Experimental Damage Identification of Carbon/ Epoxy Composite Beams Using Curvature Mode Shapes Cole S. Hamey,1 Wahyu Lestari,1 Pizhong Qiao1,* and Gangbing Song2 1 Department of Civil Engineering, The University of Akron, Akron, OH 44325-3905, USA 2 Department of Mechanical Engineering, University of Houston, Houston, TX 77204-4006, USA Many composite materials and structures are susceptible to defects, which can significantly reduce the strength of structures and may grow to failure. To avoid the catastrophic failure of structures, development of a reliable method of structural health monitoring is one of the most important keys in maintaining the integrity and safety of structures. Dynamic response-based damage detection offers a simple procedure as an alternative to the conventional nondestructive evaluation techniques. However, this technique depends on the quality of measured data for its identification accuracy. In this article, experimental aspects of dynamic response-based damage detection technique on carbon/ epoxy composites are addressed. Smart piezoelectric materials are used as sensors or actuators to acquire the curvature modes of structures. These materials are surface-bonded to the beams. An impulse hammer is used as an actuating source as well. Four types of damage detection algorithms are evaluated for several possible damage configurations with two different excitation sources. The quality of damage identification with the four different detection algorithms is discussed. These experimental damage identification techniques using curvature modes and piezoelectric materials can be effectively used in damage detection and health monitoring of composite structures. Keywords dynamic response
damage detection
composite materials
piezoelectric sensors and actuators
composite beams 1 Introduction Advanced composite materials have been extensively used in structural applications, due to their advantageous characteristics, such as high stiffness and strength-to-weight ratios, improved fatigue resistance, and superior damage tolerance capability compared to metallic structures. Carbon/ epoxy composites have higher stiffness and strength properties than other composites, such as the commonly used E-glass/epoxy composites. These advantageous properties have led to the use of carbon/epoxy composites in structures that undergo higher stresses, such as aircraft and aerospace structures. However, the carbon/epoxy composite laminates, like other composite materials and structures, are susceptible to defects, which can originate from imperfections in *Author to whom correspondence should be addressed. E-mail: Qiao@uakron.edu. Copyright ß 2004 Sage Publications, Vol 3(4): 0333–353 [1475-9217 (200412) 3:4;333–353; 10.1177/1475921704047502] 333 334 Structural Health Monitoring 3(4) manufacturing process or develop during their service life. Defects like fibre breakage, matrix cracking, debonding between fibres and matrix, and delaminations or interlayer cracks, are typical damages in composite structures. These defects can significantly reduce the strength of structures and may eventually grow to failure. When the failure occurs, it is often catastrophic, leading to loss of human life and/or monetary losses. Such failures often cause devastating effects on the psychological state of the public as well. The development of a reliable method of structural health monitoring is one of the most important keys in maintaining the integrity of structures. Some of the nondestructive evaluation equipment that utilise technologies such as X-ray imaging or eddy current can identify damages; but often these technologies are difficult to implement on the site. In this study, dynamic response-based damage detection techniques using smart materials are explored for carbon/epoxy composites. The dynamic response of structures can offer unique information on defects that may be contained within these structures. Changes in the physical properties of the structure due to damage will alter the dynamic responses such as natural frequencies, damping and mode shapes. These parameter changes can be extracted to predict damage information, such as the presence, location and severity of damage in a structure. The dynamic response-based damage detection method is an interesting method due to its simplicity of implementation. One method, in which the dynamic response is utilised, is to use the curvature mode shapes to detect damage. The curvature mode shape change due to damage has a local effect in nature; hence, it can be used to locate damage properly, provided that the changes of curvature are closely related to the changes of physical properties in the structures. The curvature mode shape methods have a potential to identify damage types that are hardly visible or lay beneath the surface, such as delamination. The challenge is to develop the ability to identify the changes of response parameters (e.g., deformations and dynamic characteristics) and interpret them in relation to the changes in physical properties of the structures. Moreover, the ability to differentiate the types of damage in a structure is also very important, since two different types of damage may result in the same changes in the parameters tested. For example, a beam with large delamination and a beam with two small delaminations may cause the same frequency changes. Moreover, damage detection in composite structures is more difficult compared to the metallic structures due to the anisotropy of the material, the conductivity of the carbon fibre, and the fact that much of the damage often occurs beneath the surface and is hence hardly detectable or visible. Some of the research studies related to dynamic response-based techniques are summarised here. Using a torsional spring to model the change in stiffness at a crack location, an analytical solution and damage identification to a cantilever beam was developed by Rizos et al. [1]. The displacement mode shapes were determined experimentally and analytically, and their comparison showed promising results. In a review of frequency-based methods, Salawu [2] discussed the effects of damage on the natural frequencies of a structure. However, the frequency-based methods might not confidently be able to determine the state of the structure if the change in the natural frequencies was less than 5%. In conclusion, the natural frequencies alone might not be sufficient for a unique identification and