Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control

Manfred Sch

onheits

, Armin Huber

and Philipp G

answ

urger

German Aerospace Center (DLR), Institute of Structures and Design, Augsburg, Germany

Keywords: Non-destructive Testing, Ultrasonic Inspection, Lamb Wave, Automation, Robotics.

Abstract:

Single-sided air-coupled ultrasonic inspection has some beneﬁcial properties compared to water-coupled ul-

trasonic inspection or double-sided ultrasonic testing. The absence of the need for water leads to easier process

handling on the one hand e.g. when manufacturing aircraft components. On the other hand, because the pro-

cess is single-sided, reachability is a minor problem compared to double-sided testing and end-effectors and

ﬁxtures can be designed in a less complex and more compact way. However, the nature of lamb waves requires

that the geometrical relation of the transmitter and the receiver varies during the inspection process. In this

paper, a prototype of an adaptive end-effector is introduced that was developed to implement this requirement

and results of ﬁrst evaluation tests are presented.

1 INTRODUCTION

Aerospace vehicles are constructed from an in-

creasing amount of carbon composites because

their strength-to-weight ratio compared to traditional

metallic components can yield weight savings that en-

able these vehicles to operate more efﬁciently. How-

ever, the manufacturing of composites is still very ex-

pensive due to the high amount of manual labor that

is involved and the use of high-performance but also

high-cost prepreg material. Therefore, it is the aim

of the Center for Lightweight Production Technology

(ZLP) of the German Aerospace Center (DLR) to de-

velop automated manufacturing process technology

for large-scale components made from carbon ﬁber

reinforced plastics (CFRP).

In order to guarantee cost-efﬁciency, process-

integrated quality assurance based on non-destructive

inspection (NDI) methods is essential. It has been

concluded that air-coupled ultrasonic testing (ACUT)

is a suitable NDI method (Ullmann et al., 2012). A

NDI method has to meet certain key criteria: Fast

measurement speed and data evaluation as well as the

ability to inspect large components. This can only be

achieved through automation of the inspection pro-

cess.

The single-sided ACUT mode, which we have uti-

lized in this work, is based on the excitation of Lamb

waves in a component to be tested. This mode is

https://orcid.org/0000-0002-5646-2111

https://orcid.org/0000-0002-5694-8293

also referred to as Focused Slanted Reﬂection Mode

(FSRM). Lamb waves have been used for NDI pur-

poses for many decades. An early description of the

ﬂaw detection of sheets and tubes immersed in water

by means of Rayleigh and Lamb waves was given by

Viktorov already back in 1967 (Viktorov, 1967). Only

a few years later, Luukkala et al. proposed a con-

tactless test method for paper and metal plates based

on Lamb waves (Luukkala et al., 1971; Luukkala

and Meril

ainen, 1973). Many applications have been

established since then, and the advent of compos-

ite materials in automotive and aerospace industries,

which took place in the early 1990s, has added sig-

niﬁcant complexity to the non-destructive testing and

evaluation processes. The ability of guided waves

to propagate many meters in a waveguide is uti-

lized for pipe inspection (Wilcox et al., 2001; Lowe

et al., 1998). They are also used for the inspection

of bonding (Lowe and Cawley, 1994), which is one

of the most challenging tasks, especially in the case

of kissing bonds (Kundu et al., 1998). Other rel-

evant studies concerning NDI and structural health

monitoring (SHM) on composite structures are found

in Refs. (Maslov and Kundu, 1997; Kessler et al.,

2002; Toyama et al., 2003; Su et al., 2006; Diamanti

and Soutis, 2010; Purekar and Pines, 2010; Ramadas

et al., 2011; Cunfu et al., 2013). The air-coupled

version of the guided wave inspection could play an

important role in future production lines. Often, the

presence of a liquid coupling medium is unwanted be-

cause it might inﬂict damage to unsealed composite

430

Schönheits, M., Huber, A. and Gänswürger, P.

Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control.

