A detecting method for spherical fuel elements in pebble-bed HTGR using eddy current detection

A detecting method for spherical fuel elements in pebble-bed HTGR using eddy current detection

NDT&E International 79 (2016) 81–91 Contents lists available at ScienceDirect NDT&E International journal homepage: www.elsevier.com/locate/ndteint ...

4MB Sizes 0 Downloads 7 Views

NDT&E International 79 (2016) 81–91

Contents lists available at ScienceDirect

NDT&E International journal homepage: www.elsevier.com/locate/ndteint

A detecting method for spherical fuel elements in pebble-bed HTGR using eddy current detection Zandong Han a,b,n, Haipeng Zhou a,b, Haiquan Zhang c, Dong Du a,b a

Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China State Key Laboratory of Tribology, Tsinghua University, Beijing 100084, China c Institute of Nuclear and New Energy Technology, Tsinghua University, Beijing 100084, China b

art ic l e i nf o

a b s t r a c t

Article history: Received 10 July 2015 Received in revised form 13 November 2015 Accepted 8 December 2015 Available online 14 December 2015

In pebble-bed high temperature gas-cooled reactors (HTGRs), spherical fuel elements move inside pipelines of a handling system, which should be controlled precisely. A detecting method for these elements is proposed in this paper, which is achieved by a detecting system with self-diagnosis function. Detecting signals are obtained by sensors installed outside pipes. A signal identification algorithm was designed for graphite ball detection. Electromagnetic simulations and detecting experiments were performed for system optimization and development of the method. The results show that the proposed method is capable for spherical fuel elements detection, and can be successfully used in practical application. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Pebble-bed HTGR Graphite ball detection Eddy current detection Electromagnetic simulation

1. Introduction As one of the most important energy sources to replace fossil fuel in the future, nuclear power and reactors have been widely researched and developed for decades. The high temperature gascooled reactor (HTGR), as an advanced type of nuclear reactor, performs excellent in both economic competitiveness and safety [1]. Unlike traditional reactors, in HTGR, nuclear fuel is distributed into numerous small fuel elements, making it possible to continuously control the loading and unloading processes without stopping the reactor [2]. Currently, the pebble-bed HTGRs are developed in Germany, Russia, China and South Africa [3,4], in which spherical fuel elements are used with graphite coats on the surface and coated fuel particles are dispersed uniformly in graphite materials of the core [5–7]. The fuel handling system of pebble-bed HTGR is one of the key techniques for continuous controlling of fuel elements’ loading and unloading. The handling system usually consists of a circulation subsystem for the fuel elements in use, together with a feed subsystem for the new elements and a discharge subsystem for the spent ones [8–11]. To ensure the system’s normal operation, it is critical to develop a reliable method to precisely detect the flow of spherical fuel elements, which move inside the pipelines of the n Corresponding author at: Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China. Tel.: þ86 10 62780012. E-mail address: [email protected] (Z. Han).

http://dx.doi.org/10.1016/j.ndteint.2015.12.005 0963-8695/& 2015 Elsevier Ltd. All rights reserved.

handling system. The issue can be simplified as to detect graphite balls moving inside stainless steel pipes, and all the balls should be distinguished and detected successfully even if they are close enough to each other as a contact mode. Several nondestructive detecting methods can be used to achieve this detecting task, including radiographic detection, acoustic detection and eddy current detection. It is supposed that the X-ray signals received in radiographic detection and the acoustic signals seized in acoustic detection can reflect information of the moving balls. However, the X-ray signals are likely to be affected by the nuclear radiation of the fuel elements themselves, while the acoustic signals can be interfered by many other kinds of sound, resulting in signal complication in both methods. Relatively, eddy current detection utilizes detecting coils to provide a detecting electromagnetic field, and the influence on the electromagnetic field caused by the moving balls will be reflected in the electrical signal of the coils. It has been proved that most of the interfering factors in this method can be recognized and removed by appropriate design of detectors and corresponding circuits. In a word, eddy current detection appears less difficult in signal processing and analysis than radiographic or acoustic methods, and it has been used in most of the existed pebble-bed HTGRs. At present, eddy current detection for graphite balls moving inside stainless steel pipes has been proposed or achieved mainly in three ways. In the first way, the pipe is drilled on its wall to form an installing hole for the detector, which can work as an electromagnetic proximity switch when the balls pass by [12]. In the