location of structural damage. More effective methods of damage detection using the curvature mode shapes have been considered and proposed. Pandey et al. [3] were among the first to develop the idea of damage detection using the curvature mode shapes. Although, the absolute difference between the displacement mode shapes of damaged and undamaged beams was not discernable between the damage location and other parts of the beam, the curvature modes showed a significant change at the damage location. Luo and Hanagud [4] developed a relationship between the dynamic properties of damaged and undamaged structures in the form of an integral equation to identify damage. The detection algorithm used the eigenvalues as well as the eigenfunction information of the system to identify the locations and corresponding severities simultaneously, with the input to the flaw detection only based on the experimental data. C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite Wahab and Roeck [5] conducted an experimental damage detection study using the curvature mode shapes of a real structure. Based on the displacement data gathered during the razing of bridge Z24 in Switzerland, the curvature mode shapes were derived by a central difference approximation. The mode shape absolute difference averaging noted as the ‘curvature damage factor’ (CDF) was used as a detection criterion. Wahab [6] further examined a model updating in combination with the curvature mode shapes for damage detection. In this method, an iterative process was initiated which allowed the parameters of the simulated beam to converge to meet the parameters of the actual beam. The convergence of the model did not improve with the inclusion of curvature data. The sensitivity of results when the curvatures were included did not change substantially. Lestari and Hanagud [7] derived a mathematical relationship between an intact beam and a damaged beam, to identify the damage location and severity simultaneously. Providing experimental data for both the healthy and damaged beams were well acquired, the damage could be identified by using a single mode, based on the curvature mode shapes information. Wang and Wang [8] explored the feasibility of using the piezoelectric materials for modal testing on a cantilever beam. Different combinations of sensors and actuators were simulated. From a comparison of the modal damping ratios and natural frequencies, it was demonstrated that the PVDF sensors and the PZT actuators were able to generate results that were similar to those of an accelerometer and an impulse excitation. Building on the reliability of curvature mode-based methods and considering the ease of application of frequency-based methods, Sampaio et al. [9] developed the frequency response function (FRF) curvature method. The benefit of this method was that there was no need to perform a complex modal analysis of the structure. The central difference approximation was applied to the FRF to obtain the second derivative FRF curvature. Using data from the Interstate 40 bridge in New Mexico, the experimental results of the FRF curvature method were demonstrated, 335 and the damage was located. In a comprehensive experimental and numerical study on the Interstate 40 bridge, different methods of damage detection (i.e. damage index, curvature mode shape, change in flexibility, change in uniform load surface curvature, and change in stiffness) were compared [10,11]. Comparisons of different damage detection algorithms indicated that the use of damage index yielded the best results, whereas the flexibility method and the stiffness method provided poor results. The main goal of this study is to develop damage detection techniques based on the dynamic response and the utilisation of smart materials. Most of the experimental curvature data in literature were obtained by approximating the second derivatives from displacement data. Experimental determination of curvature mode shapes by directly measuring the curvature modes most likely yields better results than those obtained from the displacement mode shapes. With the availability of cost-effective and easy to install piezoelectric materials, the curvature mode shapes could more easily be obtained experimentally. This paper will address some problems on experimental aspects of structural health monitoring based on dynamic response. In particular, the changes of measured curvature in the form of mode shapes or the frequency response functions (FRF) are used to identify damage in composite structures. Four damage detection algorithms based on the curvature shapes and curvature FRF are reviewed. To aid in the detection of damage, smart materials are often used. Piezoelectric materials are the most commonly used smart materials in structural health monitoring, due to their ability to act both as an actuator and as a sensor and their flexibility to be sized for specific applications. In this study, the piezoelectric materials in the form of ceramic (lead–zirconate–titanate, PZT) and polymer film (polyvinylidenefluoride, PVDF) are used as the actuator and the sensor, respectively. Unlike the PZT ceramics, the PVDF films are flexible which allows for more easy bonding to curved and non-smooth surfaces. With the use of double-sided tape, the PVDF films can be attached to a structure and used repeatedly. 336 2 Structural Health Monitoring 3(4) Damage Detection In this study, several existing detection algorithms are employed to detect damage of carbon/epoxy laminated composite beams. Contrary to most results in the literature, the directly measured curvature data are applied to the method, and comparisons among different algorithms are made and discussed. The methods being considered include the absolute difference of curvature mode shape method, the curvature damage factor method, the damage index method, and the FRF curvature method. A brief review of these detection algorithms is presented in the following. For comparison purposes between the undamaged and damaged beam modes in curvature mode-based damage detection methods, it is best to first develop weights by comparing the undamaged structure to the theoretical one wij ¼ ij, theoretical ij ð1Þ where wij is the weight, ij, theoretical is the theoretical curvature, and ij is the experimentally measured curvature for ith mode shape at location j. These weights are applied to both the damaged and undamaged structures to allow for greater ease in comparison 0ij ¼ wij ij 0ij, damaged ¼ wij ij, damaged
