Vous êtes sur la page 1sur 6

ICROS-SICE International Joint Conference 2009 August 18-21, 2009, Fukuoka International Congress Center, Japan

High Accurate Detection of Inclined Cracks in Concrete Structures Using an Acceleration Pickup
Shozou Miyoshi1 , Masayuki Okamoto1 and Shogo Tanaka1
1

Graduate School of Science and Engineering, Yamaguchi University. 2-16-1, Tokiwadai, Ube, 755-8611, Japan (Tel: +81-836-859-428; E-mail: s.tanaka@yamaguchi-u.ac.jp)

Abstract: The present paper proposes a method for the detection of inclined cracks in concrete using an acceleration pickup. The sensor on the concrete surface observes the stationary waves inside the structure whose frequencies depend on the distance to the crack. The method can recognize the prole of the inclined crack with a high accuracy by the following two procedures. The 1st procedure : we model the sensor output as an output of a linear dynamic system with an unknown parameter and measure the accurate propagation distance from the sensor to the crack by applying a maximum likelihood method at each inspection point. The 2nd procedure : we apply a least squares method considering the reection efciency of the elastic waves at the inclined crack and the directivity characteristic of the sensor. Keywords: non-destructive inspection, inclined crack, acceleration pickup, stationary waves, linear dynamic model, maximum likelihood method

1. INTRODUCTION
In recent years, accidents such as aking-off of concrete have frequently happened in a variety of concrete structures such as bridges, buildings and tunnels. As a diagnostic method, the impact acoustic technique has often been utilized[1][2][3]. However, this method has a low reliability and depends on the expertise of the operator. From the viewpoint, we previously proposed a method for inspecting cracks in concrete structures using an acceleration pickup[4][5][6]. This method assumed that the crack was parallel to the concrete surface. However, an actual crack is not necessarily parallel to the surface. As an example, while concrete is added in multiple batches during the manufacturing of tunnels, a horizontal boundary (called the cold joint) is generated. This boundary has an inclination when looked at from the concrete surface. It is also well-known that the cold joints grow into cracks with time. It is thus important to detect the inclined crack and recognize its prole with a high accuracy. However, when the crack has an inclination with respect to the surface, the propagation distance from the sensor to the reection point on the crack is not necessarily the distance to the point on the crack just below the sensor. Thus, the crack is not necessarily recognized in a right manner with our previous method. This paper proposes an improved inspection method for inclined cracks in concrete. The prole of the inclined crack can be recognized with a high accuracy by the following two procedures. We rst model the sensor output as an output of a linear dynamic system with an unknown parameter and measure the propagation distance from the sensor using the previously developed method. We next apply a least squares method to the propagation distances considering both the reection efciency of the elastic wave at the inclined crack and the directivity characteristic of the sensor.

Of course, the non-destructive inspection of cracks in concrete structure can be done with an ultrasonic sensor. However, the ultrasonic sensor is expensive and its search depth is small (about 0.4m) because of a large attenuation of the ultrasonic wave. The acceleration pickup adopted with this paper costs much less than the ultrasonic sensor and its search depth is high (about 10m) due to the use of the elastic wave.

2. EXPRESSION OF THE SENSOR OUTPUT AS AN OUTPUT OF A LINEAR DYNAMIC SYSTEM


When the surface of the inspected structure is hit with a hammer, a stationary wave is generated depending on the distance from the surface to a reection point on the crack. The angular frequency of the stationary wave of j th mode is given by where q and L are respectively the velocity of the elastic wave in the concrete structure and the propagation distance. In a short time interval, the stationary wave (elastic wave) is modeled as the sinusoidal wave with angular frequency j [4][5][6], xj (t) = j sin(j t + j ) where j and j are respectively the amplitude and the initial phase of the j th mode. When we take n modes of the stationary wave, then the sensor output can be modeled as z (t) =
n j =1

j = j 2q/(2L)

(1)

(j = 1, 2, )

(2)

xj (t)

(3)

