Real-Time Train Wheel Condition Monitoring by Fiber Bragg Grating Sensors

Wheel defects on trains, such as flat wheels and out-of-roundness, inevitably jeopardize the safety of railway operations. Regular visual inspection and checking by experienced workers are the commonly adopted practice to identify wheel defects. However, the defects may not be spotted in time. Therefore, an automatic, remote-sensing, reliable, and accurate monitoring system for wheel condition is always desirable. The paper describes a real-time system to monitor wheel defects based on fiber Bragg grating sensors. Track strain response upon wheel-rail interaction is measured and processed to generate a condition index which directly reflects the wheel condition. This approach is verified by extensive field test, and the preliminary results show that this electromagnetic-immune system provides an effective alternative for wheel defects detection. The system significantly increases the efficiency of maintenance management and reduces the cost for defects detection, and more importantly, avoids derailment timely.


Introduction
Condition monitoring measures are crucial to ensure safe and cost-effective train operation in the railroad transportation industry. A well-designed monitoring system substantially reduces hardware maintenance cost and improves service quality and overall safety. Wheel condition monitoring is one of the critical features in railway condition monitoring system. Typical train wheel defects include flat wheels, outof-roundness (OOR), spalling, and shelling [1][2][3]. If wheel defects are not detected and rectified in time, the wheel may deteriorate rapidly and induce more frictional force that inflicts further defect on both wheels and rails.
Nowadays, railway operators usually detect wheel defects by visual checking with experienced workers on a regular basis. Passenger complaint or driver report of excessive vibration is another means of identifying wheel defects. Moreover, periodical scheduled wheel reprofiling according to engineering experiences without defects identification is also employed. These methods are useful in general but they do not guarantee in-time identification of wheel defects. The railway operators have to bear the high cost for regular wheel-checking and reprofiling, and the comfort of passengers is adversely affected due to high vibration. Worst still, there is always a higher risk of derailment. Thus, a realtime monitoring system which is reliable, safety-proven, and cost-effective for handling wheel defects problem is urgently required.
Many studies have been carried out to realize wheel defect detection and most of them are based on the analysis of wheel-rail interaction [2][3][4][5][6]. Attivissimo et al. used a laser diode and a CCD camera to measure wheel and rail head profiles and hence to evaluate the wheel-rail interaction quality [2]. The system is able to discover anomalies of wheel-rail contact, but not resolving the defects of wheel or rail. In addition, the laser source and camera require precision setup which is difficult for practical railroad application. To study possible causes and effects of OOR wheels, strain gauges and accelerometers were employed to measure the vertical wheel-rail contacting force and track response [3,4], a number of 99 selected wheels were tested over a 100,000 km traveling distance. Since the sensitivity of the system is relatively low, only the wheels with local defects of 0.5 m in length can be identified. Sensors based 2 International Journal of Distributed Sensor Networks on ultrasonic [5] and acoustic [7] techniques have also been employed to measure the wheel and rail conditions, but the performance of these sensors are easily compromised under electromagnetic interference (EMI) railroad environment. Besides, EMI free fiber optics sensor [8] has been used to study the wheel defect problem, speckle interferometry in multimode optical fiber was developed to detect flat wheels of train. However, other types of defects such as local spall or polygonal wheel cannot be retrieved.
In this paper, we propose a real-time wheel-defect detection system based on fiber Bragg grating (FBG) sensors. The sensors measure the rail strain response upon wheelrail interaction and the frequency component that solely reveals the quality of the interaction are extracted from the signal and processed in order to deduce the defects of passing wheels. One of the advantages of this sensor system is that both the sensors and connecting fibers installed at the railroad side are passive to EMI and they require no electric power until the head-end measurement equipment which could be tens of kilometers away from the measurement points. This feature is particularly favorable to the modern electrified railway system since the sensing network is immunized from EMI. In addition, the system allows inservice and real-time monitoring of wheel condition, which is attractive to the railway industry. Moreover, the system can also be integrated with other railroad FBG sensing applications [9][10][11][12][13] that have been reported previously.
The organization of this paper is as follows. The basic concept of FBG sensor and its characteristics are first reviewed, followed by the descriptions of the sensor packaging and field test; the measurement results and the algorithm to generate the condition index (CI) of wheel are presented; discussions and conclusions are also given at the subsequent sections.