0ij ¼ 0ij  0ij, damaged ð5Þ where
0ij is the absolute difference in the undamaged and damaged modes. This method examines each mode individually and is classified as a single mode method. Results from this method can vary depending on the boundary conditions, damage locations, mode of interest, and sensitivity. This variation led to the development of other multiple mode methods such as the curvature damage factor and damage index methods. The Curvature Damage Factor (CDF) method involves a similar procedure as the absolute difference method, where
0ij is determined from the absolute differences. However, the curvature damage factor was developed to consider all of the modes at once [5], ð2Þ ð3Þ where 0ij and 0ij, damaged are, respectively, the weighted undamaged and damaged curvatures used for comparison. By applying the weight to the undamaged specimen, it forces the measured curvatures into the theoretical ones (Equation (2)). The weighted shapes are also then normalised such that 0 0 T ¼ 1 zero at modal nodes at different locations as compared to the damaged structure. At this particular location, a mode shape magnitude divided by a small number of the weighting function will produce large magnitude of the weighted shape. The Absolute Differences Method (ADM) is the simplest method to employ. This method takes the absolute difference in the magnitudes of the curvature mode shape as: ð4Þ It is important to note that the weighting function may cause large discrepancies at locations where the undamaged beam approaches CDFi ¼ N 1X
0ij N j¼1 ð6Þ where CDFi is the curvature damage factor at location i and N is the number of modes that will be examined. This method is considered more accurate than the absolute difference method because it eliminates the problems caused by the damage location in conjunction with certain modes. Other problems still persist, in particular the problem of sensitivity. On the other hand, the Damage Index Method (DIM) allows for greater sensitivity. This method is also more complex as compared to the other methods. In this study, the formulation of the damage index is similar to that used by C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite Farrar and Jaurequi [10,11] ij ¼ n range, the damage location will be indicated by the following expression: o2  P n o2 o2 P n imax 0
i1max 0ij þ 1 ij,damaged n o o2 n o2  P n 2 P imax 0 ij þ 1 0ij
i1max 0ij,damaged 0ij,damaged ð7Þ where ij is the damage index at location i for mode j. Considering multiple modes, the summation of ij at one location for every mode is defined as follows: i ¼ X 337
00 ð!Þij ¼ X 00 ð!Þij  00 ð!Þij, damaged ð10Þ ! where
00 ð!Þij is the absolute difference in the FRF curvatures. This method is fairly new and only a few studies have reviewed the validity of applying this method to experimental data. Therefore, the accuracy of this method will be determined by comparing the results with the ones obtained by the other three methods. ð8Þ ij j 3 where i is the damage index at location i from the summation of the single mode damage indices. This method is considered accurate and valid for damage detection; however as with the two previous methods, the modal analysis needs to be conducted in order to employ these damage detection algorithms. Alternatively, the FRF Curvature Method (FCM) offers a procedure without performing modal analysis. Usually, the FRF is measured from the displacement response. Then, the FRF curvature for each location is calculated using a central difference formulation. 00 ð!Þij ¼ 1 ð ð!Þiþ1, j  2 ð!Þij þ ð!Þi1, j Þ h2 ð9Þ where ð!Þij is the FRF measured at location i from an input force at location j. However, since the FRF measured by the PZT ceramics or PVDF films is already a function of curvature, the numerical derivative in Equation (9) is not necessary. Therefore, over a given frequency Figure 1 Specimen Considerations and Experimentation In this study, six carbon/epoxy composite beam specimens were tested. Each sample was made of carbon fibre and epoxy resins and had a [0/90]4T lay-up for a total of eight layers. The thickness of each layer was 0.22 mm (0.0086 in.) for a total thickness of 1.75 mm (0.0688 in.). Each sample had a width of 25.4 mm (1.00 in.) and a length of 241.3 mm (9.50 in.) (Figure 1). When clamped in the cantilever configuration, the beam samples had a free span length of 228.6 mm (9.00 in.). An 8 mm  12 mm piece of PZT ceramic was attached to each composite sample as an actuator. The PVDF films were used as sensors and the beam sample was divided into 16 points to best accommodate the films (Figure 2). Each point was aligned with the centre of the PVDF during the testing. The experiment began with two undamaged beams. Both of these samples were first tested, and their undamaged mode shapes were obtained. The undamaged mode shape used for comparison An undamaged carbon/epoxy composite beam specimen. 338 Structural Health Monitoring 3(4) Figure 2 Schematic of the sensor layout for the carbon/epoxy composite samples. Table 1 Type and damage location of composite beam samples. No. Damage type 1 2 3 4 5 Delaminated A Delaminated B Delaminated C Impact Saw cut Damage location, from the fixed end (mm) Damage area (mm) Damage location according to sensor location 31.75–57.15 31.75–82.55 69.85–95.25 57.15–82.55 80.55–82.15 25.4 50.8 25.4 25.4 1.6 2–4 2–6 5–7 4–6 6 was derived from an average of the two samples. Later, one of the undamaged beams was artificially damaged by cutting a notch with a handsaw. The notch had around 1.6 mm (0.0625 in.) width and cut about 60% of the beam thickness through the width of the beam. The other four beams were already damaged, three samples contained delamination at various locations and the fourth had an impact damage. The delaminations were created during the fabrication of the samples by inserting a piece of Teflon tape between the second and the third layers of the material at the desired locations. The impact damage was created by dropping an 8.0 kg (17.6 lb) mass from a height of 304.8 mm (12.0 in.) onto an undamaged carbon composite beam sample, thus allowing for the testing and comparing of five different damage conditions: one saw-cut notch, three delamination types (A, B and C), and one impact damage. The beam with a delamination type ‘‘A’’ (Delam A) has a 25.4 mm (1.00 in.) delamination beginning approximately 31.75 mm (1.25 in.) from the fixed support or