Anda di halaman 1dari 8

Rail crack detection: an infrared approach to in-service track monitoring

Richard J. Greene, 1John R. Yates, 2Eann A. Patterson

Department of Mechanical Engineering, University of Sheffield, Mappin Street, Sheffield S1 3JD, United Kingdom
2

Department of Mechanical Engineering, Michigan State University, 2555 Engineering Building, East Lansing, MI 48824

ABSTRACT The growth of fatigue cracks in rail is a cause of ongoing concern to the railway industry, and much research and development work has been conducted to develop both trackside and train mounted analysis systems which are able to monitor the structural health of railway networks. To this end a new infrared-based method for surface crack quantification at conventional train speeds is being developed which delivers crack detection capability over the full depth of the rail section, with a crack resolution capability comparable to competing detection techniques. Experimental results are presented for a laboratory three-point bend notched specimen, the geometry of which is representative of surface-connected cracks with lengths below 2 mm in sections of rail. Two analyses are considered: a simulated trackside system, where the observation point is fixed and the repeated loading event experienced by a section of rail during the passage of multi-carriage rolling stock is considered; and a simulated train-based system, where the observation point moves along the rail and the loading event produced by the passage of a single train wheel is considered. Data from the trackside simulation system clearly identifies the precise location and severity of an artificially introduced notch on the upper surface of the specimen. Initial data from the train-based simulation system identifies the notch location precisely, but is unable to quantify the magnitude of the flaw using the current processing method. The paper then describes modifications to the testing and data processing methods required to improve the performance of both systems, and outlines the future on-site testing work planned to validate the methods presented.

INTRODUCTION Significant energy has been expended in recent years to develop sensitive and repeatable crack detection systems for monitoring the structural integrity of rail in service. Much of this effort has focused on the use of ultrasonic monitoring systems, both train-mounted and handheld, which are used to detect and precisely locate flaws respectively. Portable handheld ultrasonic inspection systems are used for two broad applications: to inspect the quality of welds between sections of rail shortly after the welds have been created, but not necessarily before the first rolling stock has used the track; and to provide a high quality flaw detection system to support train-based methods, which often lack precision in determining both the precise location and magnitude of the flaw [1]. This method is effective when there is no change in material properties through an analysis region, and observes the ultrasound wave reflected from discontinuities in the rail structure. However, in cases where there is a change of material properties, this approach is ineffective due to a reflection of ultrasound at the material interfaces. For example, this occurs in welds where a filler material is used which is different from the parent material of the two rail sections being joined. In these situations the complementary method of guided wave ultrasonic detection may be employed [2].

Train mounted ultrasonic inspection systems are most commonly designed to provide a large quantity of comparatively low precision data, allowing the structural integrity of many miles of track to be assessed during the normal operation of the railway. Ultrasound waves are introduced into the rail using a specially constructed train wheel which contains probes for both the emission and detection of waves, and as with the handheld systems reflections occur at discontinuities, enabling a database of possible flaws to be developed for a length of track [3]. This data is then used to seed the handheld inspection process, which subsequently provides the precision required for accurate flaw location and quantification. The principal weakness of this method, the need for a consistent and contaminant-free contact between the rail and the instrumented wheel, can be eliminated by using air-coupled ultrasonic methods which use the rail head as a resonant chamber [4]. However, the losses experienced at the air-rail boundaries are so significant that, despite the use of high energy ultrasonic excitation, the signal to noise ratio of the data generated using this approach is lower than that for the contacting method and there is a corresponding reduction in the maximum achievable train speed during analysis [5]. It is clear that the development of a train-based system which could provide greater precision, and a trackside based system which would require less operator skill, would be of considerable attraction. Ultrasonic techniques do not require a change in loading condition of the rail in order to detect flaws. In contrast, the non-contacting technique of thermoelastic stress analysis is typically performed using repeated loading of a structure whilst recording the observed surface temperature, most commonly using an infrared imaging device [6]. If a sufficient number of infrared data frames can be recorded whilst cyclic loading occurs, and if the cyclic change in surface stress is sufficiently large, for most materials a correlation can be made between the loading cycle and the observed surface temperature, yielding the cyclic change in surface stress condition. More precisely, commercially available thermoelastic stress analysis systems [7] output a voltage from the load vs. temperature change correlation. The magnitude of this voltage S is the thermoelastic response of the observed component and is related to the cyclic change in the sum of the principal surface stresses by

