ABSTRACT
The experimental feasibility of a vibration based approach to identify damage in an advanced composite aircraft structure is presented. Analysis showed that the Modal Strain Energy Damage Index (MSE-DI) algorithm can be used to detect and localize single and multiple damage scenarios by using modal data. However, this method is less suitable to quantify the severity in terms of mechanical properties. Based on the concept of dimensional reduction it is suggested to move to a higher dimensional subset of parameters. A combination of damage sensitive features is required to enclose all levels of damage identification.
1 INTRODUCTION
The number of successful practical applications of Structural Health Monitoring (SHM) on (composite) aircraft structures is minor [1, 2] compared to applications in civil and offshore industry. This is mainly due to the complexity of the components and the high demands on safety and reliability of the SHM-system. A robust SHM-system requires the unique characterization of the presence, location and severity of the damage. A wide range of technologies, comprising global vibration and local wave propa-gation methods, is employed for health monitoring purposes [3]. No method solves all problems in all structures [1]. Defining damage sensitive parameters that are able to uniquely identify damage based on realistic measured data of large application-specific structures is one of the key challenges for these technologies. The trade-off between the damage parameters in time, time-frequency, frequency and modal domain is often far from clear-cut.
Earlier studies [4–6] revealed that structural vibration based technologies combined with the Modal Strain Energy Damage Index (MSE-DI) algorithm (introduced by Stubbs [7]) has a great potential. Delamination damage at the skin-stiffener intersection of
T.H. Ooijevaar∗, R. Loendersloot, L.L. Warnet, R. Akkerman and A. de Boer, Faculty of
Engi-neering Technology, University of Twente, P.O. Box 217, 7500AE, Enschede, The Netherlands
∗E-mail: t.h.ooijevaar@utwente.nl
Structural Health Monitoring of an
Advanced Composite Aircraft Structure
Using a Modal Approach
T. H. OOIJEVAAR, R. LOENDERSLOOT, L. L. WARNET,
R. AKKERMAN and A. DE BOER
relatively simple and small structures was detected and localised.
This paper will focus on an experimental investigation of the feasibility of this struc-tural vibration based approach to detect and localise damage in a larger and more ad-vanced composite structure, representing a typical aircraft structure. Only modal domain data is used in the analysis. The performance of a 1D and 2D formulation of the MSE-DI algorithm is compared for several damage scenarios. Special attention is paid to the perspectives of damage severity estimation based on modal domain data. This work is also considered as the validation of the numerical work presented in [8].
2 COMPOSITE SKIN-STIFFENER STRUCTURE
This research concentrates on carbon AS4D fibre reinforced thermoplastic (PEKK) skin-stiffener structures. This typical aerospace structure is vulnerable for delamination damage at the skin-stiffener interfaces. A large plate (figure 1) with three stiffeners and a non-uniform thickness distribution was manufactured by Fokker Aerostructures. The plate consists of a 16-layer quasi-isotropic midsection and 44-layer and 30-layer end-sections. The stiffeners are made of a 15-layer quasi-isotropic stack. A PEKK injection moulded filler containing short fibres is used for the connection.
27 mm x y 27 mm M1 148 mm (P29) Impact location I1 A (P203) 25 mm 150 mm 153 mm 275 mm 242 mm 35 mm M2 148 mm t = 4.14 mm t = 2.21 mm t = 6.08 mm 1304 mm Section A-A 92 mm 228 mm 456 mm Excitation point (P1) (P175) 19 mm A
Figure 1: Bottom and cross-sectional view of the panel, indicating the locations of the measuring points (dots) and the locations of the added mass (M1 and M2) and impact location (I1). The dotted vertical lines indicate the edges of the transition zones between the sections with different thicknesses.
Two damage scenarios are considered. Firstly, a structural change was obtained by adding a small mass (42 grams, ∼1.1% of total weight) to the structure (point M1 and M2 in figure 1). An added mass provides a controlled and reversible change of the structure. This is important for the validation of numerical models [8]. Secondly, actual damage was analysed. Naturally originated barely visible defects are obtained by applying a 50J impact with the help of a Dynatup 8250 Falling Weight Impact Machine. The impact was applied at a stiffener run-out (point I1 in figure 1), since this part of the skin-stiffener connection belongs to the location with the highest risk of failure. Visual and ultrasonic inspection revealed that the damage consists of interface failure, indicated by cracks, between the skin and the stiffener (figure 2) and delamination between the titanium insert and skin.