Z. Han et al. / NDT&E International 79 (2016) 81–91

second way, a tubular device takes place of a segment of the pipe, and a detecting coil with a ceramic supporting frame should be predisposed inside the device in advance. This method has already been achieved in HTR-10 of Tsinghua University in China [13,14]. The inductive reactance of the coil will be influenced by the balls, which is obtained and processed for elements’ detection and identification [15]. Obviously, in both the above two ways, pipes have to be partly damaged for the installation of detecting devices, which results in structure complication and mounting difficulty because of the higher requirements for gas tightness in HTGR. Moreover, the maintenance and replacement of the above detecting devices will become quite difficult once the reactor has been put into operation. To solve the above problem, outside-installed detectors have been proposed as the third detecting way, in which the detecting devices are directly installed outside the pipes. Li et al. fabricated a bidirectional detector based on a through-transmission eddy current testing probe [16,17], while Han et al. designed an external detection system with a sensor of two detecting coils [18]. These two detecting devices share a similar structure of double semi-ring frames wound with coils, facilitating their installation and maintenance. In the former detector, a primary coil winds on one of the frames and two pickup coils wind on the other. A stimulus signal drives the primary coil and the detecting signal is received from the pickup coils. Differently, in the latter detector of Han’s work, two pairs of semi-ring coils are combined into two whole-ring coils after the installation of frames, and both the stimulus signal and the detecting signal are loaded on the same coils. Field experiments have shown that the latter detector performs better in signal strength and detecting stability. However, Han’s work is at a preliminary stage in signal processing and identification, and is lacking a self-diagnosis function of the detecting device itself. As the detecting devices are installed outside the pipes, the signals of eddy current detection are quite weak due to the shielding of the metallic pipes. To achieve better detecting performance, an improved detecting method is proposed in this paper based on Han’s previous work, which is achieved by a detecting system with self-diagnosis function. The system is introduced in Section 2, together with a designed signal identification algorithm based on the signals obtained by the system. A finite element model for electromagnetic simulations was established for system analysis and parameters optimization, which is described in Section 3. Detecting experiments were performed for completion and verification of the detecting method, which are discussed in Section 4. The conclusions are presented in Section 5.

2. Fundamentals of the detecting method and system The detecting system proposed in this paper is derived from Han’s previous work [18] mentioned above, which is capable to seize the weak signals of eddy current detection. The system consists of an eddy current sensor and an instrument of several signal processing modules. The instrument is connected to the sensor by shielded cables, as shown in Fig. 1. The length of the cables can be more than 100 m, making it possible to place the instrument in a separate room to prevent from disposing under radiation. 2.1. Eddy current sensor For convenient installation and maintenance, the sensor was designed as a pair of semi-ring parts. Each part mainly consists of a housing and a frame wound with coils, which are also in semi-ring shapes. The two parts fit with each other and will combine into a whole-ring sensor after installation. The two semi-ring coils

Fig. 1. Structure of the detecting system.

Fig. 2. Coils after installation.

winding on two frames will form an approximate whole-ring coil, consisting of an inner circle and an outer circle with opposite current directions, as shown in Fig. 2. For electromagnetic field analysis, the parallel areas can be ignored because they contain straight wires of opposite current directions [19,20]. An upper coil and a lower coil are used in the sensor to successively obtain two signals for further processing. For clarity, only the current direction in the upper coil is shown in Fig. 2. When an alternating current signal is loaded on the detecting coils, an electromagnetic field will be stimulated inside the pipe. When a graphite ball passes the sensor, an eddy current will be inducted based on Faraday’s Induction Law. This inducted eddy current will influence the impedance of the coils, which will be detected and used to obtain ball-passing information. 2.2. Signal processing modules The detecting method is mainly achieved by several signal processing modules integrated in the instrument, including a microcomputer, a stimulus module, a pre-processing module and a self-diagnosis module, which work together as shown in Fig. 3. In Fig. 3, U1 and U2 respectively represent voltage signals obtained from the upper coil and the lower coil, while U0 represents the output signal of the pre-processing module. U0 can be

Z. Han et al. / NDT&E International 79 (2016) 81–91


Fig. 3. Relationship of signal processing modules.

Fig. 4. Typical U0 waveforms: (a) effective signals and (b) interfering signals.

considered as the difference between U1 and U2, representing the impedance difference between the upper and lower coils. The stimulus module generates a sinusoidal alternating current signal by a direct digital synthesis (DDS) chip, which drives a high power amplifier to provide the stimulus signal for the coils in the sensor. The frequency and phase of the stimulus signal are controlled by the DDS chip, while the configuration data is provided by the microcomputer. The pre-processing module obtains voltage signals U1 and U2 from the coils via a resonance bridge, and then successively processes them by differential amplification, band-pass filtering, phase sensitive detection and low-pass filtering. Consequently, the pre-processed signal U0 is generated, which is further analyzed by the microcomputer for signal identification. The identification algorithm will be further discussed in Section 2.3. Besides providing data to the DDS chip and analyzing U0 for signal identification and output, the microcomputer also provides a square signal with fixed frequency for self-diagnosis. The self-diagnosis module comprises a series of diagnosis circuits, each of which is connected to one of the modules, including the sensor, the stimulus module, the pre-processing module and the microcomputer. If any module fails, a failure signal will be

generated by the corresponding diagnosis circuit. The signals from all these circuits are collected by a logical AND operation, so only when all the parts work normally will the self-diagnosis module output a normal signal, otherwise a failure signal will be outputted. The self-diagnosis data (indicating whether all the parts work normally) together with the ball-passing data (indicating whether a ball has passed) will be outputted to an upstream processor for controlling or to a screen for display. 2.3. Signal identification algorithm The signal identification algorithm is mainly achieved by the microcomputer, based on U0 from the pre-processing module. The typical U0 waveforms are shown in Fig. 4, including the effective signals in Fig. 4(a) and the interfering signals in Fig. 4(b). A typical effective signal contains a positive pulse (Pulse þ ) and a negative pulse (Pulse  ), and the sequence of Pulse þ and Pulse  reflects the ball’s moving direction. A basic voltage (VB) will appear in U0 when no ball passes by, which is set according to demands. There are three kinds of typical interfering signals which often appear in U0 waveforms, including isolated pulses, discontinuous double pulses (with too large an interval) and unmatched double pulses.