AS

(1)

where A is the thermoelastic calibration factor and 1, 2 are the principal surface stresses [8]. However, a repetitive loading event is not mandatory for a successful thermoelastic analysis to be performed. If a sufficiently large and rapid change in surface stress condition, and hence surface temperature, is experienced by a component, an analysis of a single infrared data frame can be sufficient to yield the stress distribution. Crack tip stress distributions determined using this technique can then be readily quantified to assess the severity of the crack and, from its location in the rail section, the likely remaining fatigue life [9]. The use of thermoelastic analysis in fracture mechanics and fatigue crack detection has previously been demonstrated both for long, propagating cracks [10], and also for small cracks, with the use of specialized infrared imaging equipment [11].

TRACKSIDE CRACK MONITORING Once a flaw has been identified using currently available detection methods, the precise location of the flaw is confirmed by local ultrasonic investigation and the flaw subsequently repaired. However, this leads to costly rework of the rail, and in many cases it is presumed that benign flaws are removed at significant cost. It is therefore of considerable interest to determine whether a given flaw represents a compromise to the structural integrity of the rail, or whether the flaw does not require urgent attention and can be periodically monitored instead. In order to make this decision, some form of trackside monitoring system would be required to quantify the severity of a flaw once identified. Due to the large bending stresses induced during the passage of a wheel over the rail, the four wheel loading events which will occur during the passage of a train carriage with a pair of dual axle bogies should prove to be sufficient to calculate a surface stress distribution with a sufficiently large signal to noise ratio to identify flaws in the rail section. The design and test methodology to simulate such a loading event in the laboratory is presented below.

Specimen design and defect preparation The objective of this research program is to analyze both geometrically simplified representations of rail section and also short lengths of rail. The testing of these 1 m lengths of rail is underway and initial results have been reported elsewhere [12], and so this paper will consider the results from the geometrically simple specimens. A rectangular cross-section prismatic beam of mild steel of length 205 mm, height 25 mm, and depth 15 mm has been used as a laboratory-based representation of a section of rail in bending. Although this geometry does not include the narrow shear web which forms the central section of rail, the overall bending stress distribution is similar, allowing valuable conclusions to be drawn from this geometry. The ratio of width to depth of this beam is similar to that of rail section, and the length of the beam is sufficient to simulate typical track support spacings. The beam was abraded with 300 and 600 grit wet and dry paper to yield a uniform surface on all sides, and a vertical notch of depth 1.7 mm then electrical discharge machined (EDM) in the center of the top surface, normal to the long axis of the beam. A layer of high carbon matt black paint of approximately 30 m thickness was then applied in order to achieve a high and uniform infrared emissivity over the entire specimen surface. The geometry of the notched beam appears as figure 1 below.
205

20

80

15

25

1.7

160

Figure 1 Dimensions of the notched beam used to represent the response of a section of rail in three-point bending. The positions of the three loading points are shown for the loading case where the notch in upper surface is 20 mm to the left of the upper loading point. The hatched area shows the analysis region presented in figures 2-4. All dimensions millimeters. Cyclic loading The beam was loaded in a three-point bending rig with the notch 20 mm to the left of the upper loading point, as shown in figure 1. The span between the two lower cylindrical loading rollers was 160 mm, and the upper loading roller was equidistant from the two lower supports. The upper roller was used to apply a sinusoidal compressive load of 0.5 to 3.5 kN at 14 Hz, giving a loading range of 3.0 kN. Since the peak stress peak developed on the upper and lower surface of an unnotched beam in three-point bending is described by

peak

3PL 2dh 2

(1)

