NON-CONTACT ULTRASONIC GUIDED-WAVE DEFECT DETECTION SYSTEM
FOR RAILS
Stefano Coccia1, Robert Phillips1, Claudio Nucera1, Ivan Bartoli2, Salvatore Salamone3,
Francesco Lanza di Scalea1, Mahmood Fateh4, Gary Carr4
1NDE & Structural Health Monitoring Laboratory, University of California, San Diego
E-mails: [email protected], [email protected], [email protected], [email protected]
2 Department of Civil, Architectural and Environmental Engineering, Drexel University
E-mail: [email protected]
3Department of Civil, Structural and Environmental Engineering, State University of New York
at Buffalo
E-mail: [email protected]
4Office of Research and Development, Federal Railroad Administration
E-mails: [email protected], [email protected]
© 2011 AREMA ®
ABSTRACT
The University of California at San Diego (UCSD), under a Federal Railroad Administration
(FRA) Office of Research and Development (R&D) grant project, is developing a system for rail
defect detection. The project is also in direct response to Safety Recommendations issued by the
National Transportation Safety Board (NTSB) following the disastrous train derailments at
Superior, Wisconsin in 1992 and Oneida, New York in 2007 among others. A prototype has been
designed and field tested with the support of Volpe National Transportation Systems Center and
ENSCO, Inc. The goal of this project is to develop a rail defect detection system that provides (a)
better defect detection reliability (including internal transverse head defects under shelling and
vertical split head defects), and (b) higher inspection speed than achievable by current rail
inspection systems. A new effort for further prototype improvements envisions adding rail
surface characterization capability to the internal flaw detection capability.
The UCSD prototype uses non-contact ultrasonic probing of the rail head (laser and air-
coupled), ultrasonic guided waves, and a proprietary real-time statistical analysis algorithm that
maximizes the sensitivity to defects while minimizing false positives. The current design allows
potential inspection speeds up to 40 mph, although all field tests have been conducted only up to
15 mph so far.
This paper summarizes (a) the results of the latest technology development test conducted at
the rail defect farm of Herzog, Inc. in St Joseph, Missouri in June 2010, and (b) the construction
of a new Rail Defect Farm facility at the UCSD Camp Elliott Field Station, with in-kind
contribution from the Burlington Northern Santa Fe (BNSF) Railway.
© 2011 AREMA ®
INTRODUCTION
Conventional ultrasonic rail inspection uses piezoelectric transducers that are coupled to the
top of the rail with ultrasonic wheels or sleds filled with water or other fluids (1). The most
serious drawback of this method is that surface shallow cracks (shelling) can mask the internal
transverse defects. This limitation was the cause of train derailments in Superior, Wisconsin in
1992 and Oneida, New York in 2007, where severe problems were caused by hazardous material
spillage. In response to these accidents and others, the NTSB issued Safety Recommendations to
the FRA for improving the effectiveness of rail inspection technologies to detect internal rail
defects, particularly under shelling (2). Other drawbacks of wheel-based ultrasonic rail
inspections are the limited speed (typically less than 15 mph) and challenges in detecting
Vertical Split Head defects, also critical for rail safety.
Figure 1. Transverse Fissure (TF), Detail Fracture (DF), and FRA Safety Statistics Data for
1998-2008 – rail, joint bar and rail anchoring – all US railroads.
FRA Safety Statistics Data (3) report that train accidents caused by track failures including
rail, joint bars and anchoring resulted in 3,386 derailments and $685M in associated damage
TF DF
Type of Defect % Total Defects
Direct Damage Cost
# Derailments
Transverse/Compound Fissure
23 % (1st leading
cause)
$ 160 M(highest cost) 815
Detail Fracture12 %
(2nd leading cause)
$ 137 M(2nd highest
cost)427
Type of Defect % Total Defects
Direct Damage Cost
# Derailments
Transverse/Compound Fissure
23 % (1st leading
cause)
$ 160 M(highest cost) 815
Detail Fracture12 %
(2nd leading cause)
$ 137 M(2nd highest
cost)427
© 2011 AREMA ®
costs during the decade 1998-2008. The first leading cause of these accidents was the Transverse
Fissure (TF) defect, shown in Figure 1, found responsible for 815 derailments and $160M in cost
during the same time period. Another type of Transverse Defect is a Rolling Contact Fatigue
(RCF) defect that typically initiates at the gage corner of the railhead. The Detail Fracture (DF),
also shown in Figure 1, is the most common RCF defect, and was responsible for 427
derailments and $137 M in associated damage cost (2nd highest cost) during 1998-2008 in the
US.