Z. Han et al. / NDT&E International 79 (2016) 81–91

Fig. 5. Principles of the signal identification algorithm.

The task of the signal identification algorithm is to eliminate the interfering signals and other noises, and to identify effective signals. In the algorithm, a high threshold (VTH) and a low threshold (VTL) are used, which are determined by Eqs. (1) and (2) as below, in which Vmax represents the maximum of Pulse þ and Vmin represents the minimum of Pulse  . V TH ¼ 0:2ðV max  V B Þ þ V B


V TL ¼ 0:2ðV min V B Þ þ V B


The designed algorithm includes several steps carried out in sequence, as shown in Fig. 5 and described below. a) Segments extraction: The segments above VTH or below VTL are extracted for analysis. b) Lasting-time check for each segment: The lasting time of each extracted segment (Ts) is checked and required to be within a time scope (Tscope). Tscope should be set in advance according to the demands about the balls’ moving speed. c) Deviation check for each segment: Each extracted segment is compared with a standard segment to calculate the deviation parameter Dev. Dev is calculated based on the discrete signals. Eqs. (3) and (4) depict the calculation of Pulse þ and Pulse  respectively. W þ (n) and W  (n) represent the segments extracted from Pulse þ and Pulse  respectively, and S(n) represents the standard segment. N1 and N2 respectively represent the total number of sampling points in W þ (n) and W  (n). The standard segment should be defined and generated in advance according to the simulation and experimental results, which will be further discussed in Section 4.2. For deviation check, Dev is required to be less than a deviation threshold, DevT.  N 1   1 X  W þ ðnÞ  V TH   SðnÞ  N1 n ¼ 1 max ½W þ ðnÞ  V TH 


 N 2   1 X  V TL  W  ðnÞ   SðnÞ Dev ¼  N2 n ¼ 1 max ½V TL  W  ðnÞ


Dev ¼

required to be within an MRscope.

MR ¼

1 N1 1 N2

N1 P n¼1 N2 P

½W þ ðnÞ  V TH 


ð5Þ ½V TL  W  ðnÞ

In the above algorithm, a pulse is recognized as effective only when its extracted segment passes the lasting-time check and the deviation check. If there exist a pair of effective pulses, the interval-time check and matching check will be carried out for them. If any check fails, the former effective pulse will be abandoned and the latter one will be retained for the following pulse. Otherwise, if all the checks are satisfied, an effective ball-passing signal is identified and the ball-passing data will be generated and outputted accordingly.

3. System simulation and parameters optimization As the graphite is non-ferromagnetic material, the effect of the moving graphite balls on the coils’ impedance is usually weak [21], and the shielding of the metallic pipes makes the situation even worse. Thus, in order to obtain effective signals, the amplification factor of the whole detecting system has to be large enough, which will result in high intensity interfering signals at the same time. It is important to optimize the system for a better signal-to-noise (S/ N) ratio and a smaller amplification factor. As the sensor is the main part for signals detection, a finite element model of the eddy current sensor was established for electromagnetic simulations. The key parameters of the sensor were optimized according to the simulation results, including the stimulus frequency and structure parameters of the coils [19,22]. Moreover, the radius deviation of the graphite balls and the distance between two coils were further analyzed by this simulation method. 3.1. Finite element modeling and simulation fundamentals

d) Interval-time check for pulse-pair: The interval time between the maximum and minimum moments of a pair of pulses, identified as Ti, is checked and required to be an appropriate value. Ti is required to be less than 1.2  min(Ts1,Ts2), where Ts1 and Ts2 represent the lasting times of the two extracted segments from Pulse þ and Pulse  . e) Matching check for pulse-pair: A matching ratio (MR) is calculated to estimate that to what extent a pair of pulses match with each other. MR is calculated according to Eq. (5), which is

The low-frequency electromagnetic analysis module of software ANSYS was used to simulate the electromagnetic field generated by the sinusoidal stimulus signals in the detecting coils. As described above, the parallel areas of the coils can be ignored in analysis. Each coil can be recognized as two concentric circles with opposite current directions, and frames that contribute little to the electromagnetic field can be ignored. Moreover, the metal external housing was also ignored, as it contributes little to the electromagnetic field inside the pipe. Thus, the sensor can be simplified into an axisymmetric model, consisting of a graphite ball, a segment of the pipe, the inner circle and the outer circle of a coil (representation of either the upper coil or the lower coil), and the air surrounding them. The model was built in ANSYS in an