If the state vector is dened as x(t) = (x1 , x 1 , , xn , x n )T , then the following state equation concerning the elastic wave is obtained[4][5][6]. (t) = Ax(t) + w(t) x (4)

- 2321 -

PR0002/09/0000-2321 400 2009 SICE

0 z2

concrete surface

acceleration pickup

z1 Lmax

is performed using the following Kalman lter by assigning adequate propagation distance L at each inspection point[7][8]. k/k1 x k/k x k Pk/k1 Pk/k where k1/k1 = Fx k/k1 + Kk k = x k/k1 = yk H x = = (10) (11) (12) (13) (14)

Lmin
an inclined crack

u=az1+bz2 +c u
region of possible reflection points S

Fig. 1 Region of possible reection points

F Pk1/k1 F T + W Pk/k1 Kk HPk/k1 k = 0, 1, 2, . . .

where A = diag{A1 , A2 , , An } and Aj = 0 2 j 1 0 (0 j n) (5)

Kk k

= =

1 Pk/k1 H T k T

(15) (16)

HPk/k1 H + V

(6)

Furthermore, w(t) = (0, w1 , 0, w2 , , 0, wn )T represents the transition noise vector and wj (t) (1 j n) are assumed to be independent white Gaussian noises with zero means and a common variance 2 . These noises are introduced to compensate for a possible slight attenuation of the stationary waves in the window where the measurement is to be achieved. Discretizing the dynamic equation (4), the following equation is obtained[4][5][6][7]. xk+1 = F xk + wk (7) where xk and wk are respectively the state vector and the transition noise vector at the k th sampling time k T (T : sampling period), and F is the transition matrix dened as where s, I , L1 are respectively the Laplace operator, the unit matrix and the inverse Laplace transform. The noise vector wk has zero mean. See the reference [4] for the covariance matrix of the noise vector wk . On the other hand, by considering the sum of all the unused modes of the stationary wave and the intrinsic sensor noise as the observation noise, the observation equation can be represented as follows. yk = H xk + vk where H is the 1 2n dimensional observation matrix dened as H = [ 1, 0, 1, 0, , 1, 0 ] and yk and vk are respectively the observation and the observation noise at the k th sampling time. (k = 0, 1, ) (9) F = (T ), (t ) = L1 [(sI A)1 ] (8)

and V represents the variance of the observation noise vk . Since L is unknown, the Kalman lter cannot be directly applied. The reliability of the parameter L is evaluated by the following likelihood function[4][5][6][7] J (L) = =
K K

p(yk /L, Y k1 ) 1

k=1

k=1

(2 k (L)) k/k1 (L))2 (yk H x exp 2k (L)

(17)

where p(yk /L, Y k1 ) is a conditional probability density function of yk for a given observation sequence Y k1 k/k1 (L) and k (L) are and the assigned value of L. x respectively the estimated state vector and the covariance matrix at the k th sampling time.

4. EVALUATION FUNCTION FOR IDENTIFYING THE CRACK


The elastic wave observed with the sensor is generated between the concrete surface near the sensor point (i.e., hitting point) and a reection point on the crack. This is because the surface near the sensor is hit with a hammer. The reection point on the crack seems to be located between a point just below the sensor and the foot of the perpendicular drawn from the sensor to the crack. The reason for this is as follows : at the foot of the perpendicular drawn from the sensor to the crack, the reection efciency of the elastic wave is high despite the energy of the elastic wave generated by the impact is low. On the other hand, in a direction just below the sensor, the reection efciency is low despite the energy of the elastic wave is high. Considering the distribution of the energy of the elastic wave, the directivity of the sensor and the reection efciency at a point on the crack, it is considered that the distribution of the reection point is as shown in Fig. 1. In this gure, Lmax is the distance from the sensor to the point on the crack just below the sensor and Lmin is the

3. MEASUREMENT OF THE PROPAGATION DISTANCE FROM THE SENSOR


Since a white Gaussian property of the observation noise can be assumed[4][5], estimation of the state vector

- 2322 -

200

z1 z1 =430 z1 =630

100 output [mV]

