Smart lightning protection skin for real-time load monitoring of composite aircraft structures under multiple impacts

Composites: Part A 67 (2014) 44–54

Contents lists available at ScienceDirect

Composites: Part A journal homepage: www.elsevier.com/locate/compositesa

Smart lightning protection skin for real-time load monitoring of composite aircraft structures under multiple impacts Yoshiro Suzuki ⇑, Toyoaki Suzuki, Akira Todoroki, Yoshihiro Mizutani Tokyo Institute of Technology, Department of Mechanical Sciences and Engineering, Box I1-50, 2-12-1 Ookayama, Meguro-ku, Tokyo 152-8552, Japan

a r t i c l e

i n f o

Article history: Received 18 April 2014 Received in revised form 1 July 2014 Accepted 10 August 2014 Available online 19 August 2014 Keywords: B. Delamination B. Impact behavior D. Non-destructive testing

a b s t r a c t We experimentally test a previously reported lightning protection sheet (LPS) sensor to detect impacts within a short collision interval. The exact loading position is achieved using a two-step method. Our sensor sheet is flexible and can easily be attached to different shaped structures. Current flows in the sensor only during loading and our quick switching system identifies the real-time x, y coordinates of the loading position. Using this method multiple impacts can be detected when the time difference between two impacts is larger than 10 ms. The position of a single impact is estimated within 16 mm. This means that detailed inspection needs to be conducted only over a circular area with a radius about a few centimeters. Our method is ideally suited for aircraft components and it reduces costs and saves time as only small specific areas require inspection using ultrasonic or electrical resistance techniques. Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Carbon-fiber-reinforced polymers (CFRPs) have been used as a structural material in commercial airplanes. For instance, the Boeing 787 Dreamliner and Airbus A350 XWB use 50% and 53% CFRP by weight, respectively. The reliability of real-time inspections for CFRP components has become important. Delamination damage due to an external impact is a major concern in the design of CFRP structures. In laminated CFRPs, an out-of-plane impact can cause layers to separate with significant loss of the load-carrying capability of the structures, particularly compressive strength [1–3]. Delamination can be caused by the following impact loads.  Hailstorm during flight and bird strike during takeoff and landing.  Tailstrike during takeoff and landing that is an event in which the rear end of an airplane hits the runway.  Minor collision with other vehicles in airport such as passenger steps vehicle, cargo loader, refueling car, belt loader, towing tractor, scissor car, water wagon and so on.  Tool drop and minor collision with a working bench in hangar when the aircraft is in maintenance. Delamination is an internal damage and difficult to detect visually. Therefore ultrasonic [4], coin tapping, infrared thermography ⇑ Corresponding author. Tel./fax: +81 3 5734 3184. E-mail address: [email protected] (Y. Suzuki). http://dx.doi.org/10.1016/j.compositesa.2014.08.010 1359-835X/Ó 2014 Elsevier Ltd. All rights reserved.

[5,6], and X-ray inspections must be performed over the entire structure; these methods are time consuming and require putting the airplane out of service. A real-time system that can detect an impact and specify the position automatically would benefit flight safety and economic efficiency. As fiber-reinforced polymer (FRP) materials have lower electrical conductivity than ordinary metal, they are easy to be damaged by heat generated by a large current from lightning strikes [7,8]. High electrical conductivity and thermal conductivity could help restrain the damage by spreading current and heat from the lightning strike. To enhance both the conductivities of the FRPs, various methods are presented for adding and dispersing tiny conductive fibers or particles (e.g., carbon nanotubes [9]) in the FRPs, however, they are unpractical at least for the commercial aircraft because of the low reliability. Present composite structures of commercial aircraft are covered with a thin metal mesh or film (usually copper or aluminum) that provides lightning protection [10–12] to the airplane as shown in Fig. 1. In a previous study [13,14], we developed an in-situ method for detecting and locating impact loading by implementing resistive touchpad techniques [15–19] on a lightning protection shield (LPS). The LPS is an integral part of the load sensor circuitry and there has been little discussion so far on the repurposing of LPSs [20,21]. Our proposed sensor is a flexible thin sheet that can be attached to a curved surface. It requires no special equipment, uses a voltmeter and a limited number of power supplies, and can be replaced easily if broken. Notifying the aircrew in real time during flight that the airplane structure has been impacted would help