Based on these statistics, the primary targets of the UCSD/FRA rail inspection prototype are
Transverse Defects (TFs and DFs), including under shelling, as well as Vertical Split Heads and
Compound Fractures. Vertical Split Heads, in particular, are sometimes challenging to detect
with conventional ultrasonic search units due to their longitudinal orientation (1).
The UCSD/FRA system uses non-contact means of transduction of ultrasonic waves in the
rail head (laser and air-coupled sensors) (4). Lift-off distances for the sensors are on the order of
2” from the top of the rail head. The system also uses a proprietary signal processing algorithm
based on statistical analysis which maximizes the defect indications and minimizes false positive
indications. The type of ultrasonic waves used, contrarily to other rail ultrasonic systems that use
bulk waves (longitudinal or shear), is guided waves. The ultrasonic guided modes insonify a
large portion of the railhead and allow for a longer gage length which, in turn, increases the
achievable inspection speed. In addition, specific guided wave modes and frequencies are used to
maximize the sensitivity to the Transverse Defects as well as to the Vertical Split Head defects.
The prototype has been tested at speeds up to 15 mph in the field, although higher speeds are
potentially possible. The maximum speed potentially achievable with the current design is on the
order of 40 mph, although this speed has never been tested in the field. Higher speeds would
© 2011 AREMA ®
require some modifications to the hardware design. Figure 2 shows pictures of the prototype
towed by the FRA R-4 hy-railer during a field test.
Figure 2. The UCSD/FRA rail inspection prototype towed by the FRA R-4 hy-railer.
RESULTS OF FIELD TESTS AT HERZOG, INC.
The present section summarizes the results of two blind tests conducted during the
technology development tests of June 2010 at Herzog Services, Inc. in St. Joseph, Missouri.
ENSCO, Inc. provided field test support. Figure 3 shows pictures of the prototype and of some of
the participants to these tests.
© 2011 AREMA ®
Figure 3 – The UCSD prototype at Herzog and picture of the test participants.
The test track included twelve railhead defects, including Detail Fractures, Transverse
Defects Under shelling, Defective Field and Plant Welds, Side Drilled Holes (simulating TDs),
and Horizontal and Vertical Split Heads. Both of the blind tests were conducted at low speed (~2
mph) and mostly on tangent 136 RE track.
Two different signal processing approaches were used for the two tests. One configuration
was less sensitive to small railhead discontinuities (blind test 1) than the other one (blind test 2).
Ten out of twelve (10/12) defects were correctly detected by blind test #1 (“less sensitive
configuration”), while eleven out of twelve (11/12) defects were correctly detected by blind test
#2 (“more sensitive configuration”). Blind test #1 therefore had an 83.34 % Detection
Rate with zero False Positives (0 F. P.). Blind test #2 had a 91.67% Detection Rate at the cost of
four False Positives (4 F. P.). However, following hand-mapping of the test area, three out of the
four (3 out of 4) False Positives mapped to shallow defects under shelling. Hence the effective
False Positive detection for blind test #2 was as low as 1 F.P.
The UCSD list provided for both blind test #1 and #2 did not include an 80% Defective Plant
Weld (DFW). However, this defect was correctly detected by the system, but not included in the
© 2011 AREMA ®
original list because considered a weld. Later in the tests a method was identified to distinguish
“good welds” from “defective weld” based on their different ultrasonic signature. When this
differentiation was applied, the 80% DFW was consistently detected as a defect by the system.
Figure 4 plots the results of the two blind tests along with Industry Average and AREMA
Recommendation for reliability of defect detection (particularly TDs). The detection percentage
was computed as the number of detected defects divided by the total number of defects of a
given size class. The 21-40% size class was not tested since no defect of such size was present
on the track. The plot shows that the performance of the UCSD/FRA system compared very
favorably with Industry Average and AREMA Recommendations in all defect size classes,
including the largest size class of 81-100% once the weld differentiation method was
implemented.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
110%
1‐10% 11‐20% 21‐40% 41‐80% 81‐100%
Ultrasonic Flaw Detection Percentage
Transverse Defect Flaw Size
INDUSTRY AVERAGE
AREMA RECOMM.
UCSD blind test (list 1)
UCSD blind test (list 2 ) ‐ after weld differentiation
UCSD blind test (lists 1 and 2 ) ‐ prior to weld differentiation
Figure 4 - Results of UCSD blind tests at Herzog rail defect farm on June 15, 2010 (comparison with Industry Average and AREMA Standards).