Z. Han et al. / NDT&E International 79 (2016) 81–91


magnetic vector potential (AZ), time-integrated electric scalar potential (VOLT), electric current (CURR), and electromotive force (EMF). For the coil, both AZ and CURR DOFs were used to model it as a voltage-fed stranded coil, and all the coil’s nodes had been coupled in the CURR DOF to solve all the current values in a single equation. The cross-sectional area (Hcoiδcoil), the fill factor (Fill), the total number of coil turns (Ncoil) and other parameters of the coil were also set for solution, among which the current directions of the two circles were forced to be opposite. For the ball, the pipe segment and the surrounding air, only AZ DOF of element was used. At the symmetric axis (left boundary in Fig. 6), the direction of magnetic flux was set parallel as the boundary condition. At other boundaries, far-field elements INFIN110 were used to complete boundary conditions, which was not shown in Fig. 6 for simplicity. A sinusoidal voltage drop was loaded on the voltage-fed stranded coil as the stimulus signal, whose effective value was set to 6 V. In the simulation, the impedance of the coil was analyzed, which can be calculated by Eqs. (7) and (8), including the resistance XR and the reactance XI. XR ¼ U

XI ¼ U

Fig. 6. Axisymmetric model of finite element simulation. Table 1 Typical values and ranges of analyzed parameters and given parameters in simulation. Name

Typical value


Stimulus frequency (Fs) Distance between circles of coil (Lcoil) Total number of coil turns (Ncoil) Radius of graphite ball (Rball) Inner radius of pipe (RIpipe) Outer radius of pipe (ROpipe) Inner radius of coil (RIcoil) Radius of each wire of coil (r) Fill factor of coil (Fill) Height of coil (Hcoil) Thickness of each circle of coil (δcoil)

4 kHz 35 mm 200 30 mm 32.5 mm 38 mm 39 mm 0.2 mm 0.8 10 mm 0.016Ncoil

0.1–10 kHz 10–100 mm 50–1000 Given condition Given condition Given condition Given condition Given condition Given condition Given condition Calculated by Eq. (6)

axisymmetric way, as shown in Fig. 6. The analyzed parameters and given parameters used in the simulation are shown in Table 1, together with their abbreviated symbols shown in the parentheses. Most of the geometry parameters in Table 1 are also shown in Fig. 6. The parameters named as Fs, Lcoil and Ncoil were analyzed and optimized by the simulation method within their ranges shown in Table 1. In order to ensure that only one of the parameters was analyzed at a time, typical values of them were applied. The typical value was determined by iterative simulations. Other parameters in Table 1 were set as given conditions corresponding to practical demands. The δcoil can be calculated according to Ncoil, based on the definition of the fill factor (Fill) as described by Fill ¼

Ncoil  π r 2 δcoil  Hcoil


Due to the axisymmetric feathers of the model, the simulation was carried out using a 2-D harmonic magnetic analysis method. The 8-node element PLANE53 was applied for modeling, and each node has up to 4 degrees of freedom (DOFs): z component of the

I real I 2real þ I 2imag

I imag 2 I real þ I 2imag



Ireal and Iimag respectively represent the real part and the imaginary part of the coil’s current, and U represents the effective value of the voltage drop loaded on the coil, so U¼ 6 V. The total impedance can be expressed as Z¼ XR þjXI, and its amplitude can be calculated as |Z| ¼(XR2 þXI2)0.5. As the fundamentals of the detecting method, the change of the coil’s impedance should be analyzed, which can be calculated by comparing the coil’s impedance in the detecting situation (with a ball in position) with that in the standard situation (using an air– core coil without balls). The changing rates of XR, XI and |Z|, respectively identified as dXR, dXI, and dZ, were used as evaluating indexes for parameters optimization and further analysis. They represent the detecting sensitivity. Practically, these changing rates can be calculated as Eqs. (9)–(11), in which XRd, XId, |Zd| and XRs, XIs, |Zs| respectively represent those in the detecting situation and the standard situation. dX R ¼

X Rd  X Rs  100% X Rs

dX I ¼

X Id  X Is  100% X Is


  Z d   jZ s j  100% jZ s j


dZ ¼


3.2. Preliminary simulation Firstly, a preliminary simulation was carried out when all the analysis parameters were set to their typical values. In order to observe the generated electromagnetic field, the contour line plots of magnetic flux lines were simulated, as shown in Fig. 7. The plots indicate that the imaginary content is much stronger than the real content, mainly because the inductance of the coil (imaginary content) plays a leading role in the detection. However, the real content is less influenced by the pipe’s shielding, as the shielding effect caused by the pipe’s eddy current exerts mostly on the inductance. Practically, both the real and imaginary components contribute to the detecting signals, so the total amplitude of impedance should be mostly concerned.


Z. Han et al. / NDT&E International 79 (2016) 81–91

Fig. 7. Contour line plots of magnetic flux lines: (a) real part and (b) imaginary part.

Fig. 8. dI curves and derived curves: (a) dI curve of single coil; (b) extended dI curves of two coils; and (c) derived curve of dI difference between two coils.