250

800 1000 500

z2 =500 z2 =550
750 measurement points

100 0 2 4 time [ms]

z2
0

1000

Fig. 3 Sensor output at the inspection point (z1 , z2 ) = (550, 550) for the test specimen No.1

(a) top view acceleration pickup crack (1mm thick)

z1
likelihood J (f ) 15700 7,100Hz 15720

10,200Hz 10,800Hz

462

138 1000
15740

(b) sectional view

(unit : mm )

10000

20000

30000 frequency f [Hz]

Fig. 2 Test specimen No.1

distance from the sensor to the foot of the perpendicular drawn from the sensor to the crack. It is thus considered that the propagation distance L of the elastic wave measured by the method satises Lmin L Lmax at each inspection point. To identify the crack, the inclined crack is assumed to be a plane and the shape of the crack can be modeled as (18) u = az1 + bz2 + c The distance from the sensor to the foot of the perpendicular drawn from the sensor to the crack is expressed as Lmin = |az1 + bz2 + c|/ 1 + a2 + b2 (19)
N i=1

Fig. 4 Single-mode likelihood function J (f ) (in logarithmic scale) for the test specimen No.1 at the inspection point (z1 , z2 ) = (550, 550) (from the sensor to a reection point on the crack) for , z2 ). The reason why N is the each inspection point (z1 number of propagation distances but not the number of inspection points is that we can have several propagation distances for each inspection point.

5. EXPERIMENTS
5.1 Test specimen No.1 We apply the method to a concrete test specimen which has the prole of the crack as u = 0.575z1 115 (z1 200, 250 z2 750) as shown in Fig. 2. An acceleration pickup (TOYO Corp., PCB352C65SN) is set on the concrete structure. By hitting the surface near the sensor with a hammer and changing the point of im , z2 ) = (430, 500), (470, 550), pact as (refer to Fig.2), (z1 (510, 500),(550, 550),(590, 500),(630, 550), we obtain a series of data. Each point is hit ve times. We have thus ve sets of data for each inspection point. The sensor output is taken with a sampling period T = 1s. For reference, a sensor output at the inspection point , z2 ) = (550, 550) is shown in Fig. 3. The single(z1 mode likelihood function is obtained by applying our previous method to the sensor output and is shown in Fig. 4. Although a value of 1, 2 or 3 is acceptable for n (the number of modes), we assumed here n = 1. This is because an elastic wave of a higher mode has larger attenuations and small n decreases the complexity of the system.

The three parameters a, b, c are now determined by minimizing the following evaluation function. I (a, b, c) = [ei (a, b, c)]2 (20)

where ei (a, b, c) = min |Li w|, w [Lmin , Lmax ]


w

(21)

Expressing more concisely Lmin Li (Li < Lmin ) (Lmin Li Lmax ) ei (a, b, c) = 0 Li Lmax (Li > Lmax )

(22)

where N is the number of propagation distances and Li is the measured propagation distance of the elastic wave

- 2323 -

concrete surface measured by conventional least squares method

depth u[mm]

200

real crack measured by the proposed method

400

400

500 600 700 horizontal distance z1[mm]

Fig. 6 Measurement result for the test specimen No.1 The single-mode likelihood function is obtained by applying our previous method to ve sets of data for each inspection point (30 data in total). The propagation distances from the reection points are obtained by using L = q/(2f ) from the peak frequencies (f ) of the likelihood function. For reference, the propagation distances are drawn vertically at each inspection point (see Fig. 5). Fig. 5(a) and Fig. 5(b) respectively show the case of the scanning line for z2 = 500 and for z2 = 550. The line in Fig. 5 and the numerical value are respectively the range of the reection points and the number of reection points included within the range. The propagation distances are seen to be divided into three groups A, B and C as shown in Fig. 5. Group A corresponds to the inclined crack and there are overwhelmingly a lot of reection points in each inspection point (61 points in total, 30 points for the scanning line for z2 = 500 and 31 points for the scanning line for z2 = 550). The propagation distances belonging to the groups B and C are few, i.e., 11 and 6 points respectively (respectively 14 and 8 of the total number of the reection points). Since group As likelihood is much larger than groups B and C, it is thus easy to discriminate the group A from groups B and C. By the way, if we observe the groups B and C, it is seen that in addition to the number of propagation distances contained being few and the likelihood being very small, the depths of these groups are also mutually related by a factor of two. Therefore, considering the relationship in their frequencies, if one corresponds to the rst mode frequency then the other corresponds to the second mode frequency of the same stationary wave. The frequency 3, 570Hz that corresponds to the depth 560mm can be considered to be the rst mode frequency. In this case, the reection points of group C correspond to the bottom of the structure. The frequency 7, 140Hz that corresponds to the depth 280mm of group B becomes the second mode stationary wave frequency of group C. Thus group B can be virtually disregarded. Therefore, group A can be easily isolated by considering the magnitude of the likelihood and the integer multiplicity of the frequencies of the two other groups. By minimizing the equation (20) using the propagation distances of group A for all the inspection points, the parameters a, b and c for each impact are obtained and shown

