ULTRASONIC TOMOGRAPHY FOR RAIL FLAW IMAGING

Robert Phillips

Avanti Tech LLC.

11211 Sorrento Valley Road Suite M

San Diego, CA 92121

Tel: (619)247-4588; rphillips@avantitechllc.com

Francesco Lanza di Scalea*

University of California, San Diego

9500 Gilman Drive MC0085

La Jolla, CA 92093

Tel: (858)822-1458: Fax: (858)534-6373; flanza@ucsd.edu

Claudio Nucera

University of California, San Diego

9500 Gilman Drive MC 0085

La Jolla, CA, 92093

Tel: (858)534-5279; Fax: (858)534-6373: cnucera@ucsd.edu

Piervincenzo Rizzo

University of Pittsburgh

3700 O’Hara Street

Pittsburgh, PA 15261

Tel: (412)624-9575: Fax: (412)624-0135; pir3@pitt.edu

Leith Al-Nazer

Office of Research and Development

Federal Railroad Administration

1200 New Jersey Ave, SE

Washington, DC 20590

Tel: (202)493-6128 Fax: (202)493-6333 leith.al-nazer@dot.gov

© AREMA 2014 1

Number of words: 5618

ABSTRACT

In the railroad industry there is a need to have quantitative information on internal rail flaws, including flaw size and orientation. Having such information can ultimately lead to an increase the safety of train operations by preventing derailments using a knowledge-based decision of remedial action to be undertaken on the rail flaw. Current ultrasonic inspection methods leave such determinations to a rather subjective interpretation by the inspector which is influenced by experience and other factors. However, this quantitative information can be obtained by 3-D imaging of the rail flaws. This paper will present a method for 3-D imaging of internal rail flaws based on Ultrasonic Tomography. Combining tomographic elements from radar and sonar imaging, high resolution images of flaws can be obtained with a stationary array of ultrasonic transducers. Operated in a “full matrix capture” scheme that minimizes the number of ultrasonic transmitters, the practical implementation is simplified and the inspection time is reduced with such an array. In this method, a full 3D image of the rail volume identifies the location, size and orientation of the defect. This will help to eliminate human error involved with the typical manual, single transducer inspection methods currently used. The results of advanced numerical simulations, carried out on rail profiles with an internal defect will be presented. The simulations show the effectiveness of the technique to image a 5% Head Area Transverse Defect in the rail head. Current efforts are aimed at developing an experimental prototype based on this technology, whose design is in process.

INTRODUCTION

Safety Statistics Data from the US Federal Railroad Administration indicate that train accidents caused by track failures (including rail, joint bars and rail anchoring) resulted in 1,235 derailments and $321M in associated damage costs during the a five year period (Dec 2006- Oct 2011) (1). Among the most severe internal rail flaws are Transverse/Compound Fissures (TF) and Detail Fractures (DF) that propagate predominantly in a direction transverse to the rail running direction (Transverse Defects – TD’s).

To ensure proper rail maintenance and safety of transportation, ultrasonic rail inspections are conducted on schedules by time and gross tonnage as per FRA regulations. Currently in the U.S. the “stop and confirm” mode of rail inspection requires verifying any defect indication found by an inspection car. Such manual verification is usually performed with 70-deg ultrasonic transducers (to maximize the sensitivity to critical Transverse Defects) typically operated at frequencies of 2.25 MHz (2). However, both car inspections and manual inspections cannot currently provide accurate and repeatable information on the size, shape and orientation of a defect that has been detected. Rather, the defect size is estimated from manual scanning of the ultrasonic transducers in a subjective manner based on operator’s judgment and level of experience.

© AREMA 2014 2

The railroad industry would benefit from a system that can provide 3-D imaging of rail defects in an automated manner, to obtain more accurate information on flaw size, shape and orientation. Such a system would increase safety with a more accurate rail flaw assessment. Accurate and repeatable sizing of internal rail defects could provide for increased flexibility in the future as it may be possible to tailor maintenance mandates to the actual size of defects. Doing so could potentially improve the overall safety and efficiency of the rail transportation industry. This improvement would be particularly beneficial to high-speed rail that will require well-targeted maintenance actions within increasingly shorter maintenance windows.

TECHNOLOGY ASSESSMENT

Ultrasonic Tomography represents the “next level” in imaging performance from Ultrasonic Phased Array technology. Ultrasonic Tomography represents “the next level” in imaging performance from Ultrasonic Phased Array technology. Phased Arrays (PA), which provide steering and focusing capabilities to the ultrasonic beam, have been used successfully in the railroad industry to inspect hard-to-access locations of axles (e.g. wheel seats, brake disk seats and gear seats) (3) and to enhance the detectability of internal rail flaws (4,5). The Transportation Technology Center (TTC) in Pueblo, CO has also worked on rail flaw characterization by Ultrasonic Phased Arrays (6).

Figure 1. (a) Phased-Array Imaging. (b) Tomographic Imaging with Synthetic Aperture Focus.