began at sensor location 2 and ended at sensor location 4. Information of all five damaged-beam configurations are summarised in Table 1, and the corresponding geometry are presented in Figure 3. For beam samples with delamination, a small bump through width between sensor locations 10 and 11 was discovered. This imperfection was developed during the manufacturing process of the composite plates with delamination before it was cut into several beams. The experimental set-up of dynamic testing is presented in Figure 4. Two different sources of excitation were employed, i.e., impulse excitation and continuous excitation by using a PZT actuator. For a testing with continuous excitation, a sweep sine with a magnitude of 140 V was run through the actuators to excite the beams. A Hewlett Packard 33120 A waveform generator was used to induce the sweep sine. The sweeps took place over a frequency range from 1 to 2000 Hz over a time of 120 s. The linear and logarithmic sweeps were used to excite each beam, since an average of these two sweeps generates the best mode shape results. The responses at each point were recorded by a dSPACE data acquisition system as time domain responses. Two sets of sweep tests were conducted on the damaged beams: one set had the sensors located on the same side of the beam as the damage was located, and in the other test the sensors were located on the opposite side of the damage. C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite 339 1. Del A 2. Del B 3. Del C 4. Impact 5. Saw-cut Figure 3 Geometry of composite beam sample configurations. For impulse excitation, a PCB impulse hammer was used as the actuator. The impulse location remained stationary and was located at the free-end of the beams. A minimum of ten sets of data was collected for each sensor location. For all samples twenty data sets were acquired at each point, except for the saw-cut sample and the first undamaged sample that each had ten data sets at a point. The FRF at each point were averaged over all the measured data sets to help eliminate noise interference recorded by the sensors. These data sets were recorded at a range from 1 to 2000 Hz; however, one data set only took one second to be conducted. The procedure for data reduction of this method is the same as that for the sweep sine methods once the FRF has been determined. For both the tests, the piezoelectric film sensors were roved along the 16 340 Structural Health Monitoring 3(4) Figure 4 Figure 5 Experimental set-up. Illustration of same-sided and opposite-sided sensors. measurement locations to allow direct generation of the curvature mode shapes. The procedure for the data reduction and shape generation can be described briefly as follows. Using MATLAB code, the time domain responses are transferred into frequency response functions (FRF) and extracted as vectors. These vectors are then converted into I-DEAS functions with the aid of IMAT interface program. Using I-DEAS test module, the modal analysis of the experimental results is performed and the curvature mode shapes are generated. Once the mode shapes are generated, the shapes are exported back into MATLAB, where they are examined thoroughly and weighted to the theoretical shapes. 4 Damage Detection Using PZT as Actuator There are two possibilities of attaching the sensors into the specimens, i.e., on the same side and on the opposite side of the damage, which can be detailed as follows. A sensor is considered on the same side as the damage if, in the case of delamination, the sensors are bonded to a surface which is closer to the plane of the delamination. In the case of impact damage the sensors are located on the concave side of the compression caused by the impact. A sensor is opposite to the damage when, in the case of delamination, the sensors are located on the surface which is farthest from the delamination plane. In the case of impact damage, the sensor is located on the convex side of the tension caused by the impact. This idea is clearly illustrated in Figure 5. The PZT ceramic patch bonded to the beams near the cantilever end was used as an actuator to excite the structure using a sweep sine. Based on the results of both cases, there was an insignificant change in the ability to detect the damage, with delamination B being an exception. In the case of the same side sensor/damage configuration, it was difficult to detect large C. S. Hamey et al. Figure 6 Experimental Damage Identification of Carbon/Epoxy Composite 341 Illustration of the sensors when placed on a large delamination delamination such as delamination B. This was probably due to the fact that the delaminated portion of beam has an apparent independent vibration. When a sensor was located on the delamination region as illustrated in Figure 6, the same side sensor recorded the curvature of the delaminated area as well, as if it were a short fixed–fixed beam over the delamination span. The experimental results discussed in this article are those with the sensor and the damage on the same side, with exception of the delamination B case, of which the results from the opposite side configuration are discussed. Complete experimental results of the study can be found in Hamey [12]. 4.1 Frequency Measurements The results of frequency measurements from experiments with sweep sine excitation are presented in this section. The first three of the natural frequencies of the damaged beams are compared with the undamaged beam results. In these comparisons, an average of undamaged natural frequencies is used. Both the undamaged beams had slightly different natural frequencies for each mode, of which the maximum difference is around 2.4% at the lowest natural frequency (the first mode). These values are summarised in Table 2. An examination of the natural frequencies reveals a significant change for beams with delaminations (Table 3) and saw-cut damage (Table 4). For the beam with impact damage, the frequencies are relatively unchanged by the presence of damage. In particular, the first natural frequency demonstrates substantial changes, between 11 and 20%. At the second and third frequencies the percentage of change is smaller, but it is still noticeable, with an average of about 3%. Table 2 A comparison of the natural frequencies for the undamaged beams. Mode Undam. 1 Undam. 