© 2011 AREMA ® ®
Influence of Rail Surface Conditions
The Herzog test track contained different levels of railhead surface discontinuities, including
shells and head checks. Defects in the UCSD system are detected as peaks in a Damage Index
plot that is computed and plotted in real-time at each position along the rail by the statistical
signal processing algorithm. Different scales of visualization of the Damage Index plots were
used for the different areas of the rail. This was done to adapt the response of the system to
different surface conditions of the rail. In the different conditions of rail surface encountered in
the tests, the defect-free level of the Damage Index was varying, but the defect indications were
still distinguishable from the noise level. An Automatic Gain Control feature, similar to what
used in common ultrasonic rail inspections, should be implemented in the final configuration of
the system to rescale the data in the presence, for example, of moderate and heavy shelling.
On the other hand, the system sensitivity to different rail surface conditions could also be an
interesting feature, potentially used to estimate the severity of shelling, hence the thickness of the
layer that needs to be grinded during rail maintenance. This capability will be investigated in
depth in a future phase of this project.
Figure 5 shows one of the test runs at Herzog over a section of rail with head checks. Notice
that the Damage Index plot is sensitive to the surface condition of the rail; however, the 10%
H.A. Defective Field Weld at position 87’2” is still well recognizable over the noise floor.
© 2011 AREMA ®
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
70 72 74 76 78 80 82 84 86 88 90
Figure 5 – Damage Index plot, test run conducted at 2 mph.
Weld Signatures
One achievement of the tests was the realization that the system showed a different response
between “good welds” and “defective welds,” hence allowing for the detection of weld defects.
This differentiation is not always achievable by current ultrasonic rail inspection systems
because the coarse grain structure of welds often prevents the high frequency ultrasonic beams to
penetrate. As shown in the example of Figure 6, the prototype detected a good weld at feet 54’6”,
a 10% TD at feet 55’0” and a 50% Defective Field Weld at feet 59’2”. It can be seen in this plot
that the response to a “good weld” is a high-level stable plateau with no local minima points,
while defects and “defective welds” produce a more “jumpy” Damage Index with several local
minima points. This behavior could be used to train an automatic defect classification algorithm
or used under operator’s judgment to detect defects within welds.
10% Defective Field Weld Joint
Surface Head Checks “Clean” rail
Position (ft)
Sta
tist
ical
Dam
age
Inde
x
© 2011 AREMA ®
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
50 51 52 53 54 55 56 57 58 59 60
Figure 6 – Example of different signatures of “good weld”, “transverse defect” and “defective weld.”
Detection of Vertical Split Head Defect
The prototype primary goal is the detection of Transverse Defects that, as discussed above,
are historically the number one cause of concern for train accidents as far as rail-related defects.
However, the number two cause of concern is Vertical Split Head (VSH) defects. The tests at
Herzog demonstrated that the system has also an excellent reliability of detection of the VSH
defect. This is an important achievement, since VSH are often missed by conventional ultrasonic
rail inspections because their orientation may not generate a strong reflection of the ultrasonic
beam from ultrasonic wheel search units. Figure 7 shows an example of detection of a 1’ long
VSH defect present at Herzog’s rail defect farm.
Good weld
10% Transverse Defect 50% Defective Field Weld
Position (ft)
Sta
tist
ical
Dam
age
Inde
x
© 2011 AREMA ®
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
30 31 32 33 34 35 36 37 38 39 40
Figure 7 – Example of clear detection of a 1’ long Vertical Split Head (VSH) defect (two Joints and a Horizontal Split Head defect also shown as detected).
THE NEW UCSD RAIL DEFECT FARM
It was determined in 2009 that further development of the rail inspection prototype required a
new defect farm on site at UCSD. Such facility has now been constructed at the UCSD Camp
Elliott Field Station, about 8 miles from the main UCSD campus. The Camp Elliott Field Station
hosts also some of UCSD’s high-visibility structural testing facilities, including the world-only
Outdoor Shake Table for earthquake engineering testing and the Blast Simulator Facility for
blast studies.
1’ Vertical Split Head
Joint
2” Horizontal Split Head
Joint
Position (ft)
Sta
tist
ical
Dam
age
Inde
x
© 2011 AREMA ®
The new UCSD/FRA Rail Defect Farm (Figure 8) is a 250-ft long track, with a tangent
portion and an 8-deg curved portion. BNSF donated tracks and ties. Sopac Rail, Inc. performed
the construction. The track features about 15 natural rail defects, including TDs under shelling,
and some artificial rail defects. This facility will be used primarily for the technology
development of the FRA/UCSD rail inspection prototype. The facility will also be available to
other developers of rail inspection technologies of interest to the FRA or industry.
Figure 8 – The new Rail Defect Farm at the University of California San Diego for development of rail inspection technologies.