Figure 1(a) depicts how all the transducer elements in a Phased Array (PA) transmit the ultrasonic waves with focusing achieved with time delays of the excitation to the individual elements. A high cost is paid in Phase Array imaging with need to have many parallel transmitter and receiver channels. As an example in medical imaging a PA with 128 transducer elements requires 128 independent transmit and receive circuits. With a high channel count necessary to have a high defect imaging resolution PA imaging is

All  elements transmit withtime delays

A few elements transmit with diverging beams 

Phased‐Array (PA) ImagingTomography with Synthetic Aperture Focus (SAF) Imaging

(a) (b)

© AREMA 2014 3

not practical in meeting the power and size requirement for a hand-held system that is practical for field use on rails. In the case of 2-D arrays that are necessary to provide 3-D imaging without scanning manually the transducers, as desired by rail flaw inspection, PA imaging becomes challenging.

Alternatively, Ultrasonic Tomography with Synthetic Aperture Focusing (SAF) (7, 8, 9, and 10) represents an evolution from PA imaging, in that only a few transmitters can be used at once – Figure 1(b). As opposed to the PA method of focusing the ultrasonic wave into a beam to target select points within a volume, SAF transmits a beam that is made to be as divergent as possible to span the inspection volume. By using the divergent beam the number of active channels is greatly reduced, particularly the transmission channels. The reduction of complexity in the electronics can now multiplex a limited number of transducers at once, placing Ultrasonic Tomography with SAF at the forefront of recent Sonar, Radar and Medical Imaging. In the SAF approach, focusing is achieved in both transmit and receive with the minimum electronic complexity. Hence, a SAF-based Tomographic Imaging approach seems ideal for a hand-held rail inspection system in the field.

SYNTHETIC APERTURE FOCUSING

Figure2. Principle of Synthetic Aperture Focusing (SAF) for Ultrasonic Tomography 3D imaging.

The SAF method is most efficiently implemented using a delay-and-sum algorithm

(9, 10, and 11). Figure 2 displays the broad idea of the SAF applied in a 3D schematization. The algorithm is based on the time required for an ultrasonic wave to propagate from transmitter i to a receiver j after reflection from a focus point P, in the imaging volume. The delay-and-sum algorithm used for the SAF method can be expressed by referring to figure 2 and using to the following equation:

© AREMA 2014 4

1 1

, , ,N M

pj i

I y t i j i j

Pr (1)

In equation 1, I(rp) is the imaging intensity at focal point P(x,y,z) of the volume (associated with the 3D position vector rp), N is the total number of j receivers, M is the number of i transmitters, and the function y[tp(i,j),i,j] is the ultrasonic waveform generated by the i-th transmitter and detected by the j-th receiver. The ultrasonic waveform y[tp(i,j),i,j] is calculated at the time delay tp(i,j), that is in turn calculated from the wave travel time from transmitter i, to focal point P as follows:

, ,

,pL S L S

i jt i j

v v

P T P Rr r r r

(2)

For equation 2, rP is the 3D position vector of the focal point P, rT(i) is 3D position vector of the i-th transmitter, rR(j) is the 3D position vector of the j-th receiver, and νL,S represents the longitudinal wave velocity or the shear wave velocity in the material. Using either the longitudinal or the shear wave velocities is referred to as “Multimode Detection”.

Equations 1 and 2 “scan” all the focus points P(x,y,z) of the 3D imaging volume and attribute to each focus point an “imaging” value I(x,y,z) that corresponds to the ultrasonic intensity that is transmitted by all M transmitters, reflected by point P, and received by all N receivers. This is known as “total focusing” (i.e. focusing in both transmission and reception). More importantly, the ultrasonic tomographic approach minimizes the number of transmitters that are need to be excited at once and reducing the complexity of the electronics.

Owning to the focusing process in both transmission and reception, the Signal-to-Noise Ratio (SNR) of the image expected from the ultrasonic tomographic system with M transmitters and N receivers can be written as (12)

SNR N M (3)

It is then expected that the SNR of the resulting image will improve with the increasing number of transmitters and receivers. In practice, the hardware complexity, cost, and scan time limit the number of transmitters. Thus, here a spare array of transmitters will be used with as few as five combinations with a finer array of receivers. This approach involving the spare transmitters and fine receivers has been considered in medical imaging with the SAF technique.

MATCHED FILTERING

Used in radar imaging to compress and denoise the radar reflections from the targets, matched filtering dramatically enhances the SNR and image resolution (13). Applying matched filtering to the general SAF approach became a key step in this present study. By performing a cross-correlation between each raw waveform received by the ultrasonic transducers and the know excitation sent by the ultrasonic transmitter, the

© AREMA 2014 5

matched filter is applied. The matched filtered signal is represented by the following equation:

*z t f y t d

(4)

Such that f* is the complex conjugate of the known excitation signal sent to the transmitter (or, alternatively, the signal from an “ideal” reflector) and y(t) is the received waveform.