A A B B Titanium insert Crack A-A Crack B-B
Figure 2: Visible part of the damage, consisting of a cracked skin-stiffener interface, caused by 50J impact at the stiffener run-out. The surface at the damaged area is painted to enhance the visibility of the damage.
3 DAMAGE FEATURES FOR DAMAGE IDENTIFICATION 3.1 Damage Feature Space and Dimensional Reduction
Damage sensitive parameters, referred as features, are used to uniquely characterize damage. The internal parameters ψd(t), such as crack length and loss of stiffness, pro-vide the most direct description of damage, but are in general not directly measurable. They are related to measurable output variables r(t) like strains and velocities.
Devia-tions∆r(t) of these outputs can be used for diagnosis by solving the inverse problem [3]: ψd= F−1(∆r, ψe) (1) where ψe(t) represents the environmental and operational conditions. The goal is to maximize the damage information gathered from the measurement data and minimize the uncertainty of the damage parameter. Practically, it is desired to condense the in-formation to a lower dimensional parameter space (figure 3). The deviation in the mea-sured time domain variables are therefore regularly replaced by damage features from the time-frequency, frequency and (direct or extended) modal domain space.
Damage parameter ψd,
internal state variables
Time-Frequency and frequency domain ( ) r(ω t,) r ω • Natural frequency • Mode shape • Modal damping • ...
• Mode shape derivatives • Modal strain energy • Modal flexibility • ... Output variables, time domain ( )r t Modal domain (r ωn) Direct Extended
• Frequency Response Function • Transmissibility
• Short-time Fourier transform • Wavelet Analysis • Correlation functions
• Statistical time series analysis • ... • Crack length • Loss of stiffness/mass • Play • ...
Figure 3: An overview of damage feature sub-spaces with a gradually condensation of information.
This reduction behaviour can be illustrated by considering Frequency Response Func-tions (FRFs). FRFs are typically obtained after a windowed averaging process of the time domain signal. Depending on the time scale of the evolving damage, the system is assumed to be invariant. Modal analysis will further condense the damage information to the modal frequencies only. Information between the modal frequencies is not taken into account. Subsequently, slightly complex modal vectors are generally considered to be real normal, by neglecting non-proportional damping effects. Moreover, the modal vectors are scaled vectors without a physical quantity. Finally, in case of extended modal
damage features typically a selection of the most sensitive modes is incorporated to ob-tain an indicator. The reduced subset of parameters must be able to describe the damage scenario and is selected according to the intended level of damage identification [1, 3].
3.2 Extended Modal based Identification Method
A number of direct and extended modal domain damage features (see figure 3) are applied to the measured data of the composite aircraft structure. The most promising de-tection and localisation (level 2) results were obtained by the MSE-DI algorithm, which are presented in this paper.
A 1D formulation of the MSE-DI was introduced by Stubbs [7], while Cornwell [9] extended this approach for the 2D case. Here, the basics of the 2D formulation are explained. The derivation and assumptions are analogous to the one used for the 1D for-mulation [4, 10]. Consider a plate-like structure to be discretised in Nx× Ny elements in respectively x- and y-direction. The strain energy U , based on bending deformation, of each of the individual modes n and element ij of a vibrating structure is represented by:
Uij(n) = 1 2 Z yj yj−1 Z xi xi−1 Dij ∂2u(n)z ∂x2 !2 + ∂ 2u(n) z ∂y2 !2 + 2ν ∂ 2u(n) z ∂x2 ! ∂2u(n)z ∂y2 ! + 2 (1 − ν) ∂ 2u(n) z ∂x∂y !2 dxdy (2)
with u(n)z (x, y) the displacement amplitude of the nth participating mode shapes, D the flexural rigidity of the plate, ν the Poisson’s ratio, xi and yj the limits of element ij of the plate structure in x and y direction respectively. The total modal strain energy is approximated by the sum of equation 2 over a limited set of Nf req modes. Following the definition proposed in [9], the ratio of fractional element stiffnesses of the damaged structure over the reference structure provides the base of the damage index:
˜ γij(n).γ˜(n) γij(n).γ(n) = Ryj yj−1 Rxi xi−1w˜ (n)dxdy.Rb 0 Ra 0 w˜ (n)dxdy Ryj yj−1 Rxi xi−1dxdy . Rb 0 Ra 0 w(n)dxdy (3)
where w(n)(x, y) represents the second term between the square brackets in the
inte-grand of equation 2, γij(n)the integral of w(n)(x, y) over element ij and γ(n)the integral
w(n)(x, y) over the entire dimensions a and b. The damaged case is represented by the