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

45

Fig. 1. Schematic of the proposed sensor sheet that has two functions: lightning protection and load sensing. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

prevent serious accidents. The sensor also works on ground. For instance, it can detect a minor collision with other vehicles in airport and a tool drop when the aircraft is in maintenance. Our approach only detect an impact applied to the structure and does not detect internal damage, so other methods such as ultrasonic or electrical resistance methods [20–22] are required for detailed inspection of internal damage after detecting the impact by the proposed sensor. However, as only specified areas require inspection, this reduces the total costs and total time of the inspection. In the previous study [13,14], the position of a static indentation load applied to the sensor was estimated with an average error of 9 mm. However, the sensor has not yet been experimentally tested under real-time impacts. For practical use, the sensor sheet must locate impact loading in large FRP structures within a short collision interval. The sensor must also detect several loads if there are multiple impacts within a short time range. The present study employs the sensing method for real-time identification of positions of multiple-impact loads. To identify the real-time x, y coordinates of the loading position, we use a quick switching system. The differences between the proposed sensor in this manuscript and the sensor in our previous articles [13,14] are summarized as follows.  Real-time position estimation for a dynamic impact load applied to the sensor has not been tested in the previous study; however, it will be experimentally investigated in this manuscript (Section 4.2. Single-impact loading).  Although the previous sensor detects only single load, the new sensor can detect multiple impact loads hitting almost simultaneously (Section 4.3. Multiple-impact loading).  The previous sensor sheet, which is 100 mm  300 mm in size, estimates the load position with an average error of 9 mm. The new sensor sheet, which is 1000 mm  1000 mm in size, identifies the load position with an average error of 16 mm (See Section 4.2). As the new sensing system consists of a two-step estimation technique for the load position (Sections 2.2.1 and 2.2.2) and therefore hardly reduces the estimation accuracy even when the sensing area (i.e., area of an inspection object) becomes large. Conversely, the previous

sensor has just one-step position estimation and generates a position error that would be proportional to the length of the sensing area. 2. Principle of position specification for impact loading 2.1. Configuration of the sensor Fig. 1 shows the proposed sensor sheet composed of the following three layers.  Layer 1 is a thin copper sheet, acting as both the LPS and the signal path between the load point and voltmeter. This layer has cylindrical projections that protrude downward.  Layer 2 is a nonconductive silicone rubber sheet performing as load support and insulator between layers 1 and 3. The sheet is equipped with circular holes with diameter slightly larger than that of the cylindrical protrusion.  Layer 3 is a circuit board consisting of resistive rectangles arranged orthogonal to each other.  Layer 4 (not shown in Fig. 1) is a thin nonconductive layer that electrically insulates layers 1–3 from an FRP aircraft structure. The nonconductive layer prevents a lightning current from entering the FRP structure [10] and makes the sensor properly estimate position of an impact load. 2.2. Procedure for position specification As shown in a flowchart of the proposed method (Fig. 2), the task of position estimation is divided into two steps: specification of the loaded section and estimation of the exact position in the specified section. This two-step method results in a high degree of accuracy over large areas such as those of aircraft components. 2.2.1. Step 1: Specification of a loaded section in the entire structure Fig. 3 is a schematic of layer 3 and the switching circuit. The resistive sheets are arranged in rectangles orthogonal to each other, named 1, 2, 3,. . . and A, B, C,. . .. For example, section 3A is where rectangles 3 and A overlap. When an impact load is applied

46

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

Fig. 2. Flowchart of the procedure in the impact load identification.

Fig. 3. Schematic of layer 3 under layers 1 and 2 explaining how to specify the impacted section from the entire structure. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

to section 3A, layer 1 has electrical contact with both rectangles 3 and A. To specify an impacted rectangle from rectangles 1–3, a voltage of Vref is applied across rectangles 1–3 in the x direction as shown in Fig. 4(a). Assuming that the voltage drop across a resistive rectangle is linearly proportional to the x coordinate, the relation between the voltage read from layer 1, Vx1, and the estimated x coordinate, x1, is

x1 ¼

V x1 Lx ; V ref

ð1Þ

with

0 6 x1 < L3x Lx 6 x1 < 2L3x 3 2Lx 6 x1 < L x 3

if

rectangle 1 is loaded

if

rectangle 2 is loaded :

if

rectangle 3 is loaded

Lx is the length of layer 3 along the x axis. The impact causes layer 1 and the loaded rectangle to make electrical contact with each other. When rectangle 3 is impacted (i.e., 2Lx/3 < x1 < Lx), Vx1 becomes within one-third to two-thirds of Vref. Therefore, the loaded rectangle can be specified; this process is then repeated for the y direction (Fig. 4(b)), giving Vy1 in the following equation.

y1 ¼

V y1 Ly ; V ref

ð2Þ

with 2Ly 6 y1 < Ly 3 Ly 2L 6 y1 < 3y 3 L 0 6 y1 < 3y

if

rectangle A is loaded

if

rectangle B is loaded :

if

rectangle C is loaded

where y1 is the y coordinate estimated in the first step, Vy1 is the voltage of layer 1 when applying Vref across rectangles A–C along the y axis, and Ly is the length of layer 3 along the y axis. When 2Lx/3 < x1 < Lx and 2Ly/3 < y1 < Ly, 3A is identified as the impacted section. 2.2.2. Step 2: Estimation of the exact position of loading After specifying the loaded section (3A in this case), a more precise position for the impact is estimated using resistive touchpad techniques [15–19]). By applying a voltage of Vref across each of rectangles 1–3 along the x axis and measuring the voltage from layer 1 as shown in Fig. 4(c), the x coordinate, x2, can be estimated according to