Figure 3. Example of Matched Filtering with the top Waveform a two-cycle toneburst at 2 MHz, the center waveform a detected signal with 10 dB of white noise added, and the

bottom signal the resulting waveform after matched filtering the toneburst and the detected signals.

An example of the application of matched filtering is shown in figure 3. An excitation

waveform of two-cycle toneburst is seen at the top of the figure. The center signal is a simulated received signal with the addition of 10 dB of white noise. The result of applying the matched filtered method between the noisy signal and the excitation signal is represented in the bottom signal. In the result, it is clearly seen that the arrival of the signal is easily determined after applying the filter.

HILBERT TRANSFORM

In rail flaw detection, there is the possibility that multimodes exist as rail steel allows the two fundamental wave modes to propagate. The two modes that propagate are the longitudinal wave mode and the shear wave modes. Each of these modes travels at a different speed, typically vL=5.3 mm/µsec (0.21 in/µsec) for the longitudinal mode and vS=2.8 mm/µsec (0.11 in/µsec) for the shear mode. Using a longitudinal excitation, the

© AREMA 2014 6

reflection from a defect will generate both a longitudinal wave mode (same mode reflection) and a shear wave mode (mode conversion) that will be seen by the receivers. With the expectation of using multimode tracking, a method is needed to identify the distinct modes within a received waveform. Having the existence of two different modes allows for the use of different time delays in the delay-and-sum equations (1) and (2) and can enhance the SNR of the resulting image.

Based on time delays the Hilbert Transform allows for the identification of the individual modes. By taken the absolute values of the waveforms of the Hilbert Transform the individual wave modes are detected in an envelope. Thus, each mode is identified based on the speed and expected time delay at each point in the scanned volume. The following equation shows the analytical form of the Hilbert Transform ŷ(t) of a signal y(t):

d

t

tyty

1ˆ (5)

Figure 4 shows the Hilbert Transform envelope detection applied to an ultrasonic recorded in the rail of a 3D FEA simulation discussed later. Clearly, the Hilbert Transform captures the signal strength or amplitude at the appropriate delays at the point in the volume that can then be summed.

Figure 4. Example of envelope detection (Hilbert Transform) of an ultrasonic waveform

recorded from a 3D rail imaging FEA model.

BASELINE SUBTRACTION

Baseline subtraction is a key feature in the implementation of the 3D rail flaw imaging approach. In this step, the ultrasonic waveforms are collected from pristine defect free

© AREMA 2014 7

rail volume and used as a baseline. Using the baseline subtraction eliminates any contributions of waves that have not resulted from the direct reflection of a flaw in the railhead. By applying the baseline subtraction, reflections from the rail boundaries will be eliminated and the SNR of the defect image will be radically enhanced. In theory, a database of baselines can be developed by only having to scan a pristine section of rail for a selection of geometries such as 136 RE and 141 RE rail sizes. If the rail shows heavy wear, then a baseline can be taken from a section known to be free of defects and used on the section with the defect volume in question.

Baseline subtraction is executed in two different modalities. The first modality uses an envelope subtraction performed on the enveloped signals generated by the Hilbert Transform. In the second modality, the subtraction is performed on the resulting images. The application of both modalities greatly enhances the image resolution with the removal of spurious waveforms that are not generated by the direct reflection from the internal flaw in question.

FINITE ELEMENT MODEL FOR 3D RAIL FLAW IMAGING

Using ABAQUS a 3D FEA model was developed in order to evaluate the mathematical method for the SAF approach discussed in the pervious section. The model is a 175.2 mm (6.9 in) long section of 136 RE rail sectioned at the web height (Figure 5).

Figure 5. Rail section model for 3D FEA

The flaw in the model is a circular transverse defect extending for 5% of the rail head area and the location geometry is shown in figure 6.

© AREMA 2014 8

Figure 6. Geometric dimensions for 3D FEA model

For the FE mesh, 8-node hexahedral elements with “reduced integration” (ABAQUS

element HEXA C3D8R) were used with a characteristic size of 0.5mm (0.02 in). At the boundaries of the model, 8-node “infinite elements” (ABQUS element HEXA CIN3D8) were applied as it was important to use the infinite elements to avoid spurious reflection of the ultrasonic waves that would not exists is an actual inspection field. Figures 7 and 8 show the resulting mesh.

The parameters for the FE model were as follow:

• Total number of nodes: 3,948,855

• Total number of elements: 4,023,918

• Element type of model: linear hexahedral elements type HEXA C3D8R

• Element type on boundaries: “infinite” elements type HEXA CIN3D8

© AREMA 2014 9

Figure 7. Mesh for the 3D FEA with “infinite elements” at the boundaries to eliminate

artificial boundary reflections of the waves.

Figure 8 Sectional view of 3D FEA mesh

A planar transducer array with 525 elements at a spacing of 2 mm (0.08 in) was

implemented in the simulated model for the 3D imaging. Figure 9 shows the array and defect simulated in the FEA model.