where P is the centrally applied load, L the span of the two supports, d the depth and h the height of the beam, this cyclic load resulted in a cyclic stress range of 77 MN/m2 and a peak stress of 90 MN/m2. This leaves a comfortable margin below the yield stress of mild steel whilst achieving a reasonable dynamic stress range for thermoelastic analysis. Thermoelastic response under cyclic loading The response of the hatched section of beam identified in figure 1 is presented below. Figure 2 shows the thermoelastic response of an unnotched section of the beam, figure 2a, and the notched beam when the notch is on the upper surface of the beam and 20 mm to the left of the upper loading point, figure 2b. The thermoelastic response has been normalized by the peak response of the bottom surface of the beam.

a)

-1 b) Figure 2 Thermoelastic response of three-point bend specimen, showing: a) an un-notched section of the beam, with the expected distribution of compressive stress on the upper surface and tensile stress on the lower surface; b) a section of the beam with an artificially introduced notch of depth 1.7 mm. The data has been normalized by the peak tensile stress on the lower surface of the beam, and a 3x3 median filter applied.

-2

Figure 3 Differential crack detection image, showing the unfiltered thermoelastic response of the notch alone. The effect of the macro-scale bending stresses has been removed by subtraction of a reference image. The data is scaled using twice the peak differential thermoelastic response of the notch tip.

The data in figure 2 was recorded using a camera integration time of 1 ms, equivalent to a shutter speed of 1/1000 s, and has been filtered using a 3 by 3 pixel median filter to improve the signal to noise ratio. The median filter clarifies the stress distribution, but even without the application of this filter the location of the notch tip would be clearly visible in figure 2b. However, the automated detection and location of flaws using the data in figure 2b directly is not straightforward, since a peak detection algorithm would commonly identify the region directly beneath the upper loading point, and not the notch tip. In order to facilitate automated flaw detection, the macro-scale stress distribution associated with the contact stresses and the bending of the beam need to be removed. This is achieved by subtracting the data in figure 2a from figure 2b, the result of which appears as figure 3. Here the unfiltered result of data subtraction leaves a predominantly null field, except where the notch modifies the local stress distribution. It is then a simple task to interrogate this differential thermoelastic data to identify the notch location.

TRAIN MOUNTED FLAW DETECTION A different approach is required to achieve a successful method for detection of cracks in a rail using a trainmounted analysis system, since the repetitive loading approach used in the above trackside monitoring method is not possible. A single-pass train-mounted system, utilizing a single infrared camera system, will only have one opportunity to capture data from any given area of the rail surface, and as such cannot capture multiple image frames from the same section. The trackside repetitive loading approach is therefore inappropriate. Simulation of loading during single wheel passage In order to replicate the surface stress response of a section of rail during the passage of a single wheel along the rail, the same beam used in the trackside crack monitoring analysis and shown in figure 1 was cyclically loaded between 0.5 and 3.5 kN at 14 Hz as before, but infrared data was collected as a set of sequential data images at 15 frames per second for two seconds, again with a camera integration time of 1 ms. This stream of data was captured 1 Hz faster than the loading cycle in order to record the progression of changes in the surface stress distribution during loading, and adequately simulates the high speed capture of sequential data frames during a single wheel loading event. Thermoelastic response under simulated single wheel passage Although development of the test procedure is still underway, the results of initial single frame thermoelastic analyses of the beam are presented in figures 4-6 below.

9100

9500

Figure 4 Notch detection from a single infrared data frame captured during dynamic loading. Although the macro-scale bending stress distribution is difficult to discern, the upper contact point (chained box) and notch location (arrow) are clearly visible. The dotted line denotes the data points analyzed in figures 5 and 6 below. Data is shown in A/D units and has been corrected using a rudimentary calibration process to eliminate some of the variation in performance of individual detectors in the infrared camera, but is otherwise unprocessed.