tilde sign on top of the variable. The information in each of the mode shapes is combined in a damage index β, according to the definition proposed by Cornwell et al. [9]:
βij = Nfreq X n=1 h ˜ γij(n).γ˜(n)i ,Nfreq X n=1 h γij(n).γ(n)i (4) An overview of most common alternative formulations is presented in [8]. The damage index βij is generally normalised using the standard deviation σ and the mean µ of the damage index over all elements. This results in the value Z, defined in each element ij:
Zij =
βij − µ
4 EXPERIMENTAL ANALYSIS AND DAMAGE SCENARIOS
Vibration measurements are performed on the skin-stiffener structure before and af-ter the structural changes were applied. The complete dynamic set-up and data acquisi-tion scheme used for the experiments are similar to the one presented in [4]. However, the structure is freely suspended in vertical direction by two elastic wires. A random excitation force was applied by a shaker connected to driving point P1 (figure 1). A laser vibrometer measured the velocities along a measurement grid containing 7×29 points (figure 1). The Frequency Response Functions (FRFs) between the fixed point of excitation and the measurement points are recorded by a Siglab system. A frequency range of 50-1050 Hz (resolution: 0.3125 Hz) was selected. A measurement at each grid point consists of 20 windowed averages, with 50% overlap. The modal parameters (natural frequency, mode shapes and damping values) are obtained from the FRFs by using Experimental Modal Analysis [11] (see [4] for a description of the method used). Successive measurements showed a sufficient repeatability of the experimental set-up and testing approach (natural frequencies: σ = 0.05%, σmax = 0.53%, mode shapes: average M AC > 0.98, minimum M AC = 0.88, with M AC standing for the Modal
Assurance Criteria [12]). The mode shapes are spline interpolated and are the input for damage diagnosis by the MSE-DI algorithm.
5 RESULTS AND DISCUSSION
The 2D and the 1D formulation in x- and y-direction of the MSE-DI algorithm are applied for three damage scenarios: an added mass, impact damage and a combination of an added mass and impact damage. All measured modes within a frequency range of 200Hz-800Hz are considered in the analysis. Clear peaks in the damage index distribu-tions Zij indicate the presence and location of damage. The ratio between the damage indices of the damaged and intact area showed to be a measure for the sensitivity to identify damage [4].
5.1 Single Damage Scenarios
The damage index Zij distributions for the mass and impact damage are respectively shown in figures 4 and 5. The 2D MSE-DI algorithm, figures 4a and 5a, shows the highest indices at respectively the location of the added mass (point M1) and the im-pact damaged area (near point I1), indicating the presence and location of both damage types. It must be noticed that the effect of the mass is also clearly shown in the nat-ural frequencies and mode shapes (maximum change: δFN = 2.91%, M AC = 0.53, average: δFN = 0.54%, M AC = 0.84). However, these properties are hardly affected by the impact damage (maximum change: δFN = 0.68%, M AC = 0.84, average:
δFN = 0.19%, M AC = 0.98).
The presence and location of the mass is also predicted correctly by the 1D formu-lation in x-direction (figure 4b). The y-direction presents slightly higher peaks at the location of the mass, but also comes up with a false positive (indication of damage, but no actual change in the structure) at the top-right boundary. The difference between
x-b) 1D-x c) 1D-y a) 2D x coordinate [m] (13.2) (13.5) (5.9) (4.3) y coordinate [m] Damage Index [-] Zij y coordinate [m] y coordinate [m] Damage Index [-] Zij Damage Index [-] Zij
Figure 4: Normalized damage indices Zij for a
mass (42g) at point M1. All modes within 200-800Hz and 20×60 MSE-DI elements are used.
y coordinate [m] Damage Index [-] Zij c) 1D-y x coordinate [m] y coordinate [m] Damage Index [-] Zij b) 1D-x Damage Index [-] Zij a) 2D y coordinate [m] (7.6) (7.7)
Figure 5: Normalized damage indices Zij for the
50J impact at point I1. All modes within 200-800Hz and 20×60 MSE-DI elements are used.
and y-direction originates from the way the mode shape curvatures are affected. This effect is defined by the location of the mass with respect to the local mode shape. The structural design forces the mode shapes to show their highest amplitudes between the stiffeners. The y-coordinate of the mass almost coincides with the location of these maxima, resulting in a minor effect on the curvature. Since the stiffeners force the mode shapes in y-direction to be relatively similar for nearly all modes, the effect is less pro-nounced in the damage index compared to the x-direction. For this situation, the results in x-direction outperforms the y-direction.