Afterwards, another preliminary simulation was performed, in order to simulate the waveforms when the graphite balls pass by. The ball was placed at different heights within a scope of [– 50 mm,50 mm], while the coil’s center is at 0 mm. The actual signals are directly related to the change of current. The changing rate of the current value, identified as dI, is applied for this simulation. dI is calculated as      U  U    Z d  Z s  jZ s j  Z d  ð12Þ dI ¼   ¼   Zd U  Zs

The dI curve of a single coil is shown in Fig. 8(a), which shows the changing trend of dI when the ball pass through the coil. As the dI curve shows a shape of a normal distribution, dI can be considered to stay the same when the ball is more than 50 mm away from the coil. To simulate the actual waveforms, both coils should be taken into account. The distance between the two coils was set to 30 mm, and how to decide this distance is discussed in Section 3.4. In the simulation, the coils were separately placed at the position of 15 mm and  15 mm, and the dI curves of both coils

were extended to a position range of [  80 mm,80 mm]. The extended curves are shown in Fig. 8(b). Assuming that the electromagnetic fields generated by two coils influence little on each other, the dI difference between the two coils can be calculated by simply comparing the curves in Fig. 8(b). The derived result is shown in Fig. 8(c), which is similar to the actual detecting signals shown in Section 4.1. 3.3. Parameters optimization By the simulation method, the parameters of the detecting system have been optimized for a better detecting performance. Three parameters were analyzed, including Fs, Lcoil and Ncoil, whose value ranges in the simulation are shown in Table 1. The Fs analysis was carried out firstly, and the variation of dXR, dXI, and dZ along with Fs are shown in Fig. 9. In eddy current detection, the imaginary component plays the more important role, so XI is much stronger than XR, and the curve of dZ shares a similar changing trend with that of dXI. With the increase of Fs, dXR increases a bit at lower frequency, and it goes down to negative

Z. Han et al. / NDT&E International 79 (2016) 81–91


Fig. 9. Simulation results of Fs analysis: (a) dXR curve; (b) dXI curve; and (c) dZ curve.

Fig. 10. Simulation results of Lcoil and Ncoil analysis: (a) |dZ| curve of Lcoil and (b) |dZ| curve of Ncoil.

values at higher frequency and reached an extreme value at 6 kHz. The curves of dXI and dZ were negative at most points, and the extreme value appears at around 2 kHz. Theoretically, when the ball passes the coils, the real component XR will get bigger and the imaginary part XI will become smaller [20], in the case that the influence of the pipe has been ignored. However, most of the eddy current appears in the metal pipe, which will interact with the moving ball. This interaction effect will cause both XR and XI become smaller, and the effect gets more significant at higher frequency. Thus, the interaction between the pipe and the ball will lead to the complication of the sensor’s frequency response, resulting in the complex curves of Fig. 9. When merely considering the detecting sensitivity, 2 kHz can be the optimal frequency, as dZ is the most concerned index. However, in practical use, each of the pulses (Pulse þ and Pulse  ) is required to contain at least 10 cycles of the stimulus signal, in order to reliably distinguish ball-passing signals from interfere

signals. As the ball’s moving speed can be up to 10 m/s at most, the lasting time of each pulse can reach a minimum of about 3 ms, so it is better to set the stimulus frequency higher than 3.3 kHz. As a balance, 4 kHz was finally chosen as the actual frequency applied in practice. At 4 kHz, dZ reaches 0.051%. Afterwards, Lcoil and Ncoil were studied at the chosen frequency of 4 kHz, based on the absolute value of dZ, identified as |dZ|. dZ is a comprehensive consideration of both the real part and imaginary part, and |dZ| focuses on the value of dZ. The |dZ| curves are shown in Fig. 10, together with the chosen optimal points. In Fig. 10(a), with the increase of Lcoil, |dZ| first rises up and then falls down, and it reaches a peak value of 0.05% at 35 mm. Because of the opposite current direction, the electromagnetic field generated by the outer circle has a counteraction effect on the field generated by the inner circle, so |dZ| will increase when the outer circle moves away. Meanwhile, the impedance of the outer circle increases as its size grows larger, so the voltage loaded on the


Z. Han et al. / NDT&E International 79 (2016) 81–91

inner circle decreases, resulting in the decrease of |dZ|. When Lcoil is relatively small, the decrease of the outer circle’s counteraction effect is greater, so |dZ| increases firstly. When Lcoil gets larger enough, the voltage decrease of the inner circle is more important, so |dZ| decreases afterwards. Thus, 35 mm can be chosen as the ideal value of Lcoil. In Fig. 10(b), |dZ| decreases all the way when Ncoil increases. This is mainly because δcoil gets larger due to Eq. (6) along with the increase of Ncoil. Considering RIcoil and Lcoil remain the same, the center of each circle moves away from the pipeline when Ncoil increases, so the detecting sensitivity decreases. However, the coil turns should not be too few, as the strength of the electromagnetic field has to be high enough to improve the S/N ratio. Therefore, coils of 200 turns were selected for practical situation, and |dZ| is 0.051% at this point. 3.4. Further analysis and discussions Although Rball has been considered as a given parameter in the above simulations, its value may still change due to manufactural deviation or working abrasion. The effect of Rball’s deviation on |dZ| was analyzed utilizing the simulation model in Fig. 6. The results are shown in Table 2. The changing rate of |dZ| can be as big as over 49%, which is unacceptable for the ball’s detection with tiny detecting signals. For stable operation of the detecting system, |dZ| should not vary for more than 30%, which means that the deviation of Rball should be less than 71.2 mm or 7 4%. To analyze the distance between the coils, a particular situation has to be considered, in which several tangent balls pass the detector in a contact mode. To simulate this situation, two tangent balls were analyzed, whose center-to-center distance was 60 mm (2Rball). Two extended curves based on Fig. 8(a) were simply summed together as an approximation for the dI curve of a single coil. The result is shown in Fig. 11(a). Using a deriving method similar to that of Fig. 8, derived curves of dI difference between the two coils can be obtained. The distance between the coils was