DOI: 10.5220/0007956904300438

In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 430-438

ISBN: 978-989-758-380-3

structures. All of the following researchers have used

the air-coupled variant. Castaings et al. have done

signiﬁcant work on the single-sided ultrasonic testing

of composites by using Lamb waves (Castaings and

Cawley, 1996; Castaings et al., 1998; Castaings and

Hosten, 2001; Castaings and Hosten, 2008), while

Solodov et al. have used them for transmissive in-

spection (Solodov et al., 2004b; Solodov et al., 2004a;

Solodov et al., 2006).

In order to excite Lamb waves, the ultrasound

transducers must follow a certain orientation with re-

spect to the specimen which we call excitation angle.

This excitation angle depends upon certain properties

of the component, which might change from one lo-

cation to another. Therefore, the inspection process

has to be interrupted and the orientation of the ultra-

sonic transducers must be changed such that the cor-

rect excitation angle is met. In this paper, we present a

prototype of a fully automated, adaptive end-effector

(AEE), which enables the continuous adjustment of

the ultrasonic transducers such that Lamb waves are

excited at optimal efﬁciency at any location on the

component.

2 PROBLEM DESCRIPTION

A Lamb wave propagating through a plate causes dis-

placement throughout its whole thickness. Since ul-

trasound is reemitted along the propagation path of

the Lamb wave, it can be detected with the receiver.

As depicted in ﬁgure 1, both sender and receiver must

be oriented at the same angle with respect to the sur-

face normal when used on a ﬂat specimen. A ﬂaw

causes a change in the displacement amplitude, and

can be detected thereby. A beam shield made of card-

board is mounted between the two transducers to pre-

vent sound propagating directly from the sender to the

receiver. Both transducers can be rotated and moved

horizontally and vertically, but only manually. The

industrial robot scans the specimen while ultrasonic

pulses are triggered such that pairs of 6DOF measure-

ment points and the corresponding ultrasonic data are

generated. By that way, three-dimensional ultrasonic

images of a specimen can be obtained and ﬂaws lo-

calized. However, because the transducers are ﬁxed,

this solution is limited to parts with constant thickness

and zero or constant curvature.

According to Snell’s law, the excitation angle θ

with respect to the surface normal is given by

θ = sin

−1



Lamb



, (1)

where v

is the phase velocity of the incident plane

wave and v

Lamb

the phase velocities of the Lamb wave

Figure 1: Conventional end-effector for air-coupled ultra-

sonic inspection by using Lamb waves.

excited in the plate (Castaings and Cawley, 1996).

While v

is known, e.g., 343 m/s at room temperature

in air, v

Lamb

is not. v

Lamb

depends upon

• The stiffness of the specimen’s material,

• The layup (ﬁber orientation and layer thickness)

of the specimen,

• The ultrasonic transducer frequency since Lamb

waves are dispersive.

With the known stiffness matrix and the layup, one

can calculate Lamb wave dispersion diagrams (θ vs.

frequency) for a given propagation direction in the

laminate. A discussion of this procedure is beyond

the scope of this paper. In our previous work, we have

created an interactive software called Dispersion Cal-

culator (DC) with MATLAB



(MathWorks, Natick,

MA, USA) for the calculation of Lamb wave disper-

sion diagrams (Huber and Sause, 2018). The stand-

alone software can be downloaded free of charge on

the DLR-homepage

Real structural parts generally do not have a con-

stant thickness or a homogeneous layup. For an au-

tomated inspection process this means that the nec-

essary poses of the transducers are not constant but

depend on the location. Figure 2 shows a simpliﬁed

scenario including a part with a ﬂat surface but vari-

able thickness (indicated by magenta color), which

leads to a variable excitation angle. Even with con-

stant ﬁber orientation assumed, this would lead to a

change of the required transducer orientation along

the part. Furthermore a surface with variable curva-

ture would also make a dynamic positioning of the

transducers essential.

In order to automate the process, two major com-

ponents are therefore necessary:

https://www.dlr.de/bt/en/desktopdefault.aspx/

tabid-2478/11208 read-53373/

Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control

431

Transmitter

Receiver

Increasing Thickness

Figure 2: Change of excitation angle.

