Fast Focal Law Computing for Ndt Viena

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269162722

Fast focal law computing for Non-Destructive Testing with phased arrays

Conference Paper · January 2007


DOI: 10.3728/ICUltrasonics.2007.Vienna.1228_parrilla

CITATIONS READS
10 754

4 authors:

Montserrat Parrilla Alberto Ibañez


Spanish National Research Council Spanish National Research Council
68 PUBLICATIONS 1,062 CITATIONS 63 PUBLICATIONS 597 CITATIONS

SEE PROFILE SEE PROFILE

Jorge Camacho Carlos Fritsch


Tecnológico de Monterrey Spanish National Research Council
77 PUBLICATIONS 1,214 CITATIONS 89 PUBLICATIONS 1,507 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Alberto Ibañez on 19 December 2014.

The user has requested enhancement of the downloaded file.


Paper #1228 Presented at the International Congress on Ultrasonics, Vienna, April 9 - 13, 2007, Session R31: Underwater acoustics

Fast focal law computing for Non-Destructive Testing with phased


arrays
M. Parrilla, A. Ibáñez, J. Camacho, C. Fritsch
Instituto de Automática Industrial, CSIC, La Poveda (Arganda), 28500 Madrid, Spain
montse@iai.csic.es, aibanez@iai.csic.es, jorgecam@iai.csic.es, carlos@iai.csic.es

Abstract. Three focal-law-computing methods for Another simulation method has been used to model
dynamic focusing thorough interfaces of arbitrary shape propagation problems through interfaces in NDT [3].
are analysed. One, based on the Fermat’s principle, is Although the technique is fast, for dynamic focusing the
accurate but very time consuming. The second, derived computing time can be long.
from the Abbe’s invariant, is fast but only accurate under The methods based on the spatial impulse response
some very restrictive conditions. are widely used [4]. If changes of media are involved, the
The Fast Focal Law Computing (FFLC) method is method can still be applied [5], but the successive
proposed. It provides good accuracy and low computing convolutions at the interfaces greatly increase the
time. In most cases, iterative processing is avoided and computing time. Ray-tracing methods [6] predict sound
the delay law timing errors are kept well below the time trajectories and compute the focal laws. Nevertheless, in
resolution of the electronics. presence of complex interfaces, the method may become
Keywords. Focal law computing, dynamic focusing, very time consuming [7].
ray tracing, NDT, focusing thorough interfaces. While field simulation is useful for interpretation, in
many applications the operator needs a tool to compute
the focal laws to perform the inspection. In these cases,
A. Introduction. the shorter the time spent in focal law computing and
parameter adjustments, the better. This paper presents a
The application of the ultrasonic phased array Fast Focal Law Computing method, which is fast,
technique to Non Destructive Testing (NDT) requires accurate and applicable to interfaces with arbitrary shape.
computing the focal laws. These define the time delays
for emission and reception, for every array element and
B. Time delay resolution.
for every focus. State-of-the art equipment allows
dynamic focusing in reception, which improves Limited delay resolution produce delay quantization
resolution and sensitivity [1]. lobes that increase the noise floor and reduce the image
contrast. If the ratio of the sampling rate to the signal
A dynamically focused image composed of L lines, F
frequency is µ, the rms value of the delay quantization
foci per line and N array-elements, require Q=L·F·N
lobes to the main lobe amplitude is [8]:
delays. Typically, L=100, F=500, N=64, and Q is over 3
million delays. With static focusing Q would be about π
6,000. Thus, the application of dynamic focusing requires ηQ ≈ µ >> 1 (3)
µ 6N
fast focal law computing methods.
Focal laws for contact inspections of homogeneous On the other hand, for a point reflector, the one-way
media with sound velocity c, are easyly computed. The response is ηA =1/N at out-of focus places. The level of
delays τAF for every array element at (xA, zA) and focus at the delay quantization lobes should remain below this
(xF, zF) are obtained from the time of flight TF as: figure. Choosing a value of ηQ =1/2N,