Table 2 Simulation results of Rball’s deviation. Rball (mm)


Changing rate of |dZ|

28.0 28.8 29.6 30.0 (Standard) 30.4 31.2 32.0

0.034% 0.040% 0.047% 0.051% 0.055% 0.065% 0.076%

–32.88% –21.26%  7.66% 0 8.30% 27.06% 49.21%

successively set to 10 mm, 30 mm and 50 mm, and results are shown in Fig. 11(b)–(d). When the distance is less than 30 mm, the signals of two balls gets overlapped, resulting in a loss of the detecting sensitivity. When the distance gets larger than 30 mm, the curve of dI difference becomes deformed like Fig. 11(d), and the signal identification cannot be achieved in this situation. As a result, the distance between coils should be precisely set to 30 mm, in order to detect moving balls in a contact mode. In above simulations, the external metal housing of the sensor was ignored, as only the electromagnetic field inside the pipe was concerned for the detection of the moving ball. However, this metal housing is very important in practical situation because it can prevent the coil from outside electromagnetic influence. As described in Section 2.1, the housings are assembled outside the frames in the designed sensor, and the metal free zone between the outer coil and the housing is more than 25 mm (in both radial and axial direction), in order to decrease the influence on the detection. The relative analysis has shown that the housings affect little on the field inside the pipeline, but can significantly screen the electromagnetic field outside the coils.

4. Experimental results and discussion Based on the above detecting method and devices, detecting experiments were carried out, and the results are displayed and analyzed in the following sections. The simulation model and the signal identification algorithm have been verified based on the experimental results. Moreover, a debugging device was developed to improve the convenience of on-site debugging. To confirm the working stability of the system, electromagnetic interference (EMI) experiments were performed, and the influence of vibration effect and wobble effect was discussed. 4.1. Experimental results and model verification In the experiments, the eddy current sensor was installed outside a vertical pipe segment and the instrument was connected to the sensor. The waveforms of the pre-processed signal U0 were acquired using an oscilloscope, in which the VB, VTH and VTL had been respectively set to 2 V, 2.4 V and 1.6 V. The stimulus frequency was set to 4 kHz in these experiments. Four experiments, numbered from 1# to 4#, were carried out, in which a graphite ball was released from different heights away from the sensor. The ball speeds in these experiments can be respectively estimated as 1 m/s, 2 m/s, 3 m/s and 4 m/s, based on their releasing heights. The results of these experiments are shown in Fig. 12.

Fig. 11. Derived results for two tangent balls: (a) dI curve of single coil; (b) dI difference between two coils of 10 mm distance; (c) dI difference between two coils of 30 mm distance; and (d) dI difference between two coils of 50 mm distance.

Z. Han et al. / NDT&E International 79 (2016) 81–91


Fig. 12. Experimental results of different releasing heights (different ball speeds).

Fig. 13. Waveforms of supplemental experiments: (a) 3 tangent balls and (b) incomplete ball.

Moreover, two supplemental experiments were carried out, and the waveforms are shown in Fig. 13. Fig. 13(a) shows the waveform when 3 tangent balls passed the sensor in a contact mode, while Fig. 13(b) shows the waveform when an incomplete ball (about 1/3 crown of the ball had been cut off) passed the sensor. Although the signal of the incomplete ball was weaker than the normal balls, it can still be detected by the method, which meets the practical demand. The experimental waveforms in Fig. 12 are similar with the simulation results in Fig. 8, and the waveforms of Fig. 13(a) and Fig. 11(c) share similar feathers. These evidences can help to illustrate the validity of the simulation model used in Section 3. 4.2. Algorithm completion and verification In order to complete the signal identification algorithm described in Section 2.3, the standard segment for the deviation check need be defined primarily. The derived results of dI difference in Fig. 8(c) were used to generate the standard segments. The

waveform was extracted into two segments according to the thresholds, and then they were normalized and fitted by two quintic polynomials S1(x) and S2(x), as displayed in Fig. 14. S1(x) and S2(x) were then discretized according to the number of sampling points of experimental segments, which were acquired from signals in Fig. 12. In this way, the standard segments were generated. The values of Dev were calculated according to Eqs. (3) and (4), and whether S1(x) or S2(x) would be used to generate S(n) was determined by the relative peak position of each experimental segment. Moreover, Ts, Ti and MR of the experimental segments were also calculated, and all the calculated results are shown in Table 3. Based on some advanced research results, DevT and MRscope were separately set to 0.05 and [0.8,1.25], while Tscope was set to [2 ms,300 ms] as a corresponding scope of [0.1 m/s,10 m/s] for the ball speeds. The results in Table 3 indicate that all the experimental segments have passed the checks and all the effective ball-passing signals have been successfully identified by the designed algorithm.


Z. Han et al. / NDT&E International 79 (2016) 81–91

Fig. 14. Fitting polynomials S1(x) and S2(x).