• A kinematic system that allows dynamic position-

ing of the transducers (an adaptive end-effector in

our case)

• A software solution that, given a description of a

part, creates a program for the kinematic system

including the correct poses for the transducers

3 SPECIMEN

To validate the AEE, a testing specimen was de-

signed. The following requirements were deﬁned:

• The specimen should mimic real components and

material used in aircraft.

• It must have all possible features that cause a

change in the excitation angle, namely variable

thicknesses (layup) and curvature.

• Artiﬁcial ﬂaws must be inserted into the layup to

enable the quantiﬁcation of the probability of de-

tection (POD) of the AEE

The specimen was manufactured from

SAERTEX



7006919 dry fabric, inﬁltrated with

RIMR135 resin by vacuum assisted resin infusion.

The ﬁber volume content is 55 %. A schematic of the

layup is drawn in Figure 3. Figure 3(a) shows a cross

section through the laminate. The base laminate is a

2.2 mm thick, quasi-isotropic layup [0/90/-45/45]

. A

step-like reinforcement layup [0/90]

was placed on

top of the base laminate with a maximum thickness

of 4.4 mm. Therefore, the overall laminate thickness

varies between 2.2 and 6.6 mm. Artiﬁcial ﬂaws made

of 5×5 and 10×10 mm

square pieces of Kapton



sheet (DuPont Inc., Wilmington, DE, USA) were

laminated into different depths. As shown in Fig. 4,

the layup was performed in a small area close to the

tail of an aircraft fuselage preform where signiﬁcant

curvature gradients occur. The specimen is 935 mm

long, 505 mm wide, and has a maximum height of

110 mm. The reinforcing layup covers a length of

420 mm.

(a)

D L R

(b)

Figure 3: Cross sectional view (a) and top view (b) of the

specimen’s layup. Artiﬁcial ﬂaws are drawn in red color.

The dimensions are not true to scale.

Figure 4: The layup of the specimen was done in an aircraft

fuselage preform.

4 SOFTWARE TOOLCHAIN

It turned out that relying on commercial software was

not sufﬁcient for programming and controlling a sys-

tem like the AEE, so a software toolchain had to

be developed that provides the missing functionality.

This section discusses our requirements, the short-

comings of commercial software in this respect, and

our solution of the problem.

4.1 Requirements

To obtain a uniform scan of the part, a grid of i mea-

surement points on the specimen surface with the co-

ordinates x

in the base coordinate system needs

to be generated. This must be done based on a digital

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

432

representation of the specimen. These are then used

to create a program for the robotic system. To do so,

they are put in order in such a way that the robot scans

the specimen on a meander-like path. To fully deﬁne

the kinematic path for the end-effector, the pose for

both the transmitter and the receiver for each mea-

surement point needs to be determined. Because the

Transmitter

Receiver

Excitation

Angle

Measurement

Grid Points

Thickness

Figure 5: Determination of transmitter and receiver poses.

transducers should not collide with each other or the

specimen, they need to be offset from each other and

the surface. Also, the Lamb wave needs a minimum

length it travels through the part. To determine the

orientation of the ultrasonic heads the excitation an-

gles θ

need to be calculated. To enable this, the layup

at each measurement point must be obtained, but this

in turn requires the laminate thicknesses d

. The dis-

tances which the ultrasound propagates in air and in

the laminate also must always kept constant. The re-

sult is a list of transmitter-receiver pairs on the grid,

depicted in ﬁgure 5.