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

47

Fig. 4. Electrical potential gradient applied to layer 3 in estimating the coordinates of the load position, x1, x2, y1, and y2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

x2 ¼

V x2 lx ; V ref

ð3Þ

where x2 is the x coordinate estimated in the second step, Vx2 is the voltage of layer 1 when applying Vref across each of rectangles 1–3 along the x axis, and lx is the length of each section along the x axis. Once x2 has been obtained, the voltage is removed and a potential difference of Vref is applied across each of rectangles A–C in the y direction. The linear potential gradient along the y axis, Vy2, is then measured on layer 1 as shown in Fig. 4(d).

y2 ¼

V y2 ly ; V ref

ð4Þ

where y2 is the y coordinate estimated in the second step, Vy2 is the voltage of layer 1 when applying Vref across each of rectangles A–C along the y axis, and ly is the length of each section along the y axis. Note that the position x2, y2 estimated in the second step is regarded as more accurate than x1, y1 estimated in the first step. There are two reasons for this. The first is that, in the second step, layer 2 has fixed potential lines at intervals shorter than those in the first step. The second reason is that layer 3 has a gradient of voltage in the second step that is three times steeper than that in the first step. The steeper gradient leads to a higher signal to noise ratio. 2.2.3. Switching system All of the above coordinates x1, x2, y1, and y2 of the loading point must be measured within a short collision interval. The collision interval of an impact applied to the sensor depends on various conditions (e.g., the type of measuring materials, impact energy and velocity, and spacing of supporting parts between layers 1 and 3). However, it would always be possible to finish the position specification during loading by setting the switching time appropriately. A microcomputer and eight transistors T1, T2, ..., T8 as shown in Fig. 3 are used to switch the connection between the power supply and resistive rectangles in each quarter interval so that the entire processes can be completed within the collision interval. The DC power supply outputs a constant voltage of Vref between its anode terminal and the ground. Each transistor has three terminals, named

the base, collector, and emitter. A small current at the base terminal can switch another current between the other two terminals (from the collector to the emitter). Each edge of rectangles 1–3 and A–C is connected with the power supply or ground through these two terminals of respective transistors. Respective base terminals are connected with the respective output terminals of the microcomputer so that on–off control of the transistors is operable independently at a different timing. In other words, the transistors can connect or disconnect each edge of all the rectangles with or from the power supply or ground at an arbitrary timing. When only transistors T1 and T4 are activated (Fig. 4(a)), there is a potential difference in Vref across rectangles 1–3 along the x direction, giving Vx1 in Eq. (1). If four transistors T1–T4 are activated and Vref is applied across each of rectangles 1–3 in the x direction (Fig. 4(c)), Vx2 in Eq. (3) can be measured. These processes are then repeated for the y direction to measure Vy1 and Vy2 in Eqs. (2) and (4). Setting the total measurement period lower than the collision interval enables the sensor to finish estimating all of x1, x2, y1, and y2 before the collision object separates from the sensor. 2.3. Advantages over other inspection techniques In comparison with other inspection techniques, the proposed sensor sheet has the following advantages. 1.

2.

3.

4.

The proposed sensor can monitor the impact and indentation load on a structure regardless of the shape or construction material of the structure. Regardless of the size of the inspection target, the proposed two-step localization of loading enables both monitoring of the whole structure within half a millisecond and highly accurate estimation of the load position. The sensor can monitor several impacts separately when the difference in collision times between the impacts is larger than 10 ms. The LPS plays three roles: lightning protection, a load sensor, and wiring. This minimizes wiring and structural weight.

48

5.

6. 7.

8.

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

There is no electrical current flow or power consumption unless a load is applied. Current flows in the sensor only during loading. These factors greatly reduce cost. The sensor sheet is flexible and affixed to the structure surface. There is no need to embed foreign material under the surface. There is no risk of a reduction in structural strength due to embedding sensors Partial repair and replacement of the surface mount sheet are easy.