Fig. 5 Depths of the reection points for the test specimen No.1 Table 1 Estimates for the parameters a, b and c for the test specimen No.1 a 0.595 0.570 0.568 0.589 0.581 b 0.02 0.00 0.00 -0.05 0.00 c -122 -115 -120 -102 -125

case 1 case 2 case 3 case 4 case 5

Note that, the horizontal axis of Fig. 4 uses the frequency f (= q/(2L)) instead of the distance L. In this gure, three main peaks are observed at 7, 100HzC10, 200Hz and 10, 800Hz respectively. If we cosider these to be the rst mode frequencies, then the corresponding calculated propagation distances L (= q/(2f )) become 282mm, 196mm and 185mm respectively, where the velocity of the elastic wave in the concrete adopted here is the standard one of 4, 000 m/s. The frequencies 10, 200Hz and 10, 800Hz correspond to the elastic wave from the inclined crack and the frequency 7, 100Hz corresponds to the elastic wave from the concrete bottom. However, in most of the data, the elastic wave from the concrete bottom cannot be observed and the main peaks are solely related to the elastic waves from the crack. Moreover, since 24 main peaks are observed, therefore, it means that multiple reection points exist on the crack.

- 2324 -

z1 z1 =400 z1 =700

250

z2 =500 z2 =600
750 measurement points

1000 500

z2
0 104

1000

(a) top view acceleration pickup crack (1mm thick)


308

z1
292

1000

(b) sectional view

(unit : mm)

Fig. 7 Test specimen No.2 Fig. 8 Depths of the reection points for the test specimen No.2 in Table 1. Since an actual inclined crack has a = 0.575, b = 0.0 and c = 115.0, the parameters for each hitting are seen to be obtained in high accuracy and there is reproducibility. When these measurements are averaged, they become a = 0.581, b = 0.006 and c = 116.8. For reference, the prole of the inclined crack measured by the proposed method and the one measured by a conventional least squares method are shown in Fig. 6. It is seen from Fig. 6 that the inclined crack can be identied accurately with the proposed method. Here, because the inspection point is changed within the range of (z1 200, 250 z2 750), the inclined crack within this range is obtained. However, if the range of the inspection point is changed, the inclined crack within the changed range can be measured. 5.2 Test specimen No.2 Next, we apply the method to a concrete test specimen which has the prole of the crack as u = 0.188z1 + 104 (z1 0, 250 z2 750) as shown in Fig. 7. The same acceleration pickup that was used in section 5.1 is again used here. By hitting the surface near the sensor with a hammer and changing the point of impact as (refer to Fig.7), (z1 , z2 ) = (400, 600), (460, 500), (520, 600), (580, 500), (640, 600), (700, 500), we obtain a series of data. Each point is hit ve times. We have thus ve sets of data for each inspection point. Although the details are omitted, the single-mode likeTable 2 Estimates for the parameters a, b and c for the test specimen No.2 a 0.185 0.192 0.205 0.200 0.190 b 0.00 0.01 -0.03 -0.03 0.01 c 108 100 113 117 101