Table 3 Calculation results of signal identification algorithm for experimental segments. Experiment

Ts (ms)


Ti (ms)



29.44 (  ) 26.60 ( þ ) 18.04 (  ) 18.80 ( þ) 11.96 (  ) 12.80 (þ ) 8.72 (  ) 9.72 ( þ )

0.0256 (  ) 0.0383 (þ ) 0.0244 (  ) 0.0312 ( þ) 0.0180 (  ) 0.0324 ( þ ) 0.0246 (  ) 0.0334 ( þ )









2# 3# 4#

4.3. Debugging device development In practical use of the detecting system, the pipeline conditions, the sensor’s manufacturing deviation, the installation precision and other detecting situations are likely to vary from place to place, as the sensor will be installed at different places of the fuel handling system. At each detecting point, the detecting signals should be amplified enough without saturation distortion, which is necessary for better sensitivity and effectiveness of the detecting method. To achieve this purpose, a debugging procedure is needed before the system is put into operation. During this procedure, the graphite balls are required to pass through the pipe segment where the sensor installs, in order to adjust the parameters of the system. In laboratory, the repeated experiments are easily achieved as only pipe segments are used, but it is difficult to carry out these experiments in practical reactors as the handling system is always totally sealed up. To improve the convenience of on-site debugging, a debugging device was developed accordingly, which can simulate virtual detecting signals without balls passing through the handling system’s pipelines. The device mainly contains two sets of variable impedances, whose working principle is shown in Fig. 15(a). The device was placed between the sensor and the instrument, and each set of the variable impedances was inserted between one of the coils and the resonance bridge of the pre-processing module. The virtual signals were generated by simulating the changing trend of impedances, just like actual balls had passed by. For the practical application of these variable impedances, programmable inductors were used in the device, which can be controlled appropriately by a microcomputer and other necessary circuits. A virtual signal generated by the device is shown in Fig. 15(b), which is similar to the experimental results in Fig. 12. This debugging

device can facilitate on-site debugging, and its feasibility has already been proved by on-site application. 4.4. Discussions To ensure the working stability of the detecting system, several influence factors should be considered. Firstly, the ability of the detecting system against electromagnetic interference needs to be evaluated. Consequently, electromagnetic interference (EMI) experiments were carried out according to IEC-62236-3-2: 2008 [23], including a conducted emission test and an electromagnetic radiation test. In the former test, radio-frequency (RF) interferences were introduced to the detecting system via the power cables and transducer cables in a conducted emission way. In the latter test, the RF interferences were loaded on the detecting system via aerials as the electromagnetic radiation. During the interference periods, graphite balls passed through the sensor discontinuously. The results show that all the ball-passing signals were identified successfully without missing or false results. Furthermore, the influence of vibration effect should also be considered. In practice, the sensor is installed outside the pipeline firmly to avoid movement between the sensor and the pipeline. Thus, the vibration effect will only appear when the pipeline vibrates together with the sensor. To evaluate the ability of the detecting system against vibration of the pipeline, a vibration experiment was carried out on the vibrating platform UDSAI60H560B/ST. In the experiments, the sensor was installed on a segment of pipe, which can be vibrated in both radial and axial directions. The vibration acceleration spectral density is set to 2.0 (m/s2)2/Hz at frequency of 10–100 Hz. The results show that the system can work stably all the time. Besides, the influence of wobble effect should also be analyzed. When the ball passes by the sensor, the offset of the graphite ball from the central line of pipe is likely to occur. A wobble experiment was performed to study this situation, in which the ball was forced to pass the sensor at different distance to the symmetric axis. The results show that the influence of wobble effect is very little and does not affect the detection. In fact, wobble effect may enhance the signal in one side of the coils and weaken the signal in the other side, but the total signal of the whole sensor will almost remain the same. 5. Conclusions In this paper, a detecting method for spherical fuel elements in pebble-bed HTGR is proposed and achieved by a detecting system

Z. Han et al. / NDT&E International 79 (2016) 81–91


Fig. 15. Principle and performance of debugging device: (a) working principle of the device and (b) virtual signal generated by the device.

with a self-diagnosis function. The detecting system consists of an eddy current sensor and an instrument of several signal processing modules. By the electromagnetic simulation method, several parameters of the sensor were analyzed and optimized. Simulation results indicate that the optimal values of several key parameters should be Fs ¼ 4 kHz, Lcoil ¼35 mm and Ncoil ¼200. The effect of Rball’s deviation on the detecting performance was also analyzed, which indicates the deviation should be less than 74%. Moreover, the distance between two coils is required to be precisely 30 mm for the detection of moving balls in a contact mode. A fuel ball signal identification algorithm was designed, which requires several sequential steps, including the segments extraction, the lasting-time and the deviation check for each segment, and the interval-time and matching check for each pulse-pair. Detecting experiments were carried out and the results were analyzed. The experimental results are helpful to illustrate the validity of the simulation model, and help to complete the identification algorithm and validate its availability. Moreover, a debugging device was developed to improve the convenience of on-site debugging, as it can simplify the debugging procedure. Several influence factors have been discussed and tested to ensure the working stability of the detecting system, including the electromagnetic interference, the vibration effect and the wobble effect. Experiments has validated the working stability of the detecting system under the environment of EMI, vibration and wobble. The results of analysis and experiments have shown that the proposed detecting method is capable for spherical fuel elements detection, and can be successfully used in practical system.