3. Experimental procedure 3.1. Sensor configuration Fig. 5 shows the detailed configuration of the proposed threelayer sensor used in experiments. The sensor has four sections: the upper left is section 1A; the upper right is section 2A; the lower left is section 1B; the lower right is section 2B. The length and width of these sections are lx = 500 mm and ly = 500 mm. In the first step of the load identification, we activated only transistors T1 and T4 (2SC1815, Silicone NPN Epitaxial Type, Toshiba Corp., Tokyo, Japan) to measure Vx1 in Eq. (1). In the second step, transistor T4 was deactivated and only T1–T3 were activated to obtain Vx2 in Eq. (3). Transistors T1–T3 were then deactivated and Vy1 and Vy2 were measured by using T5–T8 in a similar way. Note that Vx1 and Vy1 enabled us to specify which section was loaded. Therefore, we identified the loaded section and the exact load position on the loaded section from Vx1, Vx2, Vy1 and Vy2 using Eqs. (1)–(4). On– off switching of the transistors was controlled by a microcomputer (PIC16F84A-20I/P, Microchip Technology Inc., Tokyo, Japan). As shown in the enlarged cross-sectional view of Fig. 5, the copper sheet and cylindrical protrusion are 0.7 and 0.3 mm in thickness, respectively. The thickness of a silicone rubber sheet (layer 2) is 0.5 mm and greater than that of the protrusion by 0.2 mm. The total thickness of the three-layer sheet is less than 2.0 mm. The copper protrusions are located at intervals of 5.0 mm. The silicone rubber (Shore hardness of 90) has circular holes that are coaxially located with the protrusions. Layer 3 is a circuit board having a grid of orthogonal copper wires under each protrusion (left of Fig. 5). Only the horizontal wire around the grid passes on the backside of layer 3 so that the two wires make no contact. Ends of these wires are connected with high-resistance nichrome wires (diameter: 0.05 mm, resistance per meter: 570 X/m). If the load is smaller than a threshold value, layer 1 has no potential as shown in Fig. 6(b). If it exceeds the threshold, the protrusion under the loading point touches two orthogonal copper wires of layer 3 as

shown in Fig. 6(c). The potential of layer 1 then becomes equal to that of the wires. It was assumed that the voltage drop across the nichrome wire was linearly proportional to the x coordinate when applying a potential difference between the two ends of the horizontal nichrome wire. The potential of layer 1 was measured and the x coordinate of the loading point was calculated. Layer 1 voltage was measured with a voltmeter (PCD-300A, Kyowa Electronic Instruments Co., Ltd., Tokyo, Japan) when applying a voltage of Vref = 4 V with a stabilized DC power supply (Dual-Tracking Multi-Output DC Power Supply, PMM35-1.2DU, Kikusui Electronics Corp., Kanagawa, Japan) to the nichrome wires. The internal electrical resistance of ordinary voltmeters is much higher than the resistance of layer 1 and the contact resistance at the loading point. Therefore, the voltage drop across layer 1 and the contact between the protrusion and copper wires does not affect the estimation accuracy of the load position. The diameters of the copper protrusion and circular holes are 2.5 and 3.0 mm, respectively. The diameters are regarded as much smaller than those of ordinary collision objects. Therefore, the silicone rubber around the hole is compressed to three-fifths of the original thickness when an impact causes layers 1 and 3 to make contact with each other. In other words, a small load does not bend layer 1 so as to make the protrusion touch the copper wires unless the silicone rubber is largely compressed. This prevents a low indentation such as regular aerodynamic loading during flight from making the contact. 3.2. Quasi-static indentation The proposed sensor must detect a load larger than that that can cause damage to structures. The extent of an out-of-plane load that can cause damage to composite structures depends on several factors such as the thickness, the stacking sequence, and the material type of the structures. There has been little information about relation between the magnitude of an impact load and the extent of the damage in the laminated composite materials because the impact damage is usually related to the kinetic impact energy and recorded. In a literature [23], as a result of an impact test with cross-ply glass fiber/epoxy polymer laminates of 2.92 mm in thickness, delamination cracking of 130 mm2 or larger in area was observed in the case where the peak value of the contact force was larger than 870 N, and nothing or delamination of smaller than 30 mm2 was detected when the peak load was smaller than 630 N. This indicates that an impact load whose peak value exceeds approximately 630 N can initiate delamination cracking in the glass fiber laminates. Main composite structures of commercial aircraft (e.g., main wing, empennage, and fuselage) are thicker than

Fig. 5. Detailed configuration of the proposed three-layer load sensor. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

49

Fig. 6. Schematic of a cross section of the load sensor. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

the laminates and therefore the value of a load causing delamination would be larger. With the above in mind, this manuscript aimed to detect an impact load larger than 1000 N. Additionally, the sensor should be designed not to detect lower loads such as the aerodynamic forces that act on aircraft. For instance, wing loading, P, is calculated from the total weight of the airplane, M, the wing area, S, and the gravitational constant, g = 9.8 m/s2.