case 1 case 2 case 3 case 4 case 5

lihood function is obtained by applying our previous method to 30 data in total. The propagation distances from the reection points are obtained by using L = q/(2f ) from the peak frequencies (f ) of the likelihood function. For reference, the propagation distances are drawn vertically at each inspection point (see Fig. 8). Fig. 8(a) and Fig. 8(b) respectively show the case of the scanning line for z2 = 500 and for z2 = 600. Here, if the ones with the large likelihood are made group A and other ones are made groups B and C, then the propagation distances are seen to be divided into three groups as shown in Fig. 8. Of course, group A corresponds to the inclined crack and groups B and C correspond to the propagation distances due to the reection from the bottom of the structure. By minimizing the equation (20) using the propagation distances of group A for all the inspection points,

- 2325 -

0 real crack depth u[mm]

concrete surface measured by the proposed method

200 measured by conventional least squares method 400 400 500 600 700 800 horizontal distance z1[mm]

Here, we proposed the accurate prole measurement of an inclined crack by using the acceleration pickup because it costs much less than the ultrasonic sensor and search depth is higher. This system can be similarly applied also using the low-cost piezoelectric sensor.

REFERENCES
[1] T.Uomoto, K.Kato and S.Kono, Non-destructive Inspection of Concrete Structures, Morikita Publishing Co.,Ltd. (1990) [2] H.Oshima and O.Yoshioka, Example of Applying Nondestructive Inspection of Concrete Structures Thickness and Cavity Survey of Concrete in Railway Tunnel, Concrete Institute, 127-3 (1989) [3] JTA maintenance committee, New Method for Inspection of Tunnels, Tunnel and Underground, Vol. 27, No. 811, (1996) [4] S.Tanaka and M.Okamoto, Dynamic Model Based Pipe Length Measurement Using Stationary Wave, Measurement, Vol.28, 249/259 (2000) [5] M.Arita, M.Okamoto and S.Tanaka, Detection of Cracks for Concrete Structures Using an Acceleration Pickup, The 14th Transactions of the Society of Instrument and Control Engineers, Chugoku Branch, 214/215 (2005) [6] S.Tanaka and M.Okamoto, High-precision Detection of Multiple Minute Cracks in Reinforced Concrete Pile Considering the Orthogonality in Stationary Waves, Transactions of the Society of Instrument and Control Engineers, Vol. 43, No.9, 725/734 (2007) [7] S.Tanaka, System Measurements Engineering, Asakura Pub. Co. (1994) [8] S.Arimoto, Kalman Filter, Sangyo-Tosho (1977) [9] M.Okamoto, H.Fijimoto and S.Tanaka, Health Monitoring of Concrete Pillars Using an Acceleration Pickup, Transactions of the Institute of Electrical Engineers of Japan.C, Vol.126, No.8, 931/937 (2006)

Fig. 9 Measurement result for the test specimen No.2 the parameters a, b and c for each impact are obtained and shown in Table 2. Since an actual inclined crack has a = 0.188, b = 0.0 and c = 103.5, the parameters for each hitting are seen to be obtained in high accuracy. When these measurements are averaged, they become a = 0.194, b = 0.008 and c = 107.5. For reference, the prole of the inclined crack measured by the proposed method and the one measured by a conventional least squares method are shown in Fig. 9. It is seen from Fig. 9 that the inclined crack can be identied accurately with the proposed method. By the way, we considered the case where the crack has a planar inclined surface in this paper. The actual crack might have a curved surface. This method can be applied in that case. For example, a curved surface can be piecewise approximated by several at inclined surfaces.

6. CONCLUSION
This paper proposed an improved inspection method for an inclined crack in concrete. The prole of the inclined crack could be recognized with a high accuracy by the following two procedures. We rst modeled the sensor output as an output of a linear dynamic system with an unknown parameter and measured the propagation distances using our previously developed method. We next applied a least squares method to the propagation distances considering both the reection efciency of the elastic wave at the inclined crack and the directivity characteristic of the sensor. The effectiveness was shown by the experiment.

- 2326 -

Vous aimerez peut-être aussi