2 Ave. Undam. % Change 1st 2nd 3rd 33.2 182.2 502.5 32.4 179.3 498.4 32.8 180.8 500.5 2.42 1.57 0.81 Compared to the undamaged beams, the first natural frequency of delamination B, where the sensor was located at the opposite side of the delamination (oppDelam B), has the highest change of 20.6%. This might be due to the lack of the sensors near the fixed end, where the first mode readings were typically analysed. Overall, the values of the frequency changes of Delam C are comparable to those of Delam A, since the delamination length in both the cases are the same, although their locations are different. In the impact damage case, the natural frequencies are actually increased slightly instead of being decreased or relatively unchanged (Table 4). The first natural frequency demonstrated less than 3% of increase. For the saw-cut damage case, the natural frequencies changed quite dramatically. The first is reduced by about 15%, the second changed by 6.4% and the third by almost 4%; these changes are substantial when compared to the changes by the other damage conditions. 4.2 Mode Shapes The undamaged beam curvature mode shapes were generated by averaging the mode shapes from two undamaged beam samples to provide the best representative shape as a base for later comparison. The average mode shapes are shown in Figure 7(a), which are better than either of the other two individually. Even before being 342 Structural Health Monitoring 3(4) Table 3 Comparisons of the natural frequencies of delaminated beams: Delam A, oppDelam B and Delam C. Mode 1st 2nd 3rd Undam. Delam A % Change oppDelam B % Change Delam C % Change 32.8 180.8 500.5 29.12 175.45 481.89 11.34 2.96 3.72 39.61 171.54 492.84 20.60 5.13 1.53 29.23 179.06 489.01 11.01 0.97 2.30 Table 4 Comparisons of the natural frequencies of impact and saw-cut damaged beams. Mode 1st 2nd 3rd Undam. Impact % Change Saw-cut % Change 32.8 180.8 500.5 33.82 181.78 505.06 2.97 0.53 0.91 27.99 169.23 481.99 14.77 6.40 3.70 weighted, the average shapes more closely resemble the theoretical shapes. Based on a visual inspection of the curvature mode shapes (Figure 7(b) and (c)), the damage location of delamination A and B cannot be discerned. The mode shapes are not as smooth as the ones obtained from the undamaged beams, even after weighting them, the damage location is still not recognisable. For Delam C, the damage location could be somewhat discerned (Figure 7(d)). In all three modes there was some distinct pattern of shapes around the locations of sensors 5, 6 and 7. This became more evident after the shapes were weighted. In particular, at point 5 where the delamination began, the change could be easily recognised. Similar indication is also noticed at the curvature mode shapes of impact-damaged beam (Figure 7(e)). The impact damage location could be predicted around location 5. For the saw-cut damaged beam, the damage location could be somewhat discerned from mode 1, around location 6 (Figure 7(f )). It is clearly evident that in the first mode there was some mode difference at location 6. Yet, in the second and third modes, there is no distinction around location 6, from which a prediction of the damage location could be made. 4.3 Damage Identification Analysis In this section, damage identification results are presented. The four damage detection algorithms introduced in Section 2 are used to locate the damage in the composite beams. Applying the ADM to the data of Delam A indicated that around the damage location (sensor 3) there is a peak in the first mode (Figure 8(a)). However, there are also peaks at locations 6 and 7 and at imperfection location (sensor 10). Thus any single mode cannot detect the location of damage in this instance. The CDF also failed to properly locate the damage for Delam A (Figure 8(a)). Neither the DIM nor the FCM fared any better on this damage condition (see Figure 8(b)). Although both the methods did show peaks at the location of sensor 4, other peaks were also recognisable. Hence, all the methods failed to exclusively locate the damaged area for beam Delam A. Some of the methods did have peaks around the damage location; however peaks at other nondamaged location were also present. The inability to locate the delamination may be due to the fact that the damage location lies near modal nodes in two of the three modes examined. Acquiring additional modes by refinement of sensors and thus the ability to more properly judge higher modes may aid in alleviating this problem. 4.3.1 Delaminated Beam ‘A’ 4.3.2 Delaminated Beam ‘B’ For the delaminated beam B, the application of ADM resulted in some peaks within the damaged area (sensors 2–6) for all three modes (Figure 9(a)) and around the imperfection (locations 10 and 11). The peak that was located near sensor 14 in the third mode caused concern, since there was no explanation for this peak. The CDF estimated that locations 6 and 10 (Figure 9(a)) as possible damage and C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite Sensor location Sensor location Sensor location Sensor location (a) (b) Sensor location Sensor location Sensor location Sensor location (c) (d) Sensor location Sensor location Sensor location Sensor location (e) 343 (f) Figure 7 The first three curvature mode shapes, from sweep sine excitation experiment, of (a) the average undamaged beam; (b) Delam A beam; (c) oppDelam B beam; (d) Delam C beam; (e) impact-damaged beam and (f) saw-cut damaged beam. 344 Structural Health Monitoring 3(4) Damage index FRF curvature Damage index CDF ABS diff FRF curvature Sensor location Sensor location (a) (b) Figure 8 Detection results based on curvature shapes experimental data for Delam A, from (a) absolute difference and CDF methods, and (b) damage index and FRF curvature methods. Damage index Sensor location (a) FRF curvature Damage index CDF ABS diff FRF curvature Sensor location (b) Figure 9 Detection results based on curvature shapes experimental data for oppDelam B beam from (a) absolute difference and CDF methods, and (b) damage index and FRF curvature methods. imperfection locations, respectively. The CDF captured a peak at location 14 as well. Both the DIM and FCM (Figure 9(b)) were able to locate the damage and the imperfection. The DIM had two peaks around the damage boundaries: one at location 4 and the other at near location 6, although false indication at location 14 was also captured. The FCM also had two peaks, one large gradual peak spanning between locations 4 and 6 where the delamination was located and the other peak was around location 10. In conclusion, the location of delamination B that spanned locations 2 to 6 was identified by all three multiple mode methods. The delamination was recognised around locations 4, 5, 6 and 7. The imperfection discussed earlier was also detected at locations 9 and 10. 