P¼

Mg : S

ð5Þ

The Boeing 787 Dreamliner (B787-800) has main wings of S = 325 m2 and its maximum take-off weight M is 228,000 kg. Therefore, the maximum wing load during flight, P, is 6875

Fig. 7. Experimental setup of the static indentation test for investigating the threshold value in detecting a load applied to the sensor. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

N/ m2. As described in Section 3.1, the copper protrusions in the proposed sensor sheet are located at intervals of 5.0 mm. Therefore, square area with a 5.0 mm side length centering at each protrusion receives the maximum load of approximately 0.17 N during flight. From those stated above, the threshold value for detecting a load applied to the sensor should be set in the range of a few newtons to 1000 N. To investigate the threshold value for load detection, we conducted a quasi-static indentation test. Fig. 7 shows the experimental setup. A hemispherical indenter of 15.9 mm diameter was set on top of the sensor sheet. A Teflon sheet was inserted between the indenter and sensor for electrical insulation. The copper layer

Fig. 8. Impactor and strain gage attached to the surface used in the falling-weight impact test.

50

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

(a) 2000 N

Fig. 10. Typical result of the single-impact loading test. Measured relation between the impact load applied at point 1 and the electrical potential of the copper layer (layer 1). (b) Shows the enlarged view of (a). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

(b) 5000 N Fig. 9. Results of repeated quasi-static indentation tests for investigation of the threshold value for load detection: (a) maximum indentation of 2000 N and (b) 5000 N. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

was indented by applying a static load using a universal testing machine (Autograph AG-I/100kN, Shimadzu Co., Kyoto, Japan) at a cross-head speed of 0.1 mm/min. Indentation loads of 2000 N and 5000 N were applied five times at two different positions, respectively. As the objective of this test was to obtain the threshold for load detection and not to estimate the load position, we did not need layer 3. Instead of layer 3, a copper film was laid under layer 2. A uniform potential distribution of 3 V was generated across the film as shown in Fig. 7, and the potential of layer 1 was measured. Although layer 1 had no potential when the indentation load was lower than the threshold, layer 1 had 3 V immediately after the load exceeded the threshold.

diameter as shown in Fig. 8. The impactor was dropped through a guide and the impact energy was calculated by multiplying the falling height by the mass of the impactor. A biaxial strain gage (KFG-2-120-D16-11, Kyowa Electronic Instruments Co., Ltd.) was pasted on the impactor surface to estimate the impact loading. The gage was 30.85 mm from the impact point. The load–strain curve of the gage was previously measured so that the loading on the impactor can be estimated from the strain gage response. The sensor sheet was impacted at each of points 1–3 on the surface by striking by a mass of 1.46 kg. Because the time of contact between the impactor and the sensor was approximately 2.0 ms, the connection between the nichrome wires and the power supply was switched in each period of 0.114 ms by the microcomputer and transistors so that all the coordinates of the loading point (x1, x2, y1, and y2 in Eqs. ((1)–(4)) can be estimated within 0.5 ms. The potential of layer 1 was measured while applying an impact to the sensor. The loading position was measured for every impact load, which verified the reproducibility and dispersions of measurement accuracy of the proposed sensor.

3.3. Single-impact loading 3.4. Multiple-impact loading Drop impact tests were conducted to verify the feasibility of the proposed position estimation of a single impact within a collision interval employing the three-layer sensor as shown in Fig. 5. The test impactor consisted of a long dart with a hemispherically shaped headstock (made of a stainless steel, SUS304) 12.7 mm in

The experimental setup used for the multiple-impact loading test was the same as that illustrated in Section 3.3, except for the drop impact testing machine. The sensor was hit manually at almost the same time with two impactors. Respective values of

Table 1 Repeated static indentation test results investigating fluctuating load detection threshold values. Number of repeated loads Threshold value for load detection (N)

Indentation of 2000 N Indentation of 5000 N

1st

2nd

3rd

4th

5th

Average

Standard deviation

577 484

116 127

143 202

134 116

123 11

219 188

201 179

51

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

Fig. 11. Actual and estimated positions of single-impact loading. Square, triangle, and cross encircled with a dashed frame show the actual (xactual, yactual), first-stepestimated (x1, y1), and second-step-estimated (x2, y2) coordinates of each impact, respectively.

loading were calculated from the responses of biaxial strain gages attached on both impactor surfaces. This experiment was performed several times, changing the time intervals between the two impacts. 4. Results and discussion 4.1. Quasi-static indentation test As stated in Section 3.2, the sensor must detect a load larger than that that can cause damage to structures (>1000 N) but should not detect lower loads such as the aerodynamic forces that act on aircraft (
Fig. 12. Actual and estimated positions of multiple-impact loading (doublebarreled impacts). Squares, triangles, and crosses show the actual (xactual, yactual), first-step-estimated (x1, y1), and second-step-estimated (x2, y2) coordinates of the impacts, respectively.