1 N
TF = ( xF − x A ) 2 + ( z F − z A ) 2 (1) µ > 2π (4)
c 6
τ AF = TK − TF (2) This way, if TR is the period of the received signal,
the delay errors should be limited to:
where TK is a constant to make all the delays positive.
But, in many cases, there are interfaces that complicate TR 6
TE < (5)
the focal-law-computing task due to refraction effects. 2π N
Field simulation tools may help in the design of the
inspection and in focal law computing. The most general The timing resolution of the electronics plus all the
approach solves the wave equation, but this process is not errors included in the computed focal laws should be
practical. A simplification is made in the Helmholtz accounted for. This way, to meet the criterion (5) for a
model by considering a continuous wave [2]. Solving the 64-element, 5 MHz array, total timing errors should be
resulting equation require a fine spatial discretisation, kept below 10 ns. For a similar 10 MHz transducer, the
which leads to a very large computing time. overall error has to be below 5 ns.

-1-
Paper #1228 Presented at the International Congress on Ultrasonics, Vienna, April 9 - 13, 2007, Session R31: Underwater acoustics

C. Focal law computing based on the Fermat’s The region that contains all the ray entry points is
defined as the interface of interest. It is bounded by points
principle. P and Q, which are the entry points of the rays from the
The well-known Fermat’s principle states that a ray extreme elements to the extreme foci (Fig. 2).
starting at a given point arrives to another point following Interpolation between P and Q provides a densely
the lowest time of flight path. Its application to the focal sampled interface. A cubic spline yields continuous
law computing problem in presence of a S-point sampled derivatives at all the points, required by some methods.
interface (xS, zS) separating two media with sound The principal rays are found by application of (6) to the
velocities c1 and c2, respectively, is shown schematically farthest focus with (xA, zA)=array centre. Foci should be
for a single focus in Fig. 1, with c2 > c1. located on these rays at the desired spatial intervals and
the angular resolution should be chosen to meet the
50
Nyquist criterion in the lateral direction.
45
F D. Focal law computing based on the Abbe’s
40
c2 invariant.
35
Fastest ray In optics, the Abbe’s invariant relates the position of
30
image and object as a function of the refraction indexes
mm 25 [9]. Strictly, it is only applicable to circumferences (2D
Entry point space) of radius R in the paraxial region (Fig. 3):
20

Interface
15
1 1 1  1 1 1 
 −  ≈  −  (7)
10 c1  R S1  c 2  R S 2 
5 Element c1
being S1 and S2 the distances from the lens to the object
0
-30 -20 -10 0 10 20 30
and to the image, respectively.
mm
F
Fig. 1. Application of the Fermat’s principle (c2 > c1).
C
The time-of-flight element-focus is (2D geometry): R
A B S2
1  O
 c ( x A − x S ) + (z A − z S ) + 
2 2 S1
TF = min  1  (6) c1 c2
 1 2
+
 c ( x F − x S ) 2
+ ( z F − z S ) 
 2  Fig. 3. Geometry for the Abbe’s invariant.
This procedure involves a long computing time: In the application to the focal-law computing
every focal law requires computing (6) S times for every problem, the object is an array element (A) and the image
focus and element. However, it yields the most accurate is a virtual array element (B). Under the paraxial
results for a given interface sampling resolution, so that it approximation, the virtual element at B behaves as if it
is used as the reference for comparison purposes. where in the second medium. Equation (7) can be used to
obtain the coordinates of B and, then, find the intersection
of the ray B-F with the circular interface at point C. The
50 focal laws for all the foci in the ray C-F are easily
45 computed from the time-of flight:
40 AC CF
TF = + (8)
35 c1 c2
30
The method is very fast, since all the parameters can
25 be obtained using closed formulae. In particular, if the
20 centre of the circumference is at (xO, zO):

S1 = − (x A − xO )2 + (z A − zO )2 − R 
15
P (9)
10  
Q
5
c1 S1 R
S2 = (10)
0
c 2 R − (c2 − c1 ) S1
-30 -20 -10 0 10 20 30

The coordinates (xB, zB) are found from the equations:


Fig. 2. Principal rays obtained from the array centre.

