans at each inspection point were post-processed in matlab to produce time of flight (TOF) C-scans which illustrate delaminations detected at varying depths in the laminates. Examples of TOF C-scans for each impact energy can be seen in Figure 1 . The plotted colours correspond with the depth of the detection within the laminate. The TOF images provided evidence of the delamination pattern of propagation found in literature for impact damaged CFRP laminates⁽⁹⁾. Specifically, the delaminations were observed to be oblong shaped, stretched in the direction of the fibre orientation of the ply immediately below, thus forming the ‘spiral staircase’ or helix shaped damage due to the stacking sequence. It is worth noting that any delamination that is located nearer to the transducer shadow those located further from it, because the ultrasonic beam is reflected from the uppermost delamination. Due to this shadow effect, only the part of the deeper delaminations not covered by those above can be detected.

Impact damage sites measured with UT were sectioned from a selection of panels for CT inspection. An X-Tek micro-focus CT system was utilised for all 4 inspections. After reconstruction the resulting .VOL files were post processed utilising a similar method to that  proposed by⁽¹⁰⁾. The commercial software package, VG-Studio MAX, was utilised for post processing. Firstly, the adaptive rectangle tool was utilised to isolate the specimen from the background voxels. A region growing technique was then employed to extract darker voxels below a chosen threshold as illustrated in Figure 2. This resulted in a region of interest consisting of only damage such as fibre/matrix cracks and delaminations. Fibre and matrix cracks were manually isolated for each ply and coloured green. Delaminations were also isolated and colour-coded individually as seen in Figure 3.

Delaminations and matrix cracks were observed to occur in predictable orientations. The well-known helix-shape of delaminations increasing in the depth of the laminate were observed, increasing in overall size with impact energy and reversing rotation from clockwise to anticlockwise at the centre point. Multiple matrix shear cracks were seen to follow the fibre direction in each ply as expected. In each ply, two ‘major’ matrix shear cracks were observed to link the adjacent delaminations. This is widely known as the trapezoidal damage shape⁽¹¹⁾. Furthermore, delaminations were observed to be confined to the area between these two matrix shear cracks in the ply below, and excluded from the area between the two matrix shear cracks in the ply above. This is illustrated in Figure 3 on the delamination in-between ply 3 and 4 where the delamination is limited to the area between the two matrix shear cracks in the ply below (blue lines) and excluded from the area between the two matrix shear cracks in the ply above (green lines). No delaminations were detected at the central interfaces as expected due to the identical orientation of the plies. Matrix shear cracks were however, observed to propagate through the two plies, linking the delaminations at interface 7 and 9.

4. Finite element model

Matlab was used to create an M-file capable of generating .jnl files for use in Abaqus CAE. A .jnl or journal file consists of a python script containing all the data needed to construct a finite element model in Abaqus.

Each ply was constructed and meshed individually with SC8R continuum shell elements with enhanced hourglass control. Mesh size in the through the thickness (Z) direction was limited to 0.26mm. Mesh size in the XY plane was 1mm, coinciding with the resolution of the UT scan data.

The TOF data acquired from UT inspection was accumulatively segmented to provide delamination predictions for each interface, and incorporated into the simulation. This was accomplished with a separate suite of M-files that extract data from a pair of output files (.VOL and .DAT) from the USL immersion tank. The three M-files wrap around a C# .dll, which performs all of the actual work.  The first script confirms the presence of the two files, reads the data required to set up the necessary storage variables and opens a connection to the main data file (.vol). The second script extracts individual a-scans based on the provided (X,Y) index. The final script closes the file connection and releases consumed resources.

The TOF C-scan data was converted to binary using an arbitrary threshold value resulting in a .mat file, which was utilised to generate the selection of elements to be assigned cohesive behaviour. A TOF C-scan of a panel impacted at 30J and the corresponding selection of elements for the cohesive surface interaction function at each ply interface can be seen in Figure 4. Note that interface 8 was assumed to contain no delamination due to the identical fibre orientation of plies 8 and 9.

Delamination onset and propagation was modelled using the maximum traction damage initiation criterion for cohesive surfaces. Values used for the interaction were calibrated by simulating double cantilever beam (DCB) and end notched flexure (ENF) tests using the same modelling methodology outlined in⁽¹²⁾. Hard frictionless contact was also assumed between adjacent plies to stop delaminated areas passing through each other.

Intralaminar damage was accounted for using the in-built Hashin failure criteria⁽¹³⁾ in Abaqus. Material properties described in Table 1 were utilised for all simulations.

The loading fixture and boundary conditions applied to simulate the test can be seen in Figure 5. Anti-buckling knife-edge supports were modelled as analytical, rigid, revolved shells with a 1mm radius representing the points of contact between the knife edge supports and the front and rear of the panels. Hard frictionless contact was assumed between each of the four cylinders and the front/rear surfaces of the panel. Each cylinder was translated to be 0.04mm away from the panel in the Z direction. The ‘encastre’ boundary condition was utilised for each cylinder.

Matrix cracks were simulated by assigning material properties with negligible shear moduli (G12, G13 G23) to a strip of elements generated for each ply which are orientated to the fibre direction passing through the centre point, and limited to the size of the delamination defined for the rear surface of the ply.

Bi-Linear strain gauges were attached to four panels containing damage created at each impact energy to monitor the horizontal and vertical strain variations at both surfaces during the experimental CAI tests. The strain gauges were named according to Table 2 using Figure 7 as reference. Each strain gauge was simulated as a pair of nodes. By measuring the change in distance between each pair of nodes in 3D space the local strain could be calculated.