4.2 State-of-the-Art

Usually computer-aided design (CAD) and of-

ﬂine programming (OLP) tools like CATIA

DELMIA

, and FASTSURF

(Dassault Syst

emes,

elizy-Villacoublay, France) are used for the path

planning of robots in an industrial context. Here,

one has a digital representation of the robotic environ-

ment, and the path as well as orientation of the end-

effector can be generated based on the specimen’s

surface. However, the required additional kinemat-

ics of the AEE as well as the algorithm-driven vari-

ability of the transducer poses cannot be modeled

neither programmed with these products in a sufﬁ-

cient way. Furthermore, the possibilities to generate

the measurement grid on the specimen as well as the

layup and thickness determintation d

are insufﬁcient

(a)

X (mm)

400

200

Y (mm)

400

600

End

Start

800

100

Z (mm)

(b)

X (mm)

400

200

Y (mm)

400

600

End

Start

800

100

Z (mm)

(c)

Figure 6: Ofﬂine programming in MATLAB



. The STL

ﬁle is loaded and ﬁtted (a). The path, tripods, and excitation

angles are generated (b). Then, the positions and orienta-

tions of the ultrasonic transducers are determined (c). In

this example, a 50 mm grid is generated whereas a spacing

of 2 mm is more realistic in a real measurement.

in CATIA

For the calculation of θ

, we purchased DIS-

PERSE

(Imperial College London, London, UK).

DISPERSE has been developed since the early 1990s

by Lowe and Pavlakovic, and was used for the valida-

tion of DC. DISPERSE is the leading software in its

ﬁeld and has some features not included in DC. How-

ever, it has a signiﬁcant price while only available

as a node-locked license and currently cannot calcu-

late laminates containing more than 64 layers. DIS-

PERSE also lacks an API, which renders us unable to

automate the calculation of θ

as we potentially have

to calculate high numbers of sample points. For in-

stance, a square grid with 2 mm spacings on the spec-

imen has more then 10

points. In DISPERSE you

must enter each layup manually, calculate the com-

plete dispersion diagram, and then extract the excita-

tion angle of A

at the transducer’s frequency. These

drawbacks have been one of the reasons why DC was

created.

4.3 Solution

The four different software solutions mentioned in

Section 4.2 which had been considered for the soft-

http://www.imperial.ac.uk/non-destructive-evaluation/

products-and-services/disperse

Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control

433

ware tool chain initially were found insufﬁcient later.

Therefore, one task was to overcome their deﬁcien-

cies and solve them by own implementation. Our tool

chain uses two software componets currently, namely

CATIA

and MATLAB



The basis is the CAD model of the specimen in

CATIA

. A separate tesselation of the top and bot-

tom surfaces of the specimen is performed and the

result is exported into two STL ﬁles. These ﬁles can

be loaded into MATLAB



. The top surface of the

specimen is shown in Fig. 6(a). The vertices of the

top surface are used as an input for the built-in MAT-

LAB



Curve Fitting app (it requires the Curve Fit-

ting Toolbox) to obtain a three-dimensional ﬁt. Then

the spacing of the measurement grid is deﬁned and

whether the transducers should move in lines paral-

lel to the x axis or y axis (called ”sweep direction”

below). Square grid coordinates x

are generated

in the x-y plane, covering the x-y range of the point

cloud of the top surface. The grid coordinates are or-

dered in such a manner that alternating sweeps in the

positive and negative sweep direction are performed.

By evaluating the ﬁtting function at x

, the yet miss-

ing third coordinates z

can be obtained. The result-

ing path is indicated by the black lines in Figs. 6(b)

and 6(c). At each measurement grid point x

, the

tripods~n

can now be calculated indicated by, re-

spectively, blue, red, and green lines in Fig. 6(b).

Next, the thicknesses d

are determined by prob-

ing. The algorithm uses the surface normals ~n

, seeks

the vertex of the lower surface, which is pierced by

, and calculates the normal distance d

between this

vertex and the corresponding grid point. The thick-

nesses d

are indicated by magenta colored lines in

Fig. 6(b) (their lengths are scaled up by a factor of

ten for clarity). With the now known d

, the local

layup can be obtained as follows. It is known that the

specimen has twenty-four layers where it is 6.6 mm

thick. Hence, the number of layers at any given mea-

surement point from d

can be deduced. Then, from

the sequence of layer orientations in the “maximum”

layup, the local layup at the ith measurement point

can be obtained.