Figure 4 presents the unfiltered infrared response at the maximum load of 3.5 kN, or a peak bending stress of 90 MN/m2. A simple calibration has been applied to account for the variation in response of individual detector elements in the infrared array, but no further processing has been performed. Although there is considerable systematic noise in this data, the large compressive stress distribution associated with the upper surface contact is clearly visible within the chained box. The macro scale bending stress distribution is also apparent, although less clear than in the cyclic loading data in figure 2. However, there is a discontinuity in this data field at the crack tip location, marked with an arrow in figure 4. The position of the notch has been verified by comparing the data with a subsequent cyclic thermoelastic test of the beam, conducted immediately after this test was complete. A line plot of data across the notch tip, the location of which is shown by the dotted line in figure 4, appears as figure 5. A linear trend line is also included to show the deviation of the data points from the macro scale distribution.
9270 9260 9250 A/D units 9240 9230 9220 9210 9200 9190 0 5 10 15 20 25 30 35

Infrared data point

Figure 5 Linear plot of infrared response in the notch region (the location of this data is shown in figure 4 as a dotted line) in uncalibrated A/D units. Data is presented for 36 individual data points, and reveals the location of the notch with a pair of outlying data points, one well above and one well below the trend line, marked as a solid line. The above data reveals the notch location at data points 19 and 20, where there is a pair of outlying data points. In order to quantify this excursion, calculation of the standard deviation of the data is required.

3 2 Standard deviations 1 0 0 -1 -2 -3 -4 Infrared data point 5 10 15 20 25 30 35

Figure 6 Standard deviations of the infrared data from figure 5. Whilst most of the data lies within 1.5 standard deviations, data points 19 and 20 denote the location of the notch with deviations of -3.6 and +2.6 respectively.

The standard deviation for the data subset shown in figure 5 is 9.7 A/D units. The individual deviations of each data point, expressed in terms of the number of standard deviations, are then presented in figure 6. It is clear from this figure that, although most data lies well within 1.5 standard deviations, the data at the crack location produces two outlying data points at -3.6 and +2.6 standard deviations respectively. This allows for the development of a reliable, automated flaw detection method which will be employed in future work.

DISCUSSION Considerable experimental challenges are expected in the next stage of the research work, which will begin with trackside analysis of known cracks in a short section of test track. The difficulties inherent in studying rail in situ, particularly the variation in surface condition of the rail and optical obscuration by the various fittings and fixtures used to hold the rail in place, are difficult to simulate in the laboratory. The influence of these variables will need to be assessed carefully during outdoor trials. The trackside analysis method is believed to be capable of yielding results in a comparatively short timeframe, with the use of existing hardware and carefully generated loading reference signals during the passage of a series of train wheels. On-site testing will enable the performance of the method to be determined and a bespoke analysis algorithm to be developed. The train-mounted analysis method will face limitations in the maximum achievable train speed at which data of sufficient signal to noise ratio can be recorded. If the entire face of the rail is to be studied during a single train journey, a sequence of infrared images must be recorded which overlap at both ends in order to capture a continuous map of the infrared response along the rail. Since the overall height of typical rail section is approximately 150 mm from the wheel contact surface to the bottom of the mounting flange, and assuming the observed infrared image has an equal number of pixels in the horizontal and vertical directions once the image overlap has been allowed for, this enables the infrared response of a 150 mm length of rail to be recorded in each data frame. The maximum train speed during analysis then becomes a function of the frame rate of the infrared camera. When operating at a frame rate of 105 frames per second, this would permit a train speed of approximately 16 m/s, or 57 km/hr, which is comparable to current ultrasonic train speeds. However, if a reduction in the spatial resolution of the infrared data image can be accommodated, the frame rate of the camera could be increased considerably, leading to a corresponding increase in maximum train speed. Using a 320 by 256 pixel infrared detector array, this would give a data image spatial resolution of 0.6 mm per pixel. In order to avoid horizontal blurring of data, the time over which each image is recorded must be sufficiently brief that no more than 0.6 mm of translation occurs during image capture. At 16 m/s, the train will travel 0.6 mm in 0.04 ms, which would result in a detector integration time 25 times shorter than the 1 ms integration time used in the train mounted simulation presented in this paper. Therefore the limiting train speed may be determined not by the camera frame rate but instead by the infrared detector integration time required to yield meaningful data images. Alternatively, a degree of horizontal image blurring may prove to be acceptable for the detection of typical rail flaws, allowing the train speed to be increased.