-2-
Paper #1228 Presented at the International Congress on Ultrasonics, Vienna, April 9 - 13, 2007, Session R31: Underwater acoustics

S2 = (x0 − x B )2 + (z 0 − z B )2 −R 1 ( X k − x A ) + M k (Z k − z A )
Vk = +
(11) c1 ( X − x )2 + (Z − z )2
z0 − z A
(x B − x A ) + z A
k A k A
zB = (15)
x0 − x A
+
1 ( X k − x F ) + M k (Z k − z F )
The entry point (xC, zC) is obtained from:
c2 ( X k − x F )2 + (Z k − z F )2
(xC − xO )2 + (zC − zO )2 = R 2 The problem is to find the index k for which Vk is
closest to zero. The Newton-Raphson method is used to
zF − zB (12)
zF = (x F − x B ) + z B obtain the integer index k that best verifies this condition:
xF − xB
 Vk 
where (xF, zF) are the coordinates of a focus in the ray CF. k i +1 = k i − round  
 (16)
Now, the time of flight from the array element (xA, zA) to  Vk +1 − Vk 
every focus (xF, zF) in that ray is given by: The process can be repeated until ki+1=ki (± 1 to
avoid oscillations). After n iterations, an index kn is
TF =
1
(xC − x A )2 + (zC − z A )2 + obtained. Substituting this value in (14), the coordinates
c1
(13) (xC, zC) of the entry point are determined, and the time-of-
+
1
(xC − x F )2 + (z C − z F )2 flight is found by (13).
c2 The process converges very rapidly. In most cases,
For other interface shapes, perpendicular lines are there is no need to perform any iteration at all: the index
k1 for a focus is obtained by direct application of (16)
drawn at P and Q. The crossing point defines the centre O
of an approximate circumference whose radius R is the using for k0 the k1 value of the precedent focus.
mean of R1=OP and R2=OQ. Figure 4 shows the
results for an element at A, virtual element at B and entry F. Evaluation of the different methods.
point at C, for all the foci in CF. The virtual array is also Parts with different shapes have been modelled and
shown. The most important drawback of the method is its focal laws for dynamic focusing have been derived
low accuracy when the required conditions are not met. following the described methods. All the processing has
been performed on the assumptions of no lossy media
and point-like array elements. Diffraction effects due to
F the finite size of the elements are not taken into account.
50

45
A wide-angle sectorial image is formed inside the
O modelled part. Transducer frequency, propagation
40
velocities, array geometry and other parameters have
35 been also varied among different simulations.
30
R Results shown in this section refer to the geometry
25
described before, for a 32-element, 5 MHz array, element
pitch 0.6 mm. The part is made of steel (c2 = 5970 m/s),
20 inspected in water immersion (c1 = 1470 m/s) and set at
15 C 10 mm from the array. A total of 40 steering directions
are generated for imaging, with an average of 425 foci
10
per line (a total of 17,000 focal laws or more than 0.5 M
5
B delays). Table I shows the MATLAB computing time
A with a Pentium 4 processor at 2.4 GHz and 1 GB RAM.
0
-30 -20 -10 0 10 20 30

Fig. 4. Obtention of the virtual elements B and entry C Table I. Computing time of the different techniques
for an array element at A and foci in the segment CF.
Method Time (s)
E. The Fast Focal Law Computing technique. Fermat 78
Consider the interface defined by: Abbe 2
x S = X (k ) = X k FFLC 6
z S = Z (k ) = Z k 1≤k≤U (14) The method based on the application of the Fermat’s
Z (k + 1) − Z (k ) principle is, by far, the more time-consuming. However,
mS = = Mk being the most accurate, the obtained values are used as a
X (k + 1) − X ( k ) reference for evaluation of the errors of other methods.
where U is the number of the interface points and Mk is By contrast, the method based on the application of
the interface slope for index k. Taking in (6) the the Abbe’s invariant is very fast; the proposed Fast Focal
derivative and substituting (14), yields: Law Computing method is 3 times slower, but more than