© AREMA 2014 10

With the future prototype design in mind, a sparse transmitter scheme was evaluated in the model. Accordingly, only five transmitters (M=5) were used, see Figure 10. By minimizing the number of transmitters, the complexity, cost and scan time of the imaging system is effectively reduced. Two cases were considered, namely: one case with all the elements receiving (N = 525), and the other case with only 140 elements receiving (N = 140). This was done to fully evaluate the effect of reducing the number of receivers on the final image resolution.

Figure 9. FEA model view showing transducer array and defect

A one-cycle tone burst of 2 MHz force inclined 70 degrees from vertical was the excitation signal sent to the five transmitter positions. The 70 degree inclination angle of the force is consistent with the 70 degree wedge used in conventional handheld ultrasonic rail defect detection units used in hand verification of Transverse Head defects in the rail head.

The first simulation was run to acquire waveforms for a pristine rail section for the baseline subtraction for both the full array (525 elements) and sparse array (140 elements) of receivers. Following the baseline acquisition, two additional simulations were run (for the full array and for the reduced array) to collect the waveforms with the defect in the volume.

© AREMA 2014 11

Figure 10. Position of the five spare transmitters in the transducer array

FULL ARRAY (525 RECEIVER ELEMENTS) IMAGE RESULTS

In this section, the 3D images generated from the reconstruction algorithm on the waveforms collected in the FEA simulation using five transmitter elements and 525 receiver elements are presented. Figures 11 and 12 show two views of the results using a longitudinal mode for the transmitter to defect path and a shear mode for the path from the defect to the receiver. The receiver signals were matched filtered and baseline subtraction was applied in the production of the image. Clearly, the defect is seen with no extraneous artifacts, image noise, or speckle noise producing an image with excellent contrast.

© AREMA 2014 12

Figure 11. 3D image of 5% Rail Head Area defect using the Longitudinal Wave mode

and the Shear Wave mode with full array (5 transmitters and 525 receivers)

Figure 12. 3D image of 5% Rail Head Area defect using the Longitudinal Wave mode and the Shear Wave mode with full array (5 transmitters and 525 receivers)

Due to the use of an estimated shear wave velocity; there are slight inaccuracies in the defect position. The horizontal position (image z direction) is underestimated and the vertical position (image y direction) is overestimated. However, the defect shape and, of the most importance, the defect size are estimated very accurately. The actual

© AREMA 2014 13

defect size is 141 mm2 (0.22 in2) and the longitudinal wave and shear wave tomographic image generated a defect of 150 mm2 (0.23 in2), giving a difference of only 0.3% of the rail H.A.

In addition to the longitudinal wave and shear wave mode combination, results from the longitudinal wave to longitudinal wave combination are presented. Here the velocities of the wave path from the transmitter to the defect and from the defect to the receiver are both the longitudinal wave speed. Again, matched filtering and baseline subtraction were applied to the receiver signals in the generation of the image. The resulting images are shown in Figures 13 and 14.

Figure 13. 3D image of 5% Rail Head Area defect using Longitudinal Wave modes with full array (5 transmitters and 525 receivers)

© AREMA 2014 14

Figure 14. 3D image of 5% Rail Head Area defect using Longitudinal Wave modes with full array (5 transmitters and 525 receivers)

Again, in the image the defect is shown with excellent contrast with no extraneous artifacts, image noise, or uncorrelated speckle noise. Using the longitudinal wave modes has corrected the position inaccuracies seen in the previous results. However, there are some irregularities with some contour lobes that were not present in the longitudinal/ shear wave modes results. When the most important measure of defect size is evaluated, again the result is very accurate. In the reconstructed image, the defect size is 154 mm2 (0.24 in2) compared to the actual size of 141 mm2 (0.22 in2), resulting in only a 0.4% difference of rail H.A.

REDUCED ARRAY (140 RECEIVER ELEMENTS) IMAGE RESULTS

In the interest of reducing complexity of a practical system, the effect of reducing the number of receiver elements was investigated. The image results of the FEA model using five transmitters and 140 receivers are presented in this section.

/

Figure 15. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave mode and the Shear Wave Mode.

© AREMA 2014 15

Figure 16. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave mode and the Shear Wave

As in the previous section, the longitudinal wave mode and shear wave mode combination (longitudinal wave speed for the transmitter to defect path and shear wave speed for the defect to receiver path) is evaluated first. The resulting images are shown in Figures 15 and 16. The same methodology was used such that the receiver signals were matched filter and baseline subtraction was applied.

The results from the reduced receiver array are very promising. While there is some drop in the image resolution, the position and sizing accuracy is well maintained. In the image result, the size of the defect is 148 mm2 (0.23 in2) in comparison to the actual 141 mm2 (0.22 in2). This is a difference of only 0.25% of the rail H.A.