For the θ

calculation, the basic algorithm used by

DC was used. Therein, depending on the sweep di-

rection, the propagation of Lamb waves is supposed

to be either along the 0 or 90

◦

direction of the layup.

Instead of complete dispersion diagrams θ

( f ), where

f covers a broad frequency range, only θ

(200kHz)

for the A

Lamb wave are calculated.

The excitation angles θ

(200kHz) vary between

14.4

◦

where the specimen is 2.2 mm thick and 13.2

◦

where it is 6.6 mm thick. The excitation angles are

indicated by gray lines in Figure 6(b).

Now that that the coordinate system frames and

excitation angles for each measurement point are

known, the transmitter and receiver poses can be cal-

culated. This is done by applying a local coordinate

system displacement that adjusts the poses of the ul-

trasonic heads according to the excitation angle and

ﬁxed distance offsets on and from the surface (see

Figure 5 ). Throughout the experiments discussed in

this paper, 50 mm for both offsets are used. The ﬁnal

prepared data set is called the ”excitation angle map”

(EAM).

5 ROBOTIC SETUP

The process requires that the position of the transmit-

ter and the receiver with respect to each other varies

over time according to the EAM. That is why a sin-

gle kinematic is not sufﬁcient to implement the pro-

cess. At the DLR site in Augsburg there are a num-

ber of industrial-grade robot cells available. The idea

to implement the process was to use one of the in-

dustrial robots and attach a second smaller kinematic

arm. The industrial robot serves as a kinematic for

Figure 7: Robot cell.

positioning the transmitter, while the second attached

arm is responsible for positioning the receiver. The

robotic cell that served as a basis for evaluation of the

adaptive testing is shown in ﬁgure 7. The robot is a

KUKA KR120 2700 HA run by a KRC4 controller.

5.1 End-effector Design

First preliminary tests have been conducted with a

KUKA LBR iiwa as a second kinematic attached to

the KR120, as this robot arm was available at the DLR

site in Augsburg already.

These tests have been

promising, that is why a dedicated end-effector was

https://www.kuka.com/en-de/products/robot-

systems/industrial-robots/kr-quantec

https://www.kuka.com/en-de/products/robot-

systems/industrial-robots/lbr-iiwa

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

434

designed. The new design should be as compact and

as simple as possible, so it was decided to reduce the

functionality to the necessary minimum (see ﬁgure 8).

Figure 8: End-effector with iiwa (left) and new end-effector.

It was found sufﬁcient that the ultrasonic transduc-

ers can move in a planar way to each other with addi-

tional rotation, so at least three degrees of freedom are

necessary. Basically this can be implemented as e.g. a

combination of two prismatic joints and one revolute

joint or three coplanar revolute joints. The latter has

been chosen because it was assessed that this way it

is easier to build a compact, collision-free design with

the neccessary workspace. We had been provided

with three X-Series actuators by HEBI

Robotics

The end effector consists of a quick-release plate

which is a standard at the DLR Augsburg, an alu-

minum proﬁle and the three X-Series actuators. The

actuators were mounted on aluminum links as close as

possible to each other without collision. The default

position of the static and the moving ultrasonic trans-

ducers are parallel to each other and the two links in a

45-90-45 degree position. With this default position,

a widely adjustable angle between the transducers is

possible. One advantage of this conﬁguration is re-

ducing the torque on axis 1 and 2 while providing a

good workspace. An option to mount a direct sound

shield (like the cardboard shield shown in ﬁgure 1)

was added. Figure 9 shows the design.

Figure 9: Re-designed end effector.

To increase the precision of the end-effector, all

of the ﬁxed parts as well as the quick release plate

of the end effector were equipped with counterbores

that can be measured with a LEICA

laser tracker

https://www.hebirobotics.com/

that is available at the DLR Augsburg site. This way

the intrinsic dimensions of the end-effector could be

calibrated with high precision, merely leaving the po-

sition accuracy of the actuators as a source of impre-

cision.