4.3.3 Delaminated Beam ‘C’ The ADM on each mode of Delam C identified the damage location C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite 345 FRF curvature Damage index CDF ABS diff Damage index FRF curvature Sensor location Sensor location (b) (a) Figure 10 Detection results based on curvature shapes experimental data for Delam C, from (a) absolute difference and CDF methods, and (b) damage index and FRF curvature methods. in the area between locations 4 and 8, which are close to the actual location (5–7) (Figure 10(a)). The first mode depicts the delamination area by showing two peaks at the locations of sensors 4 and 8. The CDF also displays a large peak along the damaged area (Figure 10(a)). Similar peaks were also displayed at the imperfection location of sensor 10. The DIM was able to accurately determine the location of the whole delamination by displaying a peak through locations 6, 7 and 8 (Figure 10(b)). A small peak was also located at point 10 as expected. The FCM was partially able to detect the damage (location 7). However, several peaks were also located in other areas as well. All the methods discussed were capable of locating the delamination C, with the exception of the FCM. The location of this damage condition made it possible to generate the first three mode shapes without having a modal node within the vicinity of the delamination. It is demonstrated that the DIM could more accurately detect the damage location compared to the other methods. In the case of impact beam, all three modes from the ADM showed peaks around location 6 (Figure 11(a)). Only the third mode had multiple peaks, with the second peak being located at sensor 11. Hence, it can be said confidently that the impact damage 4.3.4 Impact Damaged Beam location was around location 6. The CDF and FCM also displayed a significant peak around the damaged area (Figure 11(a)) with a small peak at the imperfection location. The DIM estimated accurately the location of the whole damage by displaying a peak through locations 6 and 7 (Figure 11(b)). The location of impact damage was identified properly by all the methods. The ADM and CDF methods gave a better estimation of the damage location (within 12.7 mm or 0.5 in. of the actual location); whereas the DIM and the FCM were both off by 25.4 mm (1.0 in.). Saw-Cut Damaged Beam For the saw-cut beam, only the ADM was able to display the damage location by demonstrating peaks around location 6, especially in the first and second modes (Figure 12(a)), although, for mode 3, peaks at other locations were also significantly large. In all of the other methods, the identifications near the damage location were overshadowed by peaks at other locations. The CDF displayed a large plateau over the damage location. However, it failed to accurately locate the damage since the most prominent peak was found at location 9. The DIM and FCM were unable to determine the damage location. Peaks at locations 9, 10 and 14 (Figure 12(b)) were more significant. 4.3.5 346 Structural Health Monitoring 3(4) FRF curvature Damage index CDF ABS diff Damage index FRF curvature Sensor location Sensor location (b) (a) Figure 11 Detection results based on curvature shapes experimental data for impact-damaged beam from (a) absolute difference and CDF methods, and (b) damage index and FRF curvature methods. FRF curvature Damage index CDF ABS diff Damage index FRF curvature Sensor location Sensor location (a) (b) Figure 12 Detection results based on curvature shapes experimental data for saw-cut damaged beam from (a) absolute difference and CDF methods, and (b) damage index and FRF curvature methods. 5 Damage Detection Using Impulse Hammer Excitation Experiments using the impulse hammer had the same damage configuration specimens described earlier. The tests were conducted with the sensors mounted on the same side of the beam where the damage was located. The FRF for impulse hammer excitation often contains a large amount of noise at locations where the natural frequencies do not exist. Due to the high amounts of noise in these regions, the FRF curvature method cannot effectively be applied to the FRF data obtained from impulse hammer excitations. For this reason, no FRF curvature method data is presented in this section. 5.1 Frequency Measurements The results of natural frequencies from experiments with the impulse hammer excitation are comparable to the test results using the sweep sine excitation, although all the first three natural frequencies excited by impulse hammer are a little higher. Summary of the results of the two C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite undamaged beams and their averages are presented in Table 5. In general, the changes in natural frequencies due to damage measured by impulse hammer experiment are much smaller than those with sweep sine excitation (Tables 6 and 7). The percentage of change is between 1.4 and 8.8%, ignoring changes below 1%. Some changes in the second mode are not significant (0.78% and lower) due to the fact that the location of damage is near to or at the location of modal nodes, in this case delamination A, B and impact damage. Similarly, for the third mode of saw-cut damaged beam, the location of damage is near the node of mode 3. The impact damage relatively unchanged the natural frequencies (Table 7). Results of hamDelam C indicated a different trend than those presented in hamDelam A or Delam C. The frequency changes due to saw cut also exhibited a different trend from the test using sweep sine excitation. The inconsistency in the frequency changes might be due to the measurement of natural frequency, which could slightly drift depending on the measuring equipment, 347 weather conditions, background noise, ambient vibrations, and the inability to accurately repeat the initial boundary conditions. 