DISCUSSION AND CONCLUSIONS
The performance of the UCSD/FRA rail inspection system at Herzog’s rail defect farm was
very promising. Two blind tests were conducted at slow speed (~2 mph). The system was able to
© 2011 AREMA ®
detect, reliably, Transverse Defects including some under shelling, Side Drilled Holes, artificial
Horizontal and Vertical Split Heads and Defective Field and Plant Welds. The defect detection
reliability shown during the blind tests exceeded industry average and AREMA
recommendations. The system was also sensitive to the presence of good welds, but with a
different signature than the one related to the flaws. Testing at higher speed (up to the allowed 9
mph in the Herzog’s test track) was also conducted after the blind tests. The system performed
well at these speeds, although with a decreased position resolution compared to the lower speeds.
Modifications to the system hardware should be made to achieve robust performance at the
higher speeds.
Interesting outcomes of the Herzog tests were also the excellent detectability of the Vertical
Split Head defect, and the potential for characterizing different rail surface conditions which
could be useful to better schedule rail grinding maintenance. It should be emphasized that the
VSH defect at the Herzog defect farm was an artificial "man made" defect. VSH or rail shear
defects developing from rail manufacturing processes or caused by fatigue may produce
different results. The signal-to-noise ratio of the defect indications was very satisfactory.
Clearly, a more robust assessment of the defect detection reliability of the system will require
testing on a larger variety of defects.
This paper also reported on the completion of the new UCSD Rail Defect Farm facility, a
250-ft long track with a number of artificial and real defects built with FRA funding and BNSF
in-kind support. This facility will be available for technology development of the UCSD rail
inspection system, as well as available to other developers of rail inspection technologies.
© 2011 AREMA ®
ACKNOWLEDGMENTS
This work was supported by the U.S. Federal Railroad Administration under grants DTFR53-
02-G-00011 and FR-RRD-0001-10-01-00. Mahmood Fateh from the FRA Office of Research
and Development is the Program Manager. The National Science Foundation funded the initial
research effort. John Choros of Volpe Center participated to the field tests at Herzog as advisor
and evaluator. ENSCO, Inc. provided field support for these tests, and Eric Sherrock is
particularly acknowledged for his role in this important support. Special thanks are extended to
Troy Elbert and Rick Ebersold of Herzog, Inc. for providing access to the rail defect farm, and to
John Stanford and Scott Staples of BNSF for arranging for the donation of materials for the Rail
Defect Farm in San Diego.
REFERENCES
1. Lanza di Scalea, F. Ultrasonic Testing Applications in the Railroad Industry. Chapter 15:
Special Applications of Ultrasonic Testing, in Non-destructive Testing Handbook, 3rd edition,
P.O. Moore, ed., American Society for Nondestructive Testing, 2007, pp. 535-552.
2. National Transportation Safety Board (NTSB) Reports HZM-94/01 and RAB-08/05.
3. Federal Railroad Administration. Safety Statistics Data: 1998-2008, FRA, U.S. Department
of Transportation.
4. Coccia, S., Bartoli, I., Phillips, R., Salamone, S., Lanza di Scalea, F., Fateh, M., and Carr. G.
UCSD/FRA Ultrasonic Guided-Wave System for Rail Inspection. Proceedings of the
AREMA Annual Conference, Chicago, IL, September 20-23, 2009.