Four markers that help to calibrate the part in the

robot cell have been addded to the corners of the spec-

imen part described in section 3. To position the spec-

imen part for the process, a framework to mount the

specimen on had to be manufactured. The framework

was derived in CATIA from the lower face of the part

and produced on a RIDDER

WariCut waterjet cut-

ter. This was also used on the waterjet cutter for trim-

ming the specimen. With this form ﬁtting framework,

it is easy to place the specimen precisely on a welding

table in front of the robot repeatable and so it is not

necessary to recalibrate its position after removing it.

The framework and the specimen are shown in ﬁgure

10.

Figure 10: Specimen mounted on the framework and as-

sembled end-effector (direct sound shield not mounted).

5.2 Process Control

To make the process work, two main aspects had to

be taken into consideration:

• The KUKA robot and the HEBI arm have to fol-

low the trajectory deﬁned by the transmitter / re-

ceiver poses as waypoints

• The trajectories need to be synchronized in a way

so that both robots reach each waypoint at the

same time

For the following experiments, a master-slave-

principle was implemented: The transmitter is

mounted on the KUKA robot that moves along the

specimen. The receiver is mounted on the HEBI

arm, which adjusts itself according to the position of

the transmitter. In order to generate a trajectory for

the KUKA robot, a KRL (KUKA Robot Language)

https://www.ridder.de/

Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control

435

robot program for its robot controller gets generated

from the calculated transmitter poses. This is done

by a KRL generator that was implemented as part of

our MATLAB toolchain discussed earlier (Section 4).

The HEBI robotics API includes a generic kinemat-

ics solver, but for our purpose a forward and inverse

kinematic solver was implemented speciﬁcally for our

arm conﬁguration. The poses for the HEBI arm also

get generated from MATLAB as a CSV

ﬁle.

The robot controller sends the transmitter pose

and the current waypoint number to a PC in realtime

over the KUKA robot sensor interface (RSI) in a 4 ms

(≡ 250 Hz ) cycle. The PC uses the transmitter pose

and the waypoint number to interpolate the receiver

pose in realtime and uses the HEBI API to replan the

arm trajetory in realtime accordingly. The feedback

of the HEBI arm runs at a 5 ms (≡ 200 Hz ) cycle.

6 EXPERIMENTAL EVALUATION

As a ﬁrst step, a set of parameters for the ultrasound

device was determined experimentally. This was done

by moving the end-effector across the part with man-

ual control and adjusting the settings of the ultrasound

device so that the measured signal was reasonable in

terms of amplitude, phase and other factors. Once a

good set of parameters was found, it was kept con-

stant for all the following tests. Instead of cardboard,

a piece of EPDM

rubber was used as a direct sound

shield.

Throughout all the tests, the trajectory of the

KUKA robot was generated from the calculated trans-

mitter poses, i.e. the pose of the transmitter can be

assumed to be correct all the time. The speed of the

KUKA was set to 100 mm/s. Then the following tests

have been done.

1. Keep the receiver pose in a conﬁguration that is

correct for the thick part in the middle of the spec-

imen, scan the whole part with the receiver in this

conﬁguration

2. Keep the receiver pose in a conﬁguration that is

correct for the area with the strongest curvature,

scan the whole part with the receiver in this con-

ﬁguration

3. Scan the whole part using the adaptive process,

with the receiver adjusting its pose continuously

Test 1 and 2 basically mimic a non-adaptive end-

effector (like the conventional end-effector shown in

ﬁgure 1) with the transmitter and receiver being in a

ﬁxed conﬁguration to each other.

comma separated value

ethylene propylene diene monomer

700

600

500

Y (mm)

400

300

200

100

400

200

Z (mm)

X (mm)

Amplitude

0.2

0.4

0.6

0.8

Figure 11: Result with receiver pose adjusted to thickest

area.

Figure 11 shows the result of test 1. As expected,

a reasonable amplitude can mainly be observed in the

thick area of the part. The amplitude is not at a max-

imum, which is because the power settings initially

had to be set to a lower level that does not generate

overdrive in the thinner areas.