Earlier obtained results [5] for interface damage between skin and stiffeners showed that the best results were obtained by considering the 1D MSE-DI in the direction of the stiffeners, the x-direction. This is due to the fact that mainly the bending stiffness in this direction was affected. An opposite behaviour is shown for the impact damage considered here (figures 5b and 5c). Several false positives appear for the x-direction, while the y-direction clearly predicts the presence and the location of the impact damage. The most likely explanations for the reduced effect in x-direction are: Firstly, the fact that the damage is located at the end of the stiffened section. This transition in x-direction causes a discontinuity in the curvature, which will disturb the effects caused by the damage. Secondly, the fact that the damage is located at the thickest part of the skin (figure 2). Generally, the thickest skin section showed a relatively large wavelength of the mode shape in x-direction, representing lower vibration modes, compared to the wavelength in x-direction at the mid-section and the wavelength in y-direction (figure 6). The reduced sensitivity for the x-direction is understandable, since the lower vibration
modes are in generall hardly affected compared the higher vibration modes [4]. Thirdly, the fact that the interface damage is located close to the local shear centre, in case the plate problem is reduced to a local bending and torsion problem around the middle stiff-ener. The mode shape in x-direction will be hardly affected in case the deformation around the damage represents local torsion in the yz-plane, similar to the observations presented in [4]. A y coordinate [m] 0 0.2 0.4 0 0.2 0.4 0.6 0.8 1 1.2 -1 -0.5 0 0.5 1 x coordinate [m] Displacement [-] C B t = 4.14 mm t = 2.21 mm t = 6.08 mm A B C 0 0.2 0.4 0.6 0.8 1 1.2 x coordinate [m] t = 4.14 mm t = 2.21 mm I1 t = 6.08 mm
Figure 6: Experimental mode shape, interpolated by a cubic spline function, of the intact structure (FN =
544.8Hz) showing larger wavelengths in x-direction at the thicker sections
The more pronounced effect in y-direction can be explained by the widening of the stiffener towards its end. The delamination underneath the stiffener and the titanium insert locally reduces the bending stiffness in y-direction. Moreover, the delamination is often located at a region (near inflection point) with a high shear stress in y-direction, in case it is considered as a local bending problem in y-direction. According to [13] this results in a larger effect on the flexural rigidity and therefore the mode shapes.
5.2 Multiple Damage Scenario
An added mass (point M2) and impact damage (near point I1) were considered for a multiple damage scenario. The 2D MSE-DI results are presented in figure 7. Two clear peaks are shown at the coordinates corresponding to the locations of the mass and damage. However, the effect of the mass is more pronounced. The latter implies a larger influence of the mass at this location on the entire dynamic behaviour. The changes in the natural frequencies and mode shapes for the single damage cases already endorsed this behaviour (section 5.1).
y coordinate [m] x coordinate [m] Damage Index [-] Zij (11.5) (4.1)
Figure 7: Normalized 2D damage indices Zijfor an added mass (42g) at point M1 and 50J impact at point
I1. All modes within 200-800Hz and 20×60 MSE-DI elements are used.
In some cases the MSE-DI algorithm can provide an estimation of the geometrical properties of the damage [4]. However, comparing the maximum damage index values of the single (figures 4a and 5a) and multiple damage scenarios indicate the incapabilities
for damage severity estimation in terms of stiffness loss. The single damage scenarios show almost equal damage index levels, while the multiple damage case shows signif-icant lower values for the same impact damage. This difference is originated by the fact that the MSE-DI algorithm is based on the ratio of fractional element stiffnesses (equation 3) without considering the participation of a mode in the actual vibration. The damage index is merely a mathematical rather than an absolute physical quantity. A larger influence on the dynamics, indicated by higher damage indices, does not neces-sarily imply a higher damage severity.
Multiple damage features are therefore required to be able to obtain a level 3 dam-age identification system that is capable of estimating the severity in terms of mechanical properties. As described in section 3.1, damage information is generally condensed to parameters with a lower dimensional space. However, the neglected part of this informa-tion could be valuable for a quantitative estimainforma-tion of the damage. Hence, it is suggested to move to a higher dimensional subset of parameters, like the direct modal or frequency domain. The location information, obtained by for example an extended modal domain feature, is inserted to this subset. A direct modal approach for severity estimation is presented in [8].