For completeness, the longitudinal wave modes combination (using the longitudinal wave velocity for both transmitter to defect and defect to receiver wave paths) is also presented. The results are shown in figures 17 and 18.

© AREMA 2014 16

Figure 17. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave

modes.

Figure 18. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave modes.

Here the results are similar to those seen in the full array images. This confirms that it is possible to reduce the total number of receivers as a cost benefit. Now the detected defect size is basically the same at 154 mm2 (0.24 in2) and, compared to the actual 141 mm2 (0.22 in2), it results in a difference of 0.4% of the rail H.A. Based on these simulation results, the prospect for this method to be used in a practical application is promising. With this in mind, the development of a prototype is the next step to move the method onto the rail industry.

EXPERIMENTAL PROTOTYPE DEVELOPMENT

For the generation of the 3D images of flaws without any transducer movement, a 2D matrix transducer is required rather than the conventional 1D linear array. Using a static

© AREMA 2014 17

1-D array will collect data that only represents the planer section normal to the point of the array’s placement. In order to create the 3-D image with a linear array it is required that the array collect data perpendicular to the array’s transducer alignment to collect the third dimension of the test specimen. By implementing the sparse transmitter array, only a few transducers will be used for transmission and the remaining transducers will be used for reception.

Resolving small flaw features and meeting an accuracy goal of 20% of the actual flaw size and an accuracy of 20% degrees of the actual flaw orientation is achievable with high frequency ultrasonic waves. The results of the simulations showed that an ultrasonic frequency of 2.25 MHz will be sufficient to achieve the desired accuracy. The 2.25 MHz frequency is also used in conventional ultrasonic inspection of rails, and hence this specification should aid in the transition of the prototype to industry.

Figure 19. The 16 x 16 2D ultrasonic transducer array

The final array chosen (Fig. 19) is a 16 x 16 matrix array with a central frequency of 2.25 MHz. Each element within the array is small at 1.4mm, and therefore will generated signals that have a very wide generation field. This wide generation field is expected to allow for the elimination of the traditional 70-degree wedge used in the conventional method to target the most common transverse defect orientation in rails of 20 degrees. Eliminating the wedge will greatly simplify the reconstruction algorithm, since a wedge in the beam path will complicate the geometry and therefore the time delay calculation of the ultrasonic wave path. Nevertheless, the insertion of a wedge is being considered as an option in the image reconstruction algorithm, should this option become necessary.

The array is manufactured with piezo-composite technology transducers that optimize the ultrasonic energy transmission through minimum acoustic impedance mismatches at the element interfaces.

© AREMA 2014 18

For the demonstration prototype to be developed within this project, the 16 x 16 2D ultrasonic transducer array needs to be integrated with a data acquisition system. The system will be required to generate ultrasonic signals in selected array elements and receive the response signals from a selection of the remaining elements in order to collect the data needed for the computation of the image reconstruction algorithm.

A data acquisition system that can acquire over 500 channels simultaneously is just not economically practical at this time. Thus, the system will need to have the ability to excite the transmitter multiple times, receive a discrete number of channels, and then switch to another element for transmission and repeat the acquisition of the remaining channels until all the desired excitations are completed. The switching can be accomplished using an appropriate multiplexer and matrix switch combination. A multiplexer will be needed for the signal transmitter, since the switching speed is not critical and the multiplexer can handle higher voltages with mechanical switches. The need for the higher voltage capability is yet to be determined, but having the option for the insertion of a high-powered amplifier to create the excitation signals to the transmitters is the reasoning to use the high voltage capabilities of the multiplexer. Due to the high channel count of the reception, a matrix switch is needed. Matrix switches are fast and have high channel count through the use of solid state technology.

Perhaps the most important consideration is the sampling rate of the function generator for excitation and of the digitizer for reception. With a planned central frequency of 2.25MHz and bandwidth between 1.5MHz to 3.0MHz, the sampling rate is very high for standard digital components available. According to the Nyquist Theorem, the sampling rate of a digital signal must be a minimum of twice the desired frequency to eliminate aliasing. In practice, it best to have a sampling rate that is 10 times the maximum frequency expected to be generated or received to insure that aliasing is prevented. Thus, for the demonstration prototype both the function generator and the digitizer are required to have sampling rates of 30MHz or higher.

The hardware for the data acquisition is configured on a PXI platform. Using a PXI platform allows for the use of off the shelf components thus reducing the time constraints of this current project. Furthermore, the PXI chassis has a high speed broad bandwidth of up to 4GB/s bussing configuration that will exceed the requirements for the generation of signals and the acquisition of the signal response.

Using off the shelf components also makes the prototype economically feasible. A 100 MS/s arbitrary function generator was chosen to allow for signal generation well within the desired excitation frequencies as well as allow for the signal to be configured as preferred. The excitation signal is then sent to a high voltage switch that allows for the selection of the chosen transducers to be used as transmitters. While the majority of the components are available off the shelf, a custom switching card was required, and has been fabricated, in order to handle the large channel count of the receivers. The custom switch terminal block is a 512-crosspoint matrix, which then sends the received signal to the digitizer. For the digitizer, a 32-channel Field Programmable Gated Array card was chosen with a maximum sampling rate of 50MS/s. The hardware components are shown in figure 20.