Table 2 Strain gauge labeling

Loading of the specimens was achieved by imposing a 1.5mm displacement to the top edge and clamped region of the laminate as illustrated in Figure 5. The displacements were artificially accelerated to 950mm/sec by imposing a reduced step time consequently reducing the number of increments required for the analysis. This artificial increase of the loading velocity has minimal effect on the analysis results as dynamic effects only become significant beyond a velocity of 1000 mm/sec⁽¹⁴⁾. This was confirmed by observing that the kinetic energy in each simulation remained less than 10% of the total external work. The dynamic effects were also reduced by the specification of a smooth time step resulting in gradual loading and unloading, thus minimising abrupt accelerations in the structure. Minimal change in the total energy of each model was observed, which indicates that the stability limit was not exceeded and that the models were yielding appropriate responses.

6. Results

Figure 8 to Figure 11 compare experimental and simulated results for the load/displacement and load/local strain relationships. The simulated stiffness values of the panels are well predicted, along with the failure loads, which are all within 10% of the average experimental results for each impact energy.

Simulated strain gauge results also exhibit some similarities to that observed for the experimental tests; most notable are the responses of the vertical strain gauges situated on the back of the panels (BRV and BLV) for the 22, 30 and 38J tests seen in Figure 9, Figure 10, and Figure 11 respectively. The results are indicative of local buckling occurring on the rear of the panels as delaminations propagate laterally to the edges of the panel prior to complete failure. The simulations capture this behaviour well, albeit slightly delayed when compared to that seen in the experimental tests. Other similarities include the response of the horizontal gauges (FRH, FLH, BRH and BLH) for the 14J tests (Figure 8) in which both experimental and simulated results are indicative of slight lateral bending just before final failure.

Figure 12 and Figure 13 use the 14J panel as an example of the experimental and simulated panel prior to CAI failure. It can be seen that there is good agreement between the failure location and extent of delamination propagation up until failure.

7. Conclusions

A methodology of incorporating UT data acquired from impact damaged CFRP laminates directly into simulations capable of predicting remaining residual strengths has been developed. The developed methodology requires the assumption that matrix cracking along the fibre direction always occurs in the intra ply region between delaminated interfaces.

The simulations highlight two phenomena; the propagation of delaminations leading to sub-laminate buckling, and the propagation of matrix and fibre cracks due to the buckling. These two phenomena develop together during CAI and induce final failure of the plate. The initial damage is well represented due to the incorporation of UT data and idealised matrix cracks. The CAI strength is also well predicted for the range of impact energies. Furthermore, similarities observed in the strain gauge results prior to panel failure, indicate that the buckling modes which induce final collapse are well predicted.

Acknowledgements

This work formed part of IntACom, a joint industry project funded by Rolls-Royce, GKN, Bombardier Aerospace and TWI. It was awarded funding though Engineering and Physical Sciences Research Council Industrial CASE award programme, made possible by the EU’s Convergence European Social Fund through Welsh Government.

References

1. Aoki, Y., H. Kondo, and H. Hatta, Effect of delamination propagation on mechanical behaviour in compression after impact. ICCM16, Kyoto, Japan, 2007.
2. Craven, R., L. Iannucci, and R. Olsson, Delamination buckling: A finite element study with realistic delamination shapes, multiple delaminations and fibre fracture cracks. Composites Part A: Applied Science and Manufacturing, 2010. 41(5): p. 684-692.
3. Craven, R., L. Iannucci, and R. Olsson, Delamination buckling: A finite element study with realistic delamination shapes, multiple delaminations and fibre fracture cracks. Composites Part A: Applied Science and Manufacturing, 2010. 41(5): p. 684-692.
4. Riccio, A. and E. Pietropaoli, Modeling damage propagation in composite plates with embedded delamination under compressive load. Journal of composite materials, 2008. 42(13): p. 1309-1335.
5. Zeng, L. and R. Olsson, Buckling-induced delamination analysis of composite laminates with soft-inclusion. 2002, FOI.
6. González, E.V., et al., Simulation of drop-weight impact and compression after impact tests on composite laminates. Composite Structures, 2012. 94(11): p. 3364-3378.
7. Rivallant, S., C. Bouvet, and N. Hongkarnjanakul, Failure analysis of CFRP laminates subjected to compression after impact: FE simulation using discrete interface elements. Composites Part A: Applied Science and Manufacturing, 2013. 55(0): p. 83-93.
8. Hexcel (2007) HexPly® M21 Epoxy Matrix (180°C/356°F curing matrix) Product data.
9. ASTM7136/D, D 7136/D 7136M – 05 Standard Test Method for Measuring the Damage Resistance of a Fiber Reinforced Polymer Matrix Composite to a Drop-Weight Impact Event. 2005.
10. Abrate, S., Impact on Laminated Composite Materials. Applied Mechanics Reviews, 1991. 44(4): p. 155-190.
11. McCombe, G., et al., X-ray damage characterisation in self-healing fibre reinforced polymers. Composites Part A: Applied Science and Manufacturing, 2012. 43(4): p. 613-620.
12. Rivallant, S., et al., Experimental analysis of CFRP laminates subjected to compression after impact: The role of impact-induced cracks in failure. Composite Structures, 2014. 111(0): p. 147-157.
13. Turon, A., et al., An engineering solution for mesh size effects in the simulation of delamination using cohesive zone models. Engineering Fracture Mechanics, 2007. 74(10): p. 1665-1682.
14. Hashin, Z., Failure Criteria for Unidirectional Fiber Composites. Appl. Mech, 1980. 47: p. 329-334.
15. Simulia (2011) Simulation of the Quasi-static Crushing of a Fabric Composite Plate Abaqus Technology Brief.