-3-
Paper #1228 Presented at the International Congress on Ultrasonics, Vienna, April 9 - 13, 2007, Session R31: Underwater acoustics

an order of magnitude faster than the technique based on The first method, based on the direct application of
the Fermat’s principle. the Fermat’s principle, provides the best accuracy but, by
far, requires the higher computing time. It is only advised
But, with regard to the errors, the method based on
the Abbe’s invariant is not applicable. Errors of several for computing a few focal laws as, for example, emission
focal laws and single focus applications, as well as the
hundreds of ns are found (Fig. 5), which invalidate the
requirements of (5). Perhaps, this method might be used starting point for other methods. Improvements can be
when operating with low steering angles, the distance of achieved by limiting the extent of the interface by means
the part to the array is relatively large or when the nearest of some of the described procedures.
foci are some distance apart from the interface. The method based on the application of the Abbe’s
invariant is the fastest one, but is only strictly applicable
550 to circular interfaces within the paraxial zone and foci far
500
from the interface. Procedures to adapt this method to
non-circular interfaces have been also discussed.
450
The Fast Focal Law Computing method has been
400 proposed and analysed. It is many times faster than the
Fermat’s principle method and has a similar accuracy.
350 Furthermore, the method can be applied with interfaces of
ns

300
any shape. Thus, it is the recommended method as a
general-purpose fast focal-law-computing tool.
250

200 H. Acknowledgements.
150
Work supported by CICYT project DPI2004-06470
of the Spanish Ministry for Education and Science.
100
0 5 10 15 20 25 30 35 40
Line
I. References.
Fig. 5. Maximum errors (ns) provided by the Abbe’s invariant A. Lamarre, Dynamic Focusing of Phased Arrays for
based method, with regard to the Fermat method. Nondestructive Testing: Characterization and
Applications, ndt.net, 4, 9, Sept. 1999.
By contrast, the accuracy of the FFLC technique is T. Huttunen et al., “A Full-Wave Helmholtz Model
similar to that of the Fermat’s method (Fig. 6): The for Continuous-Wave Ultrasound Transmission”, IEEE
maximum error for the considered case is below 54 ps Trans. on Ultrasonics, Ferroelectrics, and Frequency
(10 ns allowed). Furthermore, it is valid for interfaces of Control, vol. 52, no. 3, pp. 397-409, March 2005.
arbitrary shape. This turns this method the more attractive
as a general-purpose fast focal law computing tool. A. Ibañez, M. Parrilla, J. Villazón, “Simulation of
Ultrasonic Continuous Wave Fields in Homogeneous
0.054 Media with Soft Curved Interfaces”, ECNDT 2006,
Berlín, Th.3.3.1, 2006.
0.052
P. R. Stepanishen, “Transient radiation from piston in
an infinite planar baffle”, J. Acoust. Soc. Amer., vol. 49,
0.05 pp. 1629–1638, 1971.
0.048
F. Buiochi et al., “A Computational Method to
ns
Calculate the Longitudinal Wave Evolution Caused by
0.046
Interfaces Between Isotropic Media”, IEEE Trans. on
Ultrasonics, Ferroelectrics, and Frequency Control, vol.
51, no. 2, pp. 181-192, February 2004.
0.044
G. A. Deschamps, “Ray techniques in
0.042 electromagnetics”, Proc. of the IEEE, vol. 60, no. 9, pp.
1022-1035, September 1972.
0.04
0 5 10 15 20 25 30 35 40 A. Turnbull, M. Garton, “Ultrasound ray tracing in
Line arbitrary complex geometries”, Rev. Progress in QNDE,
Fig. 6. Maximum errors (ns) obtained with the FFLC method. 14A, 1105-1110, 1995.
D. K. Peterson, G. S. Kino, "Real-Time digital image
G. Conclusions. reconstruction: A description of image hardware and an
analysis of quantization errors", IEEE Trans. on Sonics
An analysis of several methods for fast focal law and Ultrasonics, vol. 31, no. 4, pp. 337-351, 1984.
computing has been carried out. They are mainly
addressed to dynamic focusing applications in presence M. Born, E. Wolf: “Principles of Optics”, Cambridge
of interfaces of arbitrary geometry, where a large number University Press, 1997.
of focal laws have to be computed.

-4-

View publication stats

You might also like