© AREMA 2014 19

Figure 20. Data Acquisition System Hardware.

In order to expedite the development of the prototype, two software platforms will be used. Each platform will have separate tasks, one for the data acquisition and array control and the other one for the image reconstruction. This is to allow for an efficient method for development.

For the data acquisition component, the National Instruments LabVIEW© software platform will be used. This platform is a Graphical User Interface (GUI) software that is used for a broad range of data acquisition applications. Under this platform, the user will be able to control the waveform for excitation to the transmitter, select the transmitter or group of transmitters, and acquire the receiver waveforms in an appropriate format to interface with the image reconstruction algorithm software.

The image reconstruction algorithm for this study is being been written in the Matlab© software platform, to take full advantage of the powerful data processing algorithms and functions.

CONCLUSION

The ultrasonic tomography method shows great promise for the imaging of defects in rails. Under the current methods of rail flaw detection, it is quite difficult to quantify the defect size and orientation, and thus defect severity is generally estimated. Because of this, having the ability to determine if a defect is at a critical orientation or it has reached a critical size is not possible. In applying the imaging algorithm to the waveform signals generated by the FEA simulations, the results showed accurate sizing of a 5% H.A. transverse defect in the head of a 136RE rail. The defect position and orientation were

Array Connectors

Multiplexer Connector

Matrix Connectors

Digitizer Connectors

Electrical Connector Box PXI Platform

32 Channel  FPGA Digitizer

512 Point Matrix

12 x1  Multiplexer

Arbitrary Waveform Generator

12 x1  Multiplexer

© AREMA 2014 20

displayed in the reconstructed image with great accuracy. This is a substantial advancement to the current defect detection technology, in which defect size and orientation is based on the inspector’s subjective determination.

In light of the promise shown by the simulations, the development of a practical prototype was undertaken. Once operational, the prototype will test the ultrasonic tomography method on real world rail defects. Testing real defects will surely require more development of the algorithm for image reconstruction. It is the goal of this project to develop a portable system that will improve defect classification in rails and ultimately improve public safety.

ACKNOWLEDGMENTS

This work was supported by U.S. Federal Railroad Administration research and development funds through the High-Speed Rail BAA, with Leith Al-Nazer as the FRA Program Manager. In addition, Cyth Systems of San Diego is acknowledged for help in the development of the data acquisition system.

REFERENCES

1) Federal Railroad Administration. Safety Statistics Data. Train Accident by Cause:

2006-2011. FRA, U.S. Department of Transportation, 2012. 2) Lanza di Scalea, 2007 “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.

3) Hansen and Hintze, “Ultrasonic testing of railway axles with phased array technique,” 8th European Conference on Nondestructive Testing, Barcelona, June 2002.

4) Wooh and Wang, “Nondestructive characterization of defect using a novel hybrid ultrasonic array sensor,” NDT&E International, Vol. 35, 2002.

5) Tscharntke, Heckel, Boehm, Erhard, Krull, Lutzner, and Hintze, “Development of ultrasonic railway inspection systems using phased array methods,” Railway Engineering International Conference, 2003.

6) Garcia and Zhang, Application of Ultrasonic Phased Arrays for Rail Flaw Inspection, Report DOT/FRA/ORD-06, July 2006

7) Karaman, Li, and O’Donnell, “Synthetic aperture imaging for small scale systems,” IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 42, pp. 429-442, 1995.

8) Jensen, Ivanov, Gammelmark, and Pedersen, “Synthetic aperture ultrasound imaging,” Ultrasonics, Vol. 44, pp. e5-e15, 2006.

9) Nikolov and Jensen, “Investigation of the feasibility of 3-D synthetic aperture imaging,” IEEE Ultrasonics Symposium, pp. 1903-1906, 2003.

© AREMA 2014 21

10) Daher and Yen, “2-D Array for 3-D ultrasound imaging using synthetic aperture techniques,” IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 53, pp. 912-924, 2006.

11) Lanza di Scalea et al.. Research and Development to Demonstrate Ultrasonic Tomography Technology for Three-dimensional Imaging of Internal Rail Flaws. Technical Report to the Federal Railroad Administration, April 2013, DOT/FRA/ORD-13/20.