decreased to 11 N and there remained a large plastic deformation in both the copper sheet and silicone rubber sheet of the sensor. However, the sensor would be able to distinguish a load that can cause damage from a small aerodynamic load during flight. If the sensor had received additional indention loads after the fifth loading, it may have provided a false detection from the aerodynamic pressure on the aircraft surface. However, it is unlikely that a load occurs multiple times at the exact same position in a large area of an airplane component. In addition, if the sensor is forcefully compressed by a force stronger than 5000 N and the threshold further decreases, the surface sheet can be partially replaced easily. Additionally, note that the detection sensitivity for the load is adjustable according to the magnitude of an impact to be detected by changing the height of the copper protrusion of layer 1 and the thickness of the silicone rubber sheet (i.e., layer 2) shown in Fig. 5. The higher protrusion and the thinner rubber sheet would make the threshold smaller, which means that a smaller impact load causes layers 1 and 3 to make contact with each other. 4.2. Single-impact loading Fig. 10(a) presents a typical result of the single-impact tests. The left and right vertical axes show the electrical potential of layer 1 and value of impact loading calculated from the strain gage response, respectively. The potential started to rise within 0.5 ms after the strain gage response. The potential then varied. As shown in the enlarged view (Fig. 10(b)), layer 1 had four values of potential, Vx1, Vx2, Vy1, and Vy2, in Eqs. ((1)–(4)) in each period of 0.114 ms. Positions of impacts applied to points 1–3 were calculated from Vx1, Vx2, Vy1, and Vy2 using Eqs. ((1)–(4)) as shown in Fig. 11. The

Table 2 Single-impact test results. Comparison between the actual loading position (xactual, yactual) and the estimated position (x1, y1, x2, y2). Loading point

Peak value of the impact (N)

Actual (mm)

First step of the proposed method (mm)

Second step of the proposed method (mm)

xactual

yactual

x1

y1

Distance between actual and estimated positions

x2

y2

Distance between actual and estimated positions

Point 1 Point 2 Point 3 Average

2.29 2.23 2.97

400 500 750

950 500 250

288 465 625

756 455 182

81.8 70.9 75.5 76.1

418 489 757

956 484 240

18.6 18.9 11.7 16.4

52

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

Table 3 Multiple-impact test results. Comparison between the actual loading position (xactual, yactual) and the loading position estimated in the second step of the method (x2, y2). Number of experiments

1st time 2nd time 3rd time

Loading point

Point 10 Point 30 Point 20 Point 40 Point 20 Point 40 Average

Time interval between two impacts (ms)

} 11.5 } 9.40 } 1.68

Peak value of the impact (N)

4.66 4.81 3.85 4.46 1.44 3.55

square, triangle, and cross encircled with a dashed frame show the actual (xactual, yactual), first-step-estimated (x1, y1), and secondstep-estimated (x2, y2) coordinates of each impact load, respectively. The sensor was impacted at several points, but not simultaneously. Table 2 presents all the results of the difference (error) between the actual and estimated positions. The average errors of the first step and second step were 76.1 mm and 16.4 mm, respectively. This error of the second step is regarded as small enough for practical use. The objective of the sensor is to specify the loaded areas of potential damage. Another method is subsequently applied for detailed inspection of damage. Importantly, the detailed inspection needs to be conducted only over a circular area with a radius of a few centimeters centering at the position estimated in the second

Actual load position (mm)

Second-step-estimated load position (mm)

xactual

yactual

x2

20 500 400 750 400 750

700 500 950 250 950 250

7.70 777 426 438 314 974 806 205 Cannot be estimated

y2

Distance between actual and estimated positions 78.1 96.8 89.1 72.0

84.0

step, x2, y2. The reproducibility of the estimation of the impact load positions was regarded as sufficiently high. One of possible causes of the estimation error for load position is as follows. Fig. 6 shows that the LPS sensor is a sandwiched structure consisting of the top copper layer, the second silicone rubber insulator (i.e., dielectric layer), and the bottom copper wires, and therefore the sensor may also behave as a capacitor. When an impact load compresses the sensor sheet and reduces the thickness of the rubber layer, the capacitance of the LPS sensor increases for an instant. This would change the voltage of layer 1. Therefore, the voltage change caused by the capacitance increase may have affected the potential measured on layer 1, which would cause an error of the position estimation for the impact. However,

Fig. 13. Result of the double-barreled impact loading test that resulted in a failure: (a) shows the measured relation between the impact loads and the electrical potential of the copper and (b) shows the enlarged view of (a). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54 Table 4 Linear temperature expansion coefficients of copper and carbon fiber composite materials. Material

Linear temperature expansion coefficient (106/°C)

Copper Copper Beryllium 25 Std. CFRP UD (fiber direction) [24] Std. CFRP UD (orthogonal direction) [24] Std. CFRP Fabric (planar isotropic) [24]

