NON-CONTACT ULTRASONIC GUIDED-WAVE DEFECT …...bulk waves (longitudinal or shear), is guided waves....

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


Direct Damage Cost

# Derailments


23 % (1st leading

cause)


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


FRA Safety Statistics data

for period 1998‐2008

Within category: rail, joint bar

and rail anchoring – all US railroads


Direct Damage Cost

# Derailments


23 % (1st leading

cause)


Detail Fracture12 %

(2nd leading cause)

$ 137 M(2nd highest

cost)427


Direct Damage Cost

# Derailments


23 % (1st leading

cause)


Detail Fracture12 %

(2nd leading cause)

$ 137 M(2nd highest

cost)427

Research Motivation


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)


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


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)


FEA simulation of TD detection

Finite Element Simulations of Ultrasonic Guided Waves: example of TD detection


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


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


Field Tests at Herzog, June 2010


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


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


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


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


New UCSD/FRA/BNSF Rail Defect FarmCamp Elliott Field Station

Univ. of California San Diego Constr. completed 2010


AREMA Committee 4 – Rail Meeting UCSD April 5, 2011


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.


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)


AREMA Committee 4 – Rail Meeting UCSD April 5, 2011


Wayside Rail NT Measurement

Possible installation

Laboratory result

NT measurement accuracy = 3 F

Date post:	16-Apr-2020
Category:	Documents
Upload:	others
View:	12 times
Download:	2 times

NON-CONTACT ULTRASONIC GUIDED-WAVE DEFECT …...bulk waves (longitudinal or shear), is guided waves....

Documents