12) Lookwood, G.R., Talman, J.R, and Brunke, S.S., “Real-time 3D ultrasonic imaging using spare synthetic aperture beam forming,” IEEE Transactions on Ultrasonic, Ferroelectrics and Frequency Control, Vol.45 pp980-988, 1998

13) Maitre, H., “Processing of Synthetic Aperture Radar Images”, Wiley and Sons, United Kingdom, 2008

FIGURES

Figure 1. (a) Phased-Array Imaging . (b) Tomographic Imaging with Synthetic Aperture Focus. Figure2. Principle of Synthetic Aperture Focusing (SAF) for Ultrasonic Tomography 3D imaging. Figure 3. Example of Matched Filtering with the top Waveform a two-cycle toneburst at 2 MHz, the center waveform a detected signal with 10 dB of white noise added, and the bottom signal the resulting waveform after matched filtering the toneburst and the detected signals. Figure 4. Example of envelope detection (Hilbert Transform) of an ultrasonic waveform recorded from a 3D rail imaging FEA model. Figure 5. Rail section model for 3D FEA

Figure 6. Geometric dimensions for 3D FEA model

Figure 7. Mesh for the 3D FEA with “infinite elements” at the boundaries to eliminate artificial boundary reflections of the waves. Figure 8. Sectional view of 3D FEA mesh Figure 9. FEA model view showing transducer array and defect Figure 10. Position of the five spare transmitters in the transducer array Figure 11. 3D image of 5% Rail Head Area defect using the Longitudinal Wave mode and the Shear Wave mode with full array (5 transmitters and 525 receivers) Figure 12. 3D image of 5% Rail Head Area defect using the Longitudinal Wave mode and the Shear Wave mode with full array (5 transmitters and 525 receivers)

Figure 13. 3D image of 5% Rail Head Area defect using Longitudinal Wave modes with

Figure 14. 3D image of 5% Rail Head Area defect using Longitudinal Wave modes with full array (5 transmitters and 525 receivers)

Figure 15. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave mode and the Shear Wave Mode.

© AREMA 2014 22

Figure 16. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave mode and the Shear Wave

Figure 17. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave modes.

Figure 18. 3D image of 5% Rail Head Area Transverse Defect using Longitudinal Wave modes.

Figure 19. The 16 x 16 2D ultrasonic transducer array

Figure 20. Data Acquisition System Hardware.

© AREMA 2014 23

Ultr

ason

ic T

omog

raph

y fo

r Rai

l Fla

w

Imag

ing

Robert Phillips, PhD

p,

Avanti Tech, LLC San

 Diego, CA

Fran

cesco Lan

za di Scalea, PhD Claudio Nucera, P

hD

UniversityofCalifornia

SanDiego

University of California, San

 Diego

Piervincenzo Rizzo, P

hD

University of Pittsburgh

LeithAl‐Nazer(FRA Program

 Man

ager)

Federal Railroad

 Administration

Funded

 by: FRA High‐Speed BAA 2010‐1 

Collaborators: C

ythSystem

s, BNSF Railway

© AREMA 2014 24

Motivation for project• Current rail flaw detection systems provide qualitative

indication of flaws, not quantitative information on flaw size.

• Stop-and-confirm ultrasonic hand verifications result in highly subjective flaw sizing.

• High-speed rail requires accurate flaw sizing for well-High speed rail requires accurate flaw sizing for welltargeted actions with short maintenance windows.

Technology evolutionUltrasonic Tomography already successfully used for 3-D imaging in other fields:• Medical imaging (ultrasonic waves, electromagnetic RF waves,

etc..)• Sonar imaging (acoustic waves)• Radar imaging (electromagnetic RF waves)• Civil structures (ultrasonic waves)

Ultrasonic Tomography with SAF1. Synthetic Aperture Focusing (from Medical,

Sonar and Radar Imaging).

2. Matched Filtering (from Radar imaging).

3. Baseline Subtraction (from Ultrasonic Testing).

4. Multi-mode Detection (from Ultrasonic Testing).

Step 1: Synthetic Aperture Focusx

z

P(x,y,z) (focus point)

Transmitter i Receiver jTransducer array

rP

rT (i)

rR (j)

(0,0,0)

Signal‐to‐Noise Ratio of Tomographic‐SAF image

y Imaging Volume

1 1

( ) ( , ), ,N M

Pj i

I y t i j i j

Pr, ,

( ) ( )( , ) T R

PL S L S

i jt i j

v v

P Pr r r rImage intensity 

at focus point P

# receivers # transmitters

Time delay Velocity of Longitudinal Wave or Shear Wave (multi‐mode detection)

SNR N M

Step 2: Matched FilteringCross‐correlation between detected ultrasonic signals and known reference signal

*( ) ( ) ( )z t y t df

Representative signal from the 3‐D rail FEA model: signal arrival efficiently cleaned‐up from noise.

Step 4: Multi-mode DetectionExploits existence of multiple wave modes that can propagate in rail steel (Longitudinal Modes and Shear 

Modes) for more accurate imaging results.

Longitudinal Waves (travelling faster) and Shear Waves (travelling slower)

Movie: Center excitation

© AREMA 2014 25

3-D Rail Flaw Imaging• Developed 3-D FEA model of 136-lb rail with 5%H.A. Transverse Defect

inclined at 20-deg from vertical direction (realistic orientation).• Implemented “Infinite Elements” at model boundaries to avoid artificial wave

reflections not existent in the field. Movie 3‐D imaging

3-D Rail Flaw Imaging: SAF array• Simulated planar array of 525 transducers at 2mm spacing for Tomographic SAF. • Transmitters: used “sparse” configuration (5) to minimize # excitation channels.• Receivers: considered “Full Array” (all 525) and “Reduced Array” (only140).