+16.6 +17.8 –0.3 +25 +2.1

this error was probably not so large in this experiment because the capacitance change is proportional to the area compressed by the impactor. The hemispherical impactor of 15.9 mm diameter was too small to generate a large change in the capacitance of the LPS sensor sheet of 1000 mm  1000 mm in size. However, note that there is still a possibility of impact load monitoring using the capacitance change in the future. 4.3. Multiple-impact loading Fig. 12 presents the estimation results for the multiple-impact loading test. Squares, triangles, and crosses show the actual, firststep-estimated, and second-step-estimated coordinates of the impact loads, respectively. Double-barreled impacts were applied to the sensor three times: first for points 10 and 30 , second and third for points 20 and 40 . Table 3 presents the distance between the actual and estimated load positions for various time intervals between the two collisions. The average error of the second-stepestimated load positions was 84.0 mm when the time difference between the two impacts was 9.4 ms or larger. As the proposed sensor was set to finish the entire process of the position estimation for each impact in 0.5 ms, it was supposed to have identified the multiple impacts separately when the time difference was larger than 0.5 ms. However, the sensor failed to distinguish two impacts hitting with a time difference of 1.68 ms (see the third multiple-impact loading results for points 20 and 40 in Table 3). The cause of this is as follows. Time of contact between the impactor and the sensor was approximately 2.0 ms under the experimental conditions. When the time difference is shorter than 2.0 ms, the sensor is compressed at multiple points exactly simultaneously as shown in Figs. 13(a) and (b). Therefore, the sensor was able to estimate the position of the first impact but the second impact was impossible to identify. In general, time of contact between the impactor and scarcely dampened materials such as FRPs is shorter than that between the impactor and low-stiffness materials. As the time of contact with each impactor is shorter, multiple impacts are easier to detect separately. However, if the contact time is too short, the process of the position estimation for each impact cannot be finished. To accurately estimate the impact load position the sensor needs to be impacted for at least 0.5 ms under conditions of no other loading under the experimental conditions. However, the time required for the position estimation can be shortened by switching the electrical transistors described in Fig. 5 more quickly. Therefore, the sensing time is adjustable in accordance with the kind and structural constitution of a target material. 5. Conclusions and future work This study presents a new switching system, which is installed on a LPS sensor sheet to measure the position of an impact load. The sensors can be used to monitor the structural indentations of composite airplane structures. A thin copper sheet acts both as lightning protection and wires up the sensor, leading to overall reductions in weight and volume of the entire sensor system. The

53

feasibility of the real-time estimation of the load position was investigated in static indentation and dynamic impact tests. The repeated static indentation test at 2000 N revealed that the threshold value for detecting a load applied to the sensor continued to decrease as loading was repeated but seemed to converge to about 100 N. This means that the sensor has good load detection repeatability. Although repeated indentation of 5000 N generated large plastic deformations in the silicone rubber sheet of the sensor and reduced the detection threshold greatly, the sensor was able to distinguish a large load that can cause damage to the structure from a small aerodynamic load during flight even after the fifth loading. Additionally, the sensor sheet can be partially replaced easily, if broken. The result of the single-impact loading test indicated that the average difference between the actual loading position and the loading position estimated in the second step of the method (x2, y2) was 16.4 mm. This error is small enough for practical use. In the multiple-impact loading test, the sensor identified two impacts individually when the time difference between the impacts was 9.4 ms or longer. However, multiple impacts were impossible to distinguish when the time difference was shorter than time of contact between the impactor and the sensor (approximately 2.0 ms in this experiment). If the collision time interval is longer than 9.4 ms, the proposed sensor can estimate the respective positions of double-barreled impacts with an average error of 84.0 mm. This error means that detailed inspection needs to be conducted only over a circular area with a radius of 100 mm centering on the estimated position. However, note that this manuscript just provided an idea of utilizing an LPS as a real-time load monitoring sensor but did not necessarily verify the practical feasibility of the LPS sensor for an actual application because of the following reasons.  Inplane aerodynamic loads and thermal stress that occur in aircraft structures during flight or on the ground were not taken into account. Airplane structures receive various kinds of inplane loads that are not related to impact loads. The upper skin of the aircraft structure is typically loaded in inplane compression during flight and inplane tension on the ground. Furthermore, temperature difference between the ground and the sky can cause a large thermal stress because of mismatch in the thermal expansion coefficients between the metal LPS skin and the FRP structure (Table 4). When the FRP is not a unidirectional laminate but consists of multi-angle plies, interlaminar thermal stress will occur in the FRP laminate due to the large difference in the thermal deformation tendency between the plies. These inplane loads can change the threshold value for detecting out-of-plane impact loads applied to the sensor and the accuracy of the position estimation for the impacts. However, all the impact tests in this manuscript were conducted under no inplane loading. It is necessary to conduct the verification tests under the inplane loads for actual practical use of the LPS sensor.  Nonconductive layer attached on the top surface of the aircraft skin is not taken into consideration. In some cases, a thin nonconductive layer (e.g., glass fiber composite layer) is attached on the top surface of the aircraft skin ([10]. If the top layer is considerably thick, the accuracy of the estimation for load positions will be reduced because the top layer disperses a collision load and increases the contact are between the copper layer (layer 1) and the copper wires in layer 3.  Impact tests without composite structures under the LPS sensor sheet were insufficient for verification of the practical feasibility of the sensor.