6 CONCLUSIONS AND RECOMMENDATIONS
The feasibility of an extended modal based damage identification methodology ap-plied to an advanced composite aircraft structure has been investigated experimentally in this work. Results showed that the MSE-DI algorithm can be used to detect and localize an added mass and impact damage, also in case of a multiple damage scenario. Each of the individual formulations (2D, 1D x- and y- direction) can be beneficial depending on the damage case and damage location with respect to the structural design. One of the main drawbacks is the number of required measurement points to accurately describe the mode shapes.
It can be concluded that higher damage indices do not necessarily imply a higher damage severity. The MSE-DI does not allow for a quantitative estimation of the dam-age severity in terms of mechanical properties. A robust SHM-system will therefore consists of a combination of damage features to enclose all levels of damage identifi-cation. The extended modal domain features are powerful in the detection and locali-sation of damage. They can also roughly estimate the damage geometry. To estimate the mechanical damage severity it is suggested to utilize a higher dimensional subset of parameters, since a more direct relation between the internal damage properties and the damage feature is required.
The key challenge for SHM is the practical application of these technologies to re-alistic measured data sets of application specific structures under real operational condi-tions. It must be mentioned that the selection and development of damage identification algorithms is made-to-measure work. Therefore, the development has to be an integral part of the structural design process. A scenario based design approach could be utilized to make the application of SHM-systems more effective.
ACKNOWLEDGEMENTS
The authors kindly acknowledge Fokker Aerostructures for manufacturing the com-posite panels used in this research. This work is carried out in the framework of the European project Clean-Sky Eco Design (grant number CSJU-GAM-ED-2008-001).
REFERENCES
[1] D. Montalvao, N.M.M. Maia, and A.M.R. Ribeiro. A review of vibration-based structural health monitoring with special emphasis on composite materials. Shock and Vibr. Digest, 38(4):295–326, 2006.
[2] W. Fan and P. Qiao. Vibration-based Damage Identification Methods: A Review and Comparative Study. Structural Health Monitoring, 10(1):83–111, 2010.
[3] C.-P. Fritzen and P. Kraemer. Self-diagnosis of smart structures based on dynamical properties.
Mechanical Systems and Signal Processing, 23(6):1830–1845, 2009.
[4] T.H. Ooijevaar, R. Loendersloot, L.L. Warnet, A. de Boer, and R. Akkerman. Vibration based Structural Health Monitoring of a composite T-beam. Composite Structures, 92(9):2007–2015, 2010. [5] T.H. Ooijevaar, R. Loendersloot, L.L. Warnet, A. De Boer, and R. Akkerman. Vibration based Structural Health Monitoring of a Composite Plate Structure with Multiple Stiffener. In European
Workshop on Structural Health Monitoring 2010, volume 92, pages 2007–2015, 2010.
[6] R. Loendersloot, T. H. Ooijevaar, L.L. Warnet, A. De Boer, and R. Akkerman. Vibration based structural health monitoring of a composite plate with stiffeners. In Proceedings ISMA2010, pages 909–924, 2010.
[7] N Stubbs and C.R. Farrar. Field Verification of a Nondestructive Damage Localization and Severity Estimation Algorithm. Proceedings of 13th IMAC Conference, pages 210–218, 1995.
[8] R. Loendersloot, T.H. Ooijevaar, A. De Boer, and R. Akkerman. Development of a Damage Quan-tification Model for Composite Skin-Stiffener Structures. In Proceedings of SMART11, 2011. [9] P.J. Cornwell, S.W. Doebling, and C.R. Farrar. Application of the Strain Energy Damage Detection
Method To Plate-Like Structures. Journal of Sound and Vibration, 224(2):359–374, 1999.
[10] A. Alvandi and C. Cremona. Assessment of vibration-based damage identification techniques.
Jour-nal of Sound and Vibration, 292(1-2):179–202, 2006.
[11] B.J. Schwarz and M.H. Richardson. Experimental Modal Analysis, 1999.
[12] R.J. Allemang. The modal assurance criteriontwenty years of use and abuse. Sound and Vibration, 1:14–21, 2003.
[13] W. Grouve, L. Warnet, A. De Boer, R. Akkerman, and J. Vlekken. Delamination detection with fibre Bragg gratings based on dynamic behaviour. Comp. Science and Tech., 68(12):2418–2424, 2008.