Results: 3-D Rail Flaw ImagingSparse Transmitter Array (5) - Full Receiver Array (525)

Longitudinal Mode and Shear Mode. Envelope baseline subtraction.

ACCURATE TRANSVERSE DEFECT SIZING

Actual defect size:5.0% H.A.

Defect size from imaging with full array:

5.3% H.A.

Results: 3-D Rail Flaw Imaging (New Rail Vs. Worn Rail)

Sparse Transmitter Array (5) with Full Receiver Array (525)Longitudinal Mode and Shear Mode. Envelope baseline subtraction.

(BNSF Tangent Track)

Prototype Development: Array Transducer

2‐128 Pin Connectors

• 16 by 16 element (256 total) square array• 1.4mm inter-element pitch • operating frequency: 2.25 MHz• piezo-composite technology

Prototype Development: Data Acquisition Hardware

Matrix Connectors

Digitizer Connectors

National Instruments PXI Platform Electrical Connector Box

32 Channel FPGA Digitizer

512 Point Matrix

12 x1 Multiplexer

Array Connectors

Multiplexer ConnectorArbitrary Waveform Generator

© AREMA 2014 26

Waveform Generator and MultiplexerArbitrary Waveform Generator Multiplexer

Chirp

ansm

itters

Toneb

urst

Pulse

Input

Output to Up to 12 Tra

Matrix to Digitizer

nput Channels 1‐16

nput Channels 17‐32

eceivers

hannels to Digitizer

512 Point Matrix 32‐Channel FPGA Digitizer

Digital Input TriggerFrom Waveform Generator

In In

Input of 256 re

Outputs 8 Banks of 32 Ch

Data Acquisition Software Interface Summary and Conclusions• Ultrasonic Tomography being developed for 3-D imaging of rail flaws.

• Simulations demonstrated successful 2-D and 3-D imaging of 5%H.A. TD in 136 RE rail with excellent defect sizing estimation.

• Simulations showed satisfactory 3-D flaw images with only 5 transmitters and 140 receivers without moving the array (further reductions possible)reductions possible).

• Development of Experimental Prototype under way.

• Demonstrations on flawed rail sections to follow in the summer/fall• Accuracy targets:

– flaw size estimation ±20% of actual flaw size– flaw orientation estimation (planar flaws) ±20deg of actual angle

EXTRAS

18

Phased-Array vs. Tomography with SAF

All  elements transmit withtime delays

A few elements transmit with diverging beams 

Phased‐Array (PA) ImagingTomography with Synthetic Aperture Focus (SAF) Imaging

Tomography with SAF allows to reduce the number of channels to multiplex, and it is therefore ideal for a portable, hand‐held system for field use

© AREMA 2014 27

Step 1: Synthetic Aperture Focus

Envelope detection (Hilbert Transform of ultrasonic signals)

1 ( )ˆ( )

y ty t d

t

Representative signal from the 3‐D rail FEA model: envelope captures well instantaneous strength of signal.

Step 3: Baseline SubtractionSubtract “pristine” volume from “damaged” volume

Option 1:”Envelope” SubtractionOption 2: “Image” Subtraction

Representative signal from the 3‐D rail FEA model:artifact signals (edge reflections, etc..) reduced, defect reflection increased.

Results: 2-D Rail Flaw Imaging

2-D imaging results w/25 dB art. noise added to synthetic signals

3-D Imaging Results /2Sparse Transmitter Array (5) - Reduced Receiver Array (140)

Longitudinal Mode and Shear Mode. Envelope baseline subtraction.

23

Little change in imaging results between 525 Receivers and 140 Receivers (greatly simplifies system!)

ACCURATE TRANSVERSE DEFECT SIZING

Actual defect size:5.0% H.A.

Defect size from imaging:5.25% H.A.

Prototype Development: Hardware Interface

National Instruments PXI Platform:

• 100 MS/s arbitrary function generator.

• High-voltage switch to route g gexc. signal to up to 12 transmitters.

• Custom 512-crosspoint matrix and 32-channel FPGA digitizer to acquire up to 256 receivers.

• LabVIEW GUI designed for full acquisition control.

Image Reconstruction Software

Initialization Module‐ Sampling frequency‐ Array geometry‐ Position of transmitters and receivers‐ L‐wave and S‐wave speed in test material‐ Size of imaging volume

Image reconstruction Module‐ Mean subtraction‐ Matched filter‐ Signal envelope‐ Image reconstruction‐ Baseline subtraction

Waveforms from array

Visualization Module‐ 3D volume visualization 

‐ Isosurface visualization‐ 2D slice visualization 

© AREMA 2014 28