FBG Sensor and Wavelength Interrogation System
FBG sensor is an in-fiber narrowband reflective optical filter resulted from periodical variation of refractive index inside the core of optical fiber. A conventional way to fabricate a FBG sensor is to illuminate a short section of optical fiber by ultraviolet laser source under a phase mask, the periodical pattern of the mask is photo-imprinted onto the optical fiber by modulating the refractive index of the fiber core. The Bragg wavelength λ B of FBG sensor written in single-mode optical fiber is given below [14]: where n is the effective refractive index of the fiber core and Λ is the period of the refractive index modulation. Figures 1(a) and 1(b) illustrate an FBG sensor and its spectral characteristics. When a broadband light source is passed onto a sensor, a narrowband optical spectrum centered at λ B is reflected while others pass through. Mechanical and thermal perturbations alter the modulation pitch as well as refractive index of the FBG and hence λ B . By measuring the wavelength change of the FBG sensor, these perturbations can be determined. Typical wavelength changes of FBG written in standard single-mode fiber at 1550 nm region due to mechanical strain and temperature variation are ∼1 pm/με and ∼11 pm/ • C, respectively. Because of its compact size, low optical loss, self-referencing, and wavelength multiplexing capability, FBG sensors have been widely applied in a wide range of condition monitoring applications [9,[15][16][17][18][19]]. An interrogator is employed to measure the wavelength of FBG sensor array. Figure 1(c) shows a simplified schematic of a typical FBG interrogator. To produce a wavelength tunable optical source with high output power for accurate wavelength measurement, scanning ring laser consists of an optical amplifier and a tunable narrowband Fabry-Perot filter in a ring cavity configuration is employed, the laser scans the FBG sensor array periodically and the signal reflected from the FBG sensors is directed to the photodetector through an optical circulator. The photodetector performs optical to electrical domain conversion, any peak in the converted electric signal reflects the existence of an FBG and the wavelength of the FBG can be retrieved by the timing of scanning signal. Commercially available interrogators usually have four measurement channels with a wavelength measuring range of ∼80 nm, depending on the spectral occupancy of the FBG sensors in the array. Up to 40 serially connected FBG sensors (assuming 1 nm measuring range with 0.5 nm guard-band at both longer and shorter wavelength sides) can be measured simultaneously at a maximum sampling rate of 2 kHz by a single measurement channel. In this study, a four-channel interrogator with a sampling rate of 1 kHz is used.

Sensor Packaging and Field Test
Phase masks and ultraviolet excimer laser are used to fabricate the 10 mm-long FBG sensors in this study. To improve the photosensitivity of the standard single-mode fiber for FBG inscription, the fiber is kept in high pressure hydrogen vessel for a week. After FBG sensor fabrication, the fiber is annealed in temperature chamber and the FBG sensor section is recoated with the acrylate material. A 1 mm thickness stainless steel sheet with a dimension of 50 mm (length) and 12 mm (width) is machined using photochemical etching technique to form the sensor package as shown in Figure 2(a). This sensor package is designed in such a way that it allows a good mechanical contact between the FBG sensor and the rail surface while providing sufficient protection to the fiber. A 10 mm (length) and 8 mm (width) rectangular-shape void at the center of the package is used to place the FBG sensor during assembling, four elliptical recessed pads with a depth of 0.7 mm are used for electric spot-welding to the rail during on-site sensor installation. Ushape recessed grooves with a width and depth of 1 mm and 0.5 mm, respectively, are designed for fiber alignment and attachment purposes.
The sensor package is tested substantially by a calibration platform shown in Figure 2(b), a stainless steel triangular beam with one end fixed and the other end coupled to a mechanical joint which converts the rotational motion generated by a motor to a bending movement of the beam.  Four sensors have been installed by spot-welding onto two pairs of rail tracks near the Ho Tung Lau (HTL) depot in the East Rail line. Train speed at the measurement point is 50 to 90 km/h with ballast track-form. The sensors are installed near the foot of the rail which is the most sensitive and practical position [13] for measuring the wheel-rail interaction. The rail has been slightly polished to provide a clean contact surface area for spot welding of the sensors. Arrangement of the sensor installation is shown in Figure 3,  The low-frequency component of the signal mainly comes from the axle load of the train as depicted by the strain peaks of Figure 4(a), those peaks also show the slightly heavy motor cars, and this signal is essential information for studying train axle counting system [13]. Figure 4(a) that indicate the measured strain responses of the first to third cars of the train accordingly. Because these results are measured by the same sensor within a short period of time, they are therefore virtually collected under the same rail condition and reflected only the difference in the condition of the wheels. Obviously, Figure 4(b) shows a smooth strain signal generated by the first car while the signal generated by the third car as shown in Figure 4(d) is the noisiest among the three. Table 1 shows the wheel condition information of the three cars which is provided by the MTR, the OOR measurement in this table is conducted by using contact gauge in maintenance depot. The OOR of the wheels of the third car is much higher than those of the other two ones, and there is some thread wear in the wheels of the second car. In other words, the wheel condition of the first car is the best and that of the third car is the worst. Imperfect wheels of the third car exert periodic impact force onto the track and hence induce the measured uneven strain impulse as depicted by the noisy signal of Figure 4(d). Therefore smoother strain signal represents better condition of the wheels and this is consistent with the studies described previously [3,6].