54

Y. Suzuki et al. / Composites: Part A 67 (2014) 44–54

In reality the mechanical response of the LPS sensor sheet due to an external impact is affected by the existing substrate (i.e., aircraft composite structure). However, all of the impact tests in this study were conducted for only the LPS sheet. It is future work to execute the impact test following an actual environment for the LPS sensor on a fiber composite structure (not the LPS sensor alone). References [1] Kondo H, Aoki Y, Hiraoka K, Hatta H. Residual indentation, delamination area and CAI strength of CFRP laminates under low-velocity impact. In: Proc. of 16th international conference on compos mater; 2007. p. 1–7. [2] Rivallant S, Bouvet C, Hongkarnjanakul N. Failure analysis of CFRP laminates subjected to compression after impact: FE simulation using discrete interface elements. Compos A 2013;55:83–93. [3] Gan KW, Stephen R, Wisnom MR. Measurement and modelling of interlaminar shear strength enhancement under moderate through-thickness compression. Compos A 2013;49:18–25. [4] Scarponi C. Ultrasonic technique for the evaluation of delaminations on CFRP, GFRP, KFRP composite materials. Compos B Eng 2000;31:237–43. [5] Bates D, Smith G, Lu D, Hewitt J. Rapid thermal non-destructive testing of airplane components. Compos B 2000;31:175–85. [6] Vijayaraghavan GK, Majumder MC, Ramachandran KP. Quantitative analysis of delaminations in GRP pipes using thermal NDTE technique. J Adv Res Mech Eng 2010;1:60–8. [7] Gagné M, Therriault D. Lightning strike protection of composites. Prog Aerosp Sci 2014;64:1–16. [8] Hirano Y, Katsumata S, Iwahori Y, Todoroki A. Artificial lightning testing on graphite/epoxy composite laminate. Compos A 2010;41(10):1461–70. [9] Hirano Y, Iwahori Y, Katsumata S, Todoroki A. Relationship of lightning damage behavior and electrical properties of CNF dispersed CFRP. In: Proceedings of Japan international SAMPE symposium & exhibition; 2009. SMS-6-4. [10] Dexmet Corporation. Lightning Strike Protection for Carbon Fiber Airplane. In: Advancement of Materials Process Engineering (SAMPE) Conference; 2007 .

[11] Kawakami H, Feraboli P. Lightning strike damage resistance and tolerance of scarf-repaired mesh-protected carbon fiber composites. Compos A 2011;42:1247–62. [12] Feraboli P, Miller M. Damage resistance and tolerance of carbon/epoxy composite coupons subjected to simulated lightning strike. Compos A 2009;40:954–67. [13] Suzuki Y, Suzuki T, Todoroki A, Mizutani Y. Smart lighting protection skin using QTC pills for aircraft realtime load monitoring. J Solid Mech Mater Eng 2014;1(4):1–11. [14] Suzuki T, Suzuki Y, Todoroki A, Mizutani Y. Use of a lightning protection shield for composite airplanes as a load sensor. In: Proc. of 9th joint Canada–Japan workshop on composites; 2012. p. 177–84. [15] Cok RS, Bourdelais RR, Kaminsky CJ. Flexible resistive touch screen. US Patent 0212599 A1 2004. [16] Downs R. Using resistive touch screens for human/machine interface. Analog Applic J 2005;3Q:5–9. [17] Stetson J. Analog resistive touch panels and sunlight readability. J Soc Info Display 2006;22:26–30. [18] Hecht DS, Thomas D, Hu L, Ladous C, Lam T, Park Y, et al. Carbon-nanotube film on plastic as transparent electrode for resistive touch screens. J Soc Info Display 2009;17(11):941–6. [19] Cross E, Moshrefzadeh R, Geaghan B. Resistive touch screen incorporating conductive polymer. US Patent 20050076824 A1 2005. [20] Suzuki Y, Todoroki A, Mizutani Y, Matsuzaki R. Impact damage detection in CFRP using statistical analysis of resistance–temperature characteristics. J Solid Mech Mater Eng 2011;5:33–43. [21] Suzuki Y, Todoroki A, Mizutani Y, Matsuzaki R. Damage detection in CFRP without data from intact structures using statistical analysis of resistance– temperature characteristics. Jpn Soc Mech Eng Series A 2011;77:181–9. [22] Takahashi K, Park JS, Hahn HT. An addressable conducting network for autonomic structural health management of composite structures. Smart Mater Struct 2010;19:1–10. [23] Watkins SE, Akhavan F, Dua R, Chandrashekhara K, Wunsch DC. Impactinduced damage characterization of composite plates using neural networks. Smart Mater Struct 2007;16:515–24. [24] ACP Composites Inc., California, US. Product Specification Reference Sheet. 1800-811-2009.