700

600

500

Y (mm)

400

300

200

100

400

200

Z (mm)

X (mm)

Amplitude

0.2

0.4

0.6

0.8

Figure 12: Result with receiver pose adjusted to strongest

curvature.

The result of test 2 is depicted in ﬁgure 12. One

can see that a signiﬁcant amplitude gets detected in

the strongly curved area. But because the pose of the

receiver is incorrect for other areas of the specimen,

no usable signal can be observed there.

The result of test 3, using the adaptive process,

is shown in ﬁgure 13. One can see that generally a

signal can be obtained throughout the whole surface

area. A less strong amplitude can be observed in the

thicker area, which is expected. This result conﬁrms

700

600

500

Y (mm)

400

300

200

100

400

200

Z (mm)

X (mm)

Amplitude

0.2

0.4

0.6

0.8

Figure 13: Result of the adaptive process.

that the adaptive process generally is working accord-

ing to our assumptions.

However, there is still a lot of aspects to investi-

gate and a lot of room for improvement. For instance,

the precision is still quite coarse and the ﬂaw detec-

tion not sufﬁciently reliable. Some ﬂaws show up in

the scan (see ﬁgure 13), some do not. Also the inﬂu-

ence of varying the process parameters has not been

investigated in detail. Multiple factors for improve-

ment have been identiﬁed:

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

436

Direct Sound Shield: The shield built into the end-

effector is mounted in a ﬁxed position. In some areas

of the part, a little gap between the surface and the part

could be observed, allowing sound directly travelling

from the transmitter to the receiver, inﬂuencing the

lamb wave amplitude detection.

Lamb Wave Propagation Length: Fixed lengths of

50 mm for the offsets on and from the surface were

used. While the distance from the surface could be re-

duced, the transducers cannot be moved more closely

together because of their size. This causes the ﬂaws

to appear ”stretched” along the sweep direction.

Process Control Accuracy: So far, no in-depth tests

have been conducted how accurate the process con-

trol system works, i.e. how precise the transducers

follow the waypoints and how well their trajectories

are synchronized.

7 CONCLUSIONS AND FUTURE

WORK

In this paper, we have introduced an approach for air-

coupled ultrasonic inspection through adaptive lamb

wave control. A software toolchain to fully calcu-

late and determine the adaptive process was devel-

oped. An adaptive end-effector was designed for ex-

perimental evaluation. For this purpose, a specimen

part was manufactured.

It was possible to verify the principle and the

toolchain as well as the adaptive end-effector were

successfully tested. While the process was found to

be working generally, still a lot of room for improve-

ment exists. Future work could include improving the

hardware such as the direct sound shield. Further-

more, diffent combinations of ultrasonic heads and

other specimens could be examined. Also, a sophisti-

cated analysis of the process control system could be

revealing.

ACKNOWLEDGEMENTS

We would like to thank HEBI Robotics

for provid-

ing us a set of X-Series Actuators for the experimental

evaluation.

REFERENCES

Castaings, M. and Cawley, P. (1996). The generation, prop-

agation, and detection of Lamb waves in plates us-

https://www.hebirobotics.com/

ing air-coupled ultrasonic transducers. J. Acoust. Soc.

Am., 100(5):3070–3077.

Castaings, M., Cawley, P., Farlow, R., and Hayward, G.

(1998). Single sided inspection of composite mate-

rials using air coupled ultrasound. J. Nondestr. Eval.,

17(1):37–45.

Castaings, M. and Hosten, B. (2001). Lamb and SH

waves generated and detected by air-coupled ul-

trasonic transducers in composite material plates.

NDT&E Int., 34:249–258.

Castaings, M. and Hosten, B. (2008). Ultrasonic guided

waves for health monitoring of high-pressure compos-

ite tanks. NDT&E Int., 41:648–655.

Cunfu, H., Hongye, L., Zenghua, L., and Bin, W. (2013).

The propagation of coupled Lamb waves in multi-

layered arbitrary anisotropic composite laminates. J.