References [1] Abram Tim, Ion Sue. Generation-IV nuclear power: a review of the state of the science. Energy Policy 2008;36(12):4323–30. [2] Xiangwen Zhou, Zhenming Lu, Jie Zhang, Bing Liu, Yanwen Zou, Chunhe Tang, Yaping Tang. Preparation of spherical fuel elements for HTR-PM in INET. Nucl Eng Design 2013;263:456–61. [3] Xu Yuanhui, Zuo Kaifen. Overview of the 10 MW high temperature gas cooled reactor—test module project. Nucl Eng Design 2002;218(1–3):13–23. [4] Zhang Zuoyi, Wu Zongxin, Wang Dazhong, Xu Yuanhui, Sun Yuliang, Li Fu, Dong Yujie. Current status and technical description of Chinese 2  250 MWth HTR-PM demonstration plant. Nucl Eng Design 2009;239(7):1212–9.

[5] Zhou XW, Tang CH. Current status and future development of coated fuel particles for high temperature gas-cooled reactors. Prog Nucl Energy 2011;53 (2):182–8. [6] Yang Min, Li Rui, Duan Jinghui, Liang Lihong, Li Xingdong, Liu Wenli, Meng Fanyong. Analysis of DR testing blind zone of spherical fuel elements for 10 MW high-temperature gas-cooled reactor. NDT E Int 2013;60:77–86. [7] Yang Min, Zhang Jianhai, Song Sung-Jin, Li Xingdong, Meng Fanyong, Kang To, Liu Wenli, Wei Dongbo. Imaging and measuring methods for coating layer thickness of TRISO-coated fuel particles with high accuracy. NDT E Int 2013;55:82–9. [8] Liu JG, Xiao HL, Li CP. Design and full scale test of the fuel handling system. Nucl Eng Design 2002;218(1–3):169–78. [9] Mu Zhao, Bo Ma, Yujie Dong. The strategic study of pebble model high temperature gas-cooled reactor plant with power generation feature and industrial application prospect. Energy Environ Prot 2011;5(25):1–4 [in Chinese]. [10] Liu Hongbing, He Ayada, Du Dong, Wang Xin, Zhang Haiquan. A deceleration system for near-diameter spheres in pipeline transportation in a pebble bed reactor based on the resistance of a pneumatic cushion. Nucl Eng Design 2014;275:61–8. [11] Zhe Sui, Jun Sun, Chunlin Wei, Yuanle Ma. The engineering simulation system for HTR-PM. Nucl Eng Design 2014;271:479–86. [12] Albert Muncke. Graphite ball detectors for the fuel handling machine of a helium cooled pebble bed reactor. Kerntechnik 1976;18(5):201–6. [13] Yicheng Jia, Qingxian Ma, Wei Jiao, Zhicheng Wang. Development of bidirectional detector for the fuel element in high temperature gas cooled reactor. Nucl Electron Detect Technol 2000;20(6):465–8 [in Chinese]. [14] Tiefu Liu, Zandong Han, Qingxian Ma. Design of the fuel balls inspecting system for pebble-bed HTGR. Nucl Electron Detect Technol 2008;28(2):290–3 [in Chinese]. [15] Jiguo Liu, Guolin Xu, Qingxian Ma, Mei Xu. Development of spherical fuel element detector. Nucl Power Eng 1998;19(5):84–8 [in Chinese]. [16] Dong Li, Guo Sun Zhen, Qiang Chen. Detection of graphite balls for the fuel handling system in HTGR using eddy current testing. Nondestrut Test Eval 2010;25(2):169–79. [17] Dong Li, Qiang Chen, Zhenguo Sun. Development of novel sensor for fuel balls detection in high temperature gas-cooled reactor. At Energy Sci Technol 2010;44(4):451–5 [in Chinese]. [18] Zandong Han, Tiefu Liu, Qingxian Ma. External detection system for ball fuel elements in high temperature gas cooled reactors. J Tsinghua Univ (Sci Technol) 2007;47(8):1259–62 [in Chinese]. [19] Liang Zhang, Zandong Han. Simulation analysis of electromagnetic field for outside-fixed counter sensor in HTGR. Instrum Tech Sensor 2009;11:11–3 [in Chinese]. [20] Liang Zhang, Zandong Han. Parameter optimization of detecting coil design for fuel ball counter in high temperature gas-cool reactor. J Tsinghua Univ (Sci Technol) 2010;50(2):191–4 [in Chinese]. [21] Dekdouk Bachir, Chapman Robert, Brown Matthew, Peyton Anthony J. Evaluating the conductivity distribution in isotropic polycrystalline graphite using spectroscopic eddy current technique for monitoring weight loss in advanced gas cooled reactors. NDT E Int 2012;51:150–9. [22] Wilde Jürgen, Lai Yuqing. Design optimization of an eddy current sensor using the finite-elements method. Microelectron Reliab 2003;43(3):345–9. [23] IEC-62236-3-2 (Edition 2.0, 2008-12). Railway applications—electromagnetic compatibility—Part 3-2: rolling stock—apparatus.