5.2 Mode Shapes The curvature mode shapes generated for each undamaged beam were also averaged to generate the best shape. The average mode shapes from undamaged beams 1 and 2 are displayed in Figure 13(a)). The curvature mode shapes generated by measurement with impulse hammer excitation were less smooth compared to the curvatures generated by sweep sine excitation. The lack of smoothness in the mode shapes increases the difficulty in determining a location solely based on their appearance. Based on visual inspection, a flattening at locations 2–4 in the first mode and small peak at location 3 may indicate the location of delamination A (Figure 13(b)). However, this observation is not conclusive. Especially after weighting them, the damage location is not recognisable. For delamination B, the indication of damage location is more pronounced (Figure 13(c)). There Table 5 Comparison of the natural frequencies for the undamaged beams excited by impulse hammer. Mode hamUndam. 1 hamUndam. 2 Ave. hamUndam. % Change 29.17 178.38 469.47 30.18 177.27 487.46 29.67 177.82 478.47 3.38 0.63 3.76 1st 2nd 3rd Table 6 Comparisons of the natural frequencies of delaminated beams from impulse hammer experiments: hamDelam A, hamDelam B and hamDelam C beams. Mode 1st 2nd 3rd hamUndam. hamDelam A % Change hamDelam B % Change hamDelam C % Change 29.67 177.82 478.47 30.43 177.73 493.11 2.53 0.05 3.06 32.29 176.43 517.35 8.81 0.78 8.13 32.08 184.81 501.33 8.10 3.93 4.78 Table 7 Comparisons of the natural frequencies of delaminated beams from impulse hammer experiments: hamImpact and hamSaw-Cut damaged beams. Mode 1st 2nd 3rd hamUndam hamImpact % Change hamSaw-Cut % Change 29.67 177.82 478.47 29.26 177.04 495.59 1.41 0.06 3.58 29.60 170.62 478.90 0.25 4.05 0.09 348 Structural Health Monitoring 3(4) Sensor location Sensor location Sensor location Sensor location (a) (b) Sensor location Sensor location Sensor location Sensor location (c) (d) Sensor location Sensor location Sensor location Sensor location (e) (f) Figure 13 The first three curvature mode shapes, from impulse hammer excitation experiment, of (a) the average undamaged beam; (b) hamDelam A beam; (c) hamDelam B beam; (d) hamDelam C beam; (e) hamImpact damaged beam and (f) hamSaw-Cut damaged beam. C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite are clear peaks in the first and second modes at locations 3 or 4. This became less evident after the shapes were weighted; but it was still recognisable. Additionally, the second mode clearly indicates the existence of imperfection at location 10. The minor indications of delamination C were exhibited by all the three mode shapes (Figure 13(d)), between locations 5 and 7. This became more evident after the shapes were weighted. In particular at point 5, the change could easily be recognised. Small inconsistencies in all three modes around locations 5 and 6 (Figure 13(e)) indicate the location of impact damage. Saw-cut damage effect on the first mode shape (Figure 13(f )) was quite obvious, where there were some variations at locations 5, 6 and 7, which surround the damage location. However, in the second and third modes, there is no distinction around the point from which a prediction of the damage location could be made. 349 between sensors 2–4. Yet, the location of the imperfection was identified quite properly by all the methods except the absolute difference of the first mode. These results reinforce the earlier remarks that the location of the damage, which lies near the modal nodes in two of the three modes examined, made damage identification to be not feasible. Refinement of sensor configuration may solve this problem. The first and third modes of ADM showed partial damage boundary locations (Figure 15(a)), at locations 4 and 7, respectively. The location of the damage fell at or near the modal node of the second mode and made it difficult for one to detect the damage using this mode. The CDF also displayed a double peak situated at the beginning and end of the delamination (Figure 15(b)). The DIM for hamDelam B had peaks at locations 2 and 7, which corresponded with the limits of the delamination. A peak near location 10 was not very recognisable. Although, the three methods presented in this article were able to locate some part of the delamination B, such large delamination may be difficult to detect completely, considering the probability that some part of damage is located at the vicinity of modal nodes. 5.3.2 Delaminated Beam ‘B’ 5.3 Damage Identification Analysis 5.3.1 Delaminated Beam ‘A’ Similar to the results from sweep sine excitation, all the methods could not properly locate the damaged area (Figure 14). A large peak at location 6 in all the three modes gave a false indication of damage location since the actual location is Delaminated Beam ‘C’ Using the first mode of the ADM, the damage was estimated 5.3.3 Sensor location (a) CDF ABS diff Damage index i Sensor location (b) Figure 14 Detection results based on curvature shapes experimental data for hamDelam A beam, from (a) absolute difference methods, and (b) damage index and CDF methods. 350 Structural Health Monitoring 3(4) CDF ABS diff Damage index i Sensor location Sensor location (a) (b) Figure 15 Detection results based on curvature shapes experimental data for hamDelam B beam, from (a) absolute difference methods, and (b) damage index and CDF methods. Sensor location (a) CDF ABS diff Damage index i Sensor location (b) Figure 16 Detection results based on curvature shapes experimental data for hamDelam C, from (a) absolute difference methods, and (b) damage index and CDF methods. at locations 3 and 6, whereas from both the second and third modes at locations 5, 7 and 8 (Figure 16(a)). When combined together in the CDF method (Figure 16(b)), there was a large peak spanning through these four locations. This provided a good indication of the damage, since the actual damage location is between locations 5 and 7. In addition, the imperfection at location 10 was also identified. The DIM identified the location of the whole delamination by displaying peaks through locations 6, 7 and 8. This result was off by one location, since the actual damage is located between locations 5 and 7. A peak indicating that the imperfection was located at location 10 was obtained as expected. These results corresponded closely with the results from the sweep sine excitation tests for beam Delam C. 5.3.4 Impact Damaged Beam The ADM showed that all the three modes produced peaks around location 6 (Figure 17(a)). Only the third mode had multiple peaks with the second peak located at location 11. Because the peak at location 6 appeared in all the three modes, it was determined reliably that the damage is C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite 351 Sensor location (a) CDF ABS diff Damage index i Sensor location (b) Figure 17 Detection results based on curvature shapes experimental data for hamImpact damaged beam, from (a) absolute difference methods, and (b) damage index and CDF methods. located near location 6. The CDF also displayed a large peak along the damaged area (Figure 17(b)), while the other peaks were quite small. The DIM was also able to determine the damage location by displaying a peak through location 6. In this damage configuration, the damage location was identified accurately by all the three methods with a similar trend of indication. 