Sound Vib., 332:7243–7256.

Diamanti, K. and Soutis, C. (2010). Structural health mon-

itoring techniques for aircraft composite structures.

Prog. Aeronaut. Sci., 46:342–352.

Huber, A. M. A. and Sause, G. R. (2018). Classiﬁcation

of solutions for guided waves in anisotropic compos-

ites with large numbers of layers. J. Acoust. Soc. Am.,

144(6):3236–3251.

Kessler, S. S., Spearing, S. M., and Soutis, C. (2002).

Damage detection in composite materials using Lamb

wave methods. Smart Mater. Struct., 11:269–278.

Kundu, T., Maji, A., Gosh, T., and Maslov, K. (1998). De-

tection of kissing bonds by Lamb waves. Ultrasonics,

35:573–580.

Lowe, M. J. S., Alleyne, D. N., and Cawley, P. (1998). De-

fect detection in pipes using guided waves. Ultrason-

ics, 36:147–154.

Lowe, M. J. S. and Cawley, P. (1994). The applicability

of plate wave techniques for the inspection of adhe-

sive and diffusion bonded joints. J. Nondestr. Eval.,

13(4):185–200.

Luukkala, M., Heikkila, P., and Surakka, J. (1971). Plate

wave resonance - A contactless test method. Ultra-

sonics, 9(4):201–208.

Luukkala, M. and Meril

ainen, P. (1973). Metal plate testing

using airborne ultrasound. Ultrasonics, 11(5):218–

221.

Maslov, K. and Kundu, T. (1997). Selection of Lamb modes

for detecting internal defects in composite laminates.

Ultrasonics, 35:141–150.

Purekar, A. S. and Pines, D. J. (2010). Damage detection in

thin composite laminates using piezoelectric phased

sensor arrays and guided Lamb wave interrogation. J.

Intell. Mater. Syst. Struct., 21:995–1010.

Ramadas, C., Balasubramaniam, K., Joshi, M., and Krish-

namurthy, C. V. (2011). Sizing of interface delami-

nation in a composite T-joint using time-of-ﬂight of

Lamb waves. J. Intell. Mater. Syst. Struct., 22:757–

768.

Solodov, I., Pﬂeiderer, K., Gerhard, H., and Busse, G.

(2004a). Nonlinear acoustic approach to material

characterisation of polymers and composites in tensile

tests. Ultrasonics, 42:1011–1015.

Air-coupled Ultrasonic Inspection with Adaptive Lamb Wave Control

437

Solodov, I., Pﬂeiderer, K., Gerhard, H., Predak, S., and

Busse, G. (2006). New opportunities for NDE with

air-coupled ultrasound. NDT&E Int., 39(3):176–183.

Solodov, I., Stoessel, R., and Busse, G. (2004b). Material

characterization and NDE using focused slanted trans-

mission mode of air-coupled ultrasound. Res. Nonde-

str. Eval., 15(2):65–85.

Su, Z., Ye, L., and Lu, Y. (2006). Guided Lamb waves for

identiﬁcation of damage in composite structures: A

review. J. Sound Vib., 295:753–780.

Toyama, N., Noda, J., and Okabe, T. (2003). Quantitative

damage detection in cross-ply laminates using Lamb

wave method. Compos. Sci. Technol., 63:1473–1479.

Ullmann, T., Shi, Y., Rahner, N., Schm

ucker, M., Fey, P.,

Busse, G., and Becker, S. (2012). Quality Assurance

for the Manufacturing of Oxide Fiber Reinforced Ce-

ramic Composites for Aerospace Applications. In 4th

International Symposium on NDT in Aerospace, pages

1–11.

Viktorov, I. (1967). Rayleigh and Lamb Waves: Physical

Theory and Applications. Plenum, New York.

Wilcox, P. D., Lowe, M. J. S., and Cawley, P. (2001). Mode

and transducer selection for long range Lamb wave

inspection. J. Intell. Mater. Syst. Struct., 12:553–565.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

438