CONCLUSIONS The initial work presented here demonstrates the viability of employing thermoelastic analysis systems both for a trackside approach for the structural health monitoring of rail, and also for a train-based flaw detection system. The fundamental viability of both approaches has been demonstrated in the detection of a notch of length 1.7 mm, and it is believed that the lower bound to the detectable crack length will ultimately be below 1 mm, which is comparable to competing flaw detection methods currently in use by the rail industry. Whilst the trackside method appears to be readily achievable using existing technology, significant further development is required to develop a train-based system which is capable of real-time defect detection. Future on-site testing of pre-cracked rail will allow the techniques to be tested in a realistic track environment, where an assessment of the experimental difficulties of flaw detection in rails will be conducted.

ACKNOWLEDGEMENTS This work has been funded using a Higher Education Funding Council for England (HEFCE) grant, with guidance provided by Chris Kirby of Sheffield University Enterprises Ltd (SUEL). The support of David Thompson of Balfour Beatty Rail Technologies Ltd, the provision of welded rail sections by Balfour Beatty Rail, and the technical advice given by Archie Crawford of Network Rail, are also warmly acknowledged.

REFERENCES [1] Clark, R. and Singh, S., The inspection of thermite welds in railroad rail - a perennial problem, Insight - NonDestructive Testing and Condition Monitoring 45(6) 387-393 (2003) [2] Wilcox, P., Evans M., Pavlakovic B., Alleyne D., Vine K., Cawley P. and Lowe M., Guided wave testing of rail, Insight - Non-Destructive Testing and Condition Monitoring 45(6) 413-420 (2003) [3] R Crocker, Ultrasonic vehicle-based rail inspection, Non-Destructive Testing and Condition Monitoring 46(6) 323-325 (2004) [4] McNamara, J. and Lanza di Scalea, F. Air coupled ultrasonic testing of railroad rails, Mat. Eval. 60(12) 1431-1437 (2002) [5] Lanza di Scalea, F., Rizzo, P., Coccia, S., Bartoli, I., Fateh, M., Viola, E. and Pascale, G., Non-contact ultrasonic inspection of rails and signal processing for automatic defect detection and classification, Insight NDT and condition monitoring, Special Issue on NDT of Rails 47(6) 346-353 (2005) [6] Dulieu-Barton, J.M. and Stanley, P., Development and applications of thermoelastic stress analysis, J. Strain Anal. for Eng. Design, 33(2) 93-104 (1998) [7] Lesniak, J. and Boyce, B.R., A high-speed differential thermographic camera, Proc. SEM Spring Conf. on Exptl Mech., Baltimore, 491-497 (1994) [8] Pitarresi G.. and Patterson E.A., A review of the general theory of thermoelastic stress analysis, J. Strain Anal. for Eng. Design, 38(5) 405-417 (2003) [9] Dulieu-Barton, J.M., Fulton, M.C. and Stanley, P., The analysis of thermoelastic isopachic data from cracktip stress fields, Fatigue Fract. Engng Mater. Struct., 23(4) 301-313 (2000) [10] Diaz, F.A., Patterson, E.A., Tomlinson, R.A. and Yates, J.R., Measuring stress intensity factors during fatigue crack growth using thermoelasticity, Fatigue Fract. Engng Mater. Struct. 27(7) 571-584 (2004) [11] Hartman, G.A., Infrared Damage Detection System (IDDS) for real-time, small-scale damage monitoring, Proc. SEM Ann. Conf. on Exptl Mech., Charlotte, North Carolina (2003) [12] Greene, R.J., Yates, J.R. and Patterson, E.A., In-service fatigue crack detection of rail using thermal techniques, Proc. 9th Int. Conf. on Fatigue, Atlanta (2006)

Anda mungkin juga menyukai