5.3.5 Saw-Cut Damaged Beam The ADM indicated that the first and third modes showed peaks around damage location (i.e., location 6) (Figure 18(a)), although peaks at other locations were quite significant as well. The second mode had undulation with the highest peak at location 11. The CDF displayed a peak over the damage location (Figure 18(b)). However, it generated peaks at locations 9 and 11. The DIM identified the damage location more accurately by displaying a substantially more dominant peak at location 6. All the methods were able to moderately localise the damage around location 6 for beam hamSaw-Cut. 6 Conclusions In general, the damage detection methods with an impulse hammer excitation generated better identification results compared to the continuous excitation using PZT. Delamination C, the impact damage, and the saw-cut damage were identified properly by the curvature mode-based damage detection technique presented in this study. However, for delamination A and delamination B, the results are limited and inconclusive. Damage configuration and location affect the ability of the method. From this study of using both the sweep sine and impulse hammer excitations, the following concluding remarks can be drawn: 1. Condition delamination A does not contain a damage configuration that is conducive to identification of the damage appropriately. This limits the method to properly identify damages that are in close proximity to the clamped end and/or modal node points. 2. The Damage Index Method (DIM) detects and isolates the damages better than any of the other methods studied. The FRF Curvature Method (FCM) does not seem to work as well as the other methods. However, the FCM may work better with a system that generates smooth FRF curves. 3. A large delamination, as in the condition of delamination B, might be identified by multiple peaks at the edges of the delamination by all the methods under study. However, this could cause 352 Structural Health Monitoring 3(4) CDF ABS diff Damage index i Sensor location Sensor location (b) (a) Figure 18 Detection results based on curvature shapes experimental data for hamSaw-Cut damaged beam, from (a) absolute difference methods, and (b) damage index and CDF methods. misleading interpretations, such that the peaks are viewed as a multiple instance of some highly localised damage. 4. For the large delamination configuration (e.g., in delamination B), identification procedure will generate better results when the sensors were located opposite to the delamination side. This will reduce the effect of vibration of the delaminated part. In the cases of relatively localised damage, such as the impact damage or saw-cut damage, the location of the sensor with respect to the damage side has little effect on the identification results. 5. The frequency changes varied widely from one test to another and from sample to sample, especially at low natural frequencies. Thus, frequencies are inadequate to be used as a parameter in the damage magnitude prediction. 6. Finally, all the methods presented exhibited that the curvature modes measured by the piezoelectric sensors can be used as promising alternatives in damage detection techniques. The excitation sources used in this study, impulse hammer and continuous sweep sine excitations, work equally well, and the results were often nearly identical. Therefore, each excitation type with its own benefits, is a good candidate for being an excitation source in the implementation of a detection method. Acknowledgements The carbon/epoxy composite test samples used in this study were provided by Honeywell, Inc. This study was partially supported by the College of Engineering at the University of Akron and Ohio Aerospace Institute – Collaborative Core Research Program (OAI-CCRP#2002-04). References 1. Rizos, P.F. Aspragathos N. and Dimarogonas, A. (1990). Identification of crack location and magnitude in a cantilever beam from the vibration modes. Journal of Sound and Vibration, 138(3), 381–388. 2. Salawu, O. (1997). Detection of structural damage through changes in frequency: a review. Engineering Structures, 19, 718–723. 3. Pandey, A.K., Biswas, M. and Samman, M.M. (1991). Damage detection from changes in curvature mode shapes. Journal of Sound and Vibration, 145(2), 321–332. 4. Luo, H. and Hanagud, S. (1997). An integral equation for changes in the structural dynamics characteristics of damaged structures. Int. J. Solids Structures, 34, 4557–4579. 5. Wahab, M. and De Roeck, G. (1999). Damage detection in bridges using modal curvatures: application to a real damage scenario. Journal of Sound and Vibration, 226(2), 217–235. 6. Wahab, M. (2001) Effect of modal curvature on damage detection using model updating, Mechanical Systems and Signal Processing, 15(2), 439–445. C. S. Hamey et al. Experimental Damage Identification of Carbon/Epoxy Composite 7. Lestari, W. and Hanagud, S. (2001). Detection of an edge notch defect by using a single model based method. In: Proceedings of 3rd International Workshop, Stanford University, Stanford, CA, pp. 1343–1355. 8. Wang, B. and Wang, B. (1997). Feasibility analysis of using piezoceramic transducers for cantilever beam model testing. Smart Materials and Structures, 6, 106–116. 9. Sampaio, B. Maia, N. and Silva, J. (1999). Damage detection using the frequency-response-function curvature method. Journal of Sound and Vibration, 226(5), 1029–1042. 353 10. Farrar, C. and Jaurequi, D. (1998). Comparative study of damage identification algorithms applied to a bridge: I Experiment. Smart Materials and Structures, 7, 709–719. 11. Farrar, C. and Jaurequi, D. (1998). Comparative study of damage identification algorithms applied to a bridge: II Numerical study. Smart Materials and Structures, 7, 720–731. 12. Hamey, C.S. (2003). Dynamic response-based damage detection of composite structures using smart materials. MS Thesis, The University of Akron, Akron, OH.