Figures 4(b) to 4(d) show the enlarged sections of
A condition index (CI) system has been developed to express the wheel condition quantitatively. This system processes the strain signal of the four wheels (left or right side) of a train car and generates a CI that essentially indicates the condition of the four wheels (if the condition of any wheel is not in good condition, the system can generate a CI to indicate the wheels of the train car need to be maintained, but cannot distinguish which wheel is not in good condition because the system processes the strain signal of the four wheels as a whole. The reason of processing four wheels as a whole is that railway operators normally do wheel reprofiling on a train car basis). The low-frequency component of the strain signal is first filtered out by using a low-pass filter with a cutoff frequency of 20 Hz, Figure 5(a) gives the low-pass filter result of the signal of Figure 4(d). This signal basically consists of the information of axle loading of the train induced by its weight. The train speed (v) can be deduced from the distance between two wheels and the time between the peaks of the signal. Figure 5(b) shows the high-pass-filtered signal which carries the wheel condition information, and the signal is then decomposed to different frequency components by fast Fourier transform (FFT), as shown in Figure 5(c). The distribution shows that the majority of the strain change is located in the frequency range of 20 to 200 Hz while others are mainly noise at the frequency above 200 Hz. Further experimental works has confirmed that higher strain change over this frequency range represents a worse wheel or rail conditions. Moreover  CI is therefore given by where ε is the average value of the strain changes in the frequency range of 20 to 200 Hz, A is a scaling factor requires for matching the CI value within a range of 0 to 10.
The FBG monitoring system with CI was tested continuously over ten months, and the CIs of the twenty-nine passenger trains running in East Rail have been collected.  460  440  420  400  380  360  340  320  300  280  260  240  220  200  180  160  140  120  100  80  60  40  20  0 Frequency (Hz)  the measurement point once. It can be noted that the CI gradually increases for the first two/three months during the test until the wheels are reprofiled by depot grinding machine, the CI is substantially reduced from an average value of ∼7 to less than ∼1. After wheel reprofiling, the CI increases slightly and deteriorates again after ∼4 months in operation. This can be explained by the fact that newly turned wheels produce very little uneven strain on the rail and once a defect formed on a single wheel, deterioration of this wheel is accelerated and this also affects the other wheels International Journal of Distributed Sensor Networks 7 of the same bogie and hence degrades the overall wheel condition of the train car measured by the sensor system. It is also interesting to note from the figures that under normal train operation condition, the CI takes about 12 months to rise from ∼1 of newly turned wheels to ∼7 of just before wheels reprofiling, this time frame matches very well with the regular wheels reprofiling schedule exercised by the railway operator.

Conclusions
A real-time train wheel condition monitoring system using FBG sensors to measure the track strain response upon wheel-rail interaction is proposed and developed. The principle of FBG sensor and its optical characteristics are briefly reviewed, together with the description of the sensor packaging and field test. An algorithm to generate the CI value for indicating the train wheel condition is described, followed by the field test results and discussions. The system has been proven to be able to continuously monitor the conditions of the wheel and identify wheel defects in time. The system has the advantage of providing real-time and in-service measurement of wheel condition. Moreover, FBG sensors used in this system are very low in optical loss and immunity from EMI is particular suitable for electrified railways with track spanning over tens or even hundreds of kilometers. Further test and verifications with wheel and rail condition data are still required to ensure the long-term reliability of the proposed system. To improve the measurement accuracy and overall integrity of the system, a more intelligent analysis technique based on the previous wheel and rail conditions information and the results of a measurement point with more FBG sensors installed is now under investigation. A more advanced FBG-based system which can differentiate the defects of wheel and rail, as well as their types and levels, is the next stage of the real-time railway condition monitoring development.