© 2011 AREMA ®
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Non-contact Ultrasonic Guided-Wave Defect Detection System for Rails
S. Coccia, R. Phillips, C. Nucera,F. Lanza di Scalea
University of California, San Diego
I. BartoliDrexel University, Philadelphia
S. SalamoneSUNY Buffalo, New York
M. Fateh, G. CarrFederal Railroad Administration
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
FRA Safety Statistics data
for period 1998‐2008
Within category: rail, joint bar
and rail anchoring – all US railroads
Type of Defect % Total Defects
Direct Damage Cost
# Derailments
Transverse/Compound Fissure
23 % (1st leading
cause)
$ 160 M(highest cost) 815
Detail Fracture12 %
(2nd leading cause)
$ 137 M(2nd highest
cost)427
Type of Defect % Total Defects
Direct Damage Cost
# Derailments
Transverse/Compound Fissure
23 % (1st leading
cause)
$ 160 M(highest cost) 815
Detail Fracture12 %
(2nd leading cause)
$ 137 M(2nd highest
cost)427
Research Motivation
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Project Objectives
Develop a defect detection system for rails that, compared to current technology, can provide:
(1) increased reliability of defect detection to prevent derailments
(2) increased test speed- current rail inspections <15 mph
(3) New objective: characterize rail surface defects (e.g. depth and density of RCF)
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Prototype Concept• Ultrasonic guided waves used as main probing means, in
alternative to ultrasonic bulk waves (L- and S-waves) used in current wheel search units
• Rail flaws detected by analyzing the guided wave measurements through unique statistical pattern recognition algorithm (USPTO)
• Current configuration uses non-contact means of exciting and detecting the ultrasonic guided waves in the rail
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
These simulations have helped designing the UCSD/FRA inspection prototype
Finite Element Simulations of Ultrasonic Guided Waves: examples of different guided wave modes in rails
(proprietary SAFE and COMSOL analysis)
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
FEA simulation of TD detection
Finite Element Simulations of Ultrasonic Guided Waves: example of TD detection
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Prototype Development
1st field test (Gettysburg, PA) Mar 2006
Laboratory test (UCSD) Jan 2006
4th field test (Gettysburg, PA) Dec 2008
6th field test (TTC, Pueblo, CO) Jun 2009
7th field test (Herzog, Inc., St. Joseph, MO) Jun 2010
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Prototype Performance
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 150
2000
4000
6000
8000
10000
Position (Feet)
Test # 2 on 03-May-2010 First D.I. feature # 18 20 21 2
Sta
tist.
Dam
age
inde
x
10%H.A. TD under “light” shelling
Flaw detection threshold
10%H.A. TD under “severe” shelling
join
t
join
t
join
t
join
t
8%H.A. TD(no shelling)
Example of defect detection under shelling – UCSD rail test site
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Field Tests at Herzog, June 2010
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Field Tests at Herzog, June 2010
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
110%
1‐10% 11‐20% 21‐40% 41‐80% 81‐100%
Ultrasonic Flaw Detection Percentage
Transverse Defect Flaw Size
INDUSTRY AVERAGE
AREMA RECOMM.
UCSD blind test (list 1)
UCSD blind test (list 2 ) ‐ after weld differentiation
UCSD blind test (lists 1 and 2 ) ‐ prior to weld differentiation
Results of two blind tests: exceeded AREMA standards and industry ave.
Blind list 1: 92% Overall Detection Rate, 0 False PositivesBlind list 2: 100% Overall Detection Rate, 4 False Positives (3 of which questionable)
This defect size not tested
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Field Tests at Herzog, June 2010Transverse Defects: excellent detectionWeld signatures: differentiate good welds from defective welds
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
50 51 52 53 54 55 56 57 58 59 60
Good weld
10% Transverse Defect 50% Defective Field Weld
Position (ft)
Statistical D
amage Index
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Field Tests at Herzog, June 2010Clear detection of VSH and HSH defects
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
30 31 32 33 34 35 36 37 38 39 40
1’ Vertical Split Head
Joint
2” Horizontal Split Head
Joint
Position (ft)
Statistical D
amage Index
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Field Tests at Herzog, June 2010Influence of rail surface condition: of interest to grinding operations
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
70 72 74 76 78 80 82 84 86 88 90
10% Defective Field Weld Joint
Surface Head Checks “Clean” rail
Position (ft)
Statistical D
amage Index
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
New UCSD/FRA/BNSF Rail Defect FarmCamp Elliott Field Station
Univ. of California San Diego Constr. completed 2010
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
AREMA Committee 4 – Rail Meeting UCSD April 5, 2011
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Conclusions• UCSD system uses ultrasonic guided waves and statistical signal
processing to provide excellent detectability of rail head flaws.
• System proven on internal rail flaws (TDs, VSHs and Defective Field Welds ) during blind tests at Herzog in June 2010 (performance exceeded AREMA and industry standards).
• Potential also shown for characterization of rail surface conditions (density and depth of surface cracks) – of interest to grinding ops.
• Current prototype uses non-contact ultrasonic probing. The same approach (guided waves + statistical analysis) can be deployed using contact probing (e.g. ultrasonic wheels or sleds).
• Further development underway to increase inspection speed to > 5 mph and to add surface characterization to internal defect detection.
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Large-scale Rail NT/Buckling Testbed@ Powell Structural Laboratories
Sliding concrete block
Actuators for pretension
Fixed concrete block
Strong floor
Rail, tie, plates, sleepers donated by BNSF
Ballast
Heating rods (to achieve rail temperatures up to
150 F)
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
AREMA Committee 4 – Rail Meeting UCSD April 5, 2011
2011 ANNUAL CONFERENCESeptember 18-21, 2011 | Minneapolis, MN
Wayside Rail NT Measurement
Possible installation
Laboratory result
NT measurement accuracy = 3 F