Depicting corrosion-born defects in pipelines with combined neutron/γ ray backscatter: a biomimetic approach

The identification of corrosion, cracks and defects in pipelines used for transporting oil and gas can reduce the possibility of leaks, and consequently, it can limit the extent of an environmental disaster, public hazard and the associated financial impact of such events. Typically, corrosion in oil pipelines is measured with non-destructive ultrasonic or electromagnetic techniques, on the basis that corrosion and defects are often manifest as a change of thickness in the steel from which pipelines are made. However, such approaches are not practical for underground pipelines and their deployment can be complicated for the case of pipelines covered by insulation. In this paper, we present an innovative, non-destructive testing technique, which exploits the backscatter of a combination of fast-neutron and γ radiation from steel samples of a variety of thicknesses consistent with changes that might arise due to corrosion of a pipe wall. Our research demonstrates the potential to measure and characterise different steel thicknesses by detecting both the elastic, fast-neutron backscatter and the Compton-scattered γ radiations, simultaneously. Further, we demonstrate that the presence of insulation yields a consistent and separable influence on the experimental, wall-thickness measurements. The data from experimental measurements are supported by a comprehensive Monte Carlo computer simulation study.


Results
Detection system and neutron-γ collimator. The neutron and γ-ray flux backscattered from a variety of different steel slabs was measured with a system designed for this purpose, comprising: a mechanical rig, collimator and an array of small, organic scintillation detectors (Scionix, Netherlands) which were connected to a multiple-channel, mixed-field analyzer (Hybrid Instruments Ltd.) and to a bespoke embedded control system. Measurements were carried out at the low-scatter neutron facility at the National Physical Laboratory (NPL), in Teddington, London, UK. The rig-source-collimator-detector system consists of an aluminum frame in which two stepper motors have been configured to afford the system two degrees of freedom (in the X-Y plane, Fig. 1a,b). The collimator is designed to constrain the neutron and γ radiation generated by the 252 Cf to a defined area of interest of the steel sample, comprising cylinders of high-density polyethylene (HDPE) and lead (Pb).
Organic liquid scintillation detectors of type EJ-301 40 are secured in a fixed angle of orientation with adjustable arms. Up to eight detectors can be used in the geometrical configuration shown in Fig. 1. However, it has been found that four detectors enable the total geometric efficiency to be optimised, i.e., maximising the area of the hemisphere subtended by the detectors. The position of the radiation-sensitive liquid-volume in the detectors has been configured to be in a region of the space below the collimator sheltered from the radiation flux that streams from the 1-cm diameter, collimator pinhole (Figs. 1b and 2a). With this particular approach, the detectors are isolated from the incident flux but, at the same time, they are positioned sympathetically with the trajectory of the backscattered flux generated at the focus of the radiation beam collimated by the pinhole. The beam-and-detector focus-area is the point where the radiation beam reaches the surface of a given sample, and it is located approximately 10 cm after the collimator aperture (Fig. 2a).
An extensive Monte Carlo computer simulation study was performed in order to identify the best collimator geometry and detector position. MCNP6 41,42 (Monte Carlo N-Particle transport code), developed in Los Alamos National Laboratory 43 , is the tool used to model the experiment and compute the simulations. The results are presented in Fig. 2a,b. Neutron and γ-ray flux are shown in 3-dimensions for the X-Z, X-Y and Y-Z planes. In particular, the plots show the flux beyond the collimator aperture, with the Y-Z plane modelled at X = 40 cm, i.e., approximating to the radiation focus at the sample. Detectors are located at circa 5 cm from the beam, in the position of minimum flux. The elevation (top) and plan (bottom) CAD views of the system, with a focus on the collimator-detector combination. The collimator is made of layers of lead and polyethylene, height: 30 cm, external diameter: 10 cm and internal diameter of the pinhole: 1 cm. The radiation source is placed above the collimator, concentrically with the pinhole. (c) A photograph of the system built, deployed and used at the low-scatter facility of the National Physical Laboratory. (d) A sequential schematic of the experimental set-up: the mechanical rig, controlled remotely by software running on a laptop, drives the system comprising the source, collimator and detectors. The detectors are connected to a multiple-channel, mixed-field analyser, which digitises the detected events, separates the neutron and γ-ray signals and sends a corresponding transistor-transistor logic signal for each event to an embedded digital counter that is linked by Ethernet to a computer. The entire system is coordinated and controlled by a graphical user interface.
www.nature.com/scientificreports www.nature.com/scientificreports/ calibration. The EJ-301 detectors were calibrated prior to the backscatter measurements by setting the high-voltage of the photomultiplier of each detector, to yield a balanced response across all four units, and the pulse-shape discrimination parameters were adjusted via the MFA to optimise the discrimination between neutrons and γ rays. Using a 17 MBq 137 Cs source, the voltage of the scintillator photomultiplier was adjusted to align the caesium Compton edges (at ~478 keV) to the same ADC channel (Fig. 3, left, inner plot on the top). The use of the 137 Cs source ensures a response to γ rays of a single energy (662 keV), thus making possible to set the discrimination threshold on each individual scintillator, using only the γ plume (Fig. 3, scatter plot on the left). Pulse-shape discrimination is performed by the MFA via a pulse gradient analysis algorithm 44 . The discrimination value (i.e., the ratio between discrimination amplitude and signal peak amplitude) of each γ event generated by the aforementioned caesium source was also calculated. The results (Fig. 3 left, inner plot on the bottom) show the presence of a single peak, consistent with only γ radiation being present for the case of the 137 Cs source. Subsequently, 252 Cf was used in order to verify the n-γ response of the scintillators and to verify the PSD threshold settings. Two separate plumes were observed (Fig. 3, scatter plot on the right) consistent with an appropriate PSD setting; two peaks are also observed when the discrimination value of the mixed neutron/γ field events is plotted as a histogram. neutron-γ backscatter. When a narrow, collimated radiation field hits a material, part of its radiation is transmitted, part is absorbed and part is backscattered (Fig. 4a). The flux of backscattered radiation, as a function of the material thickness, is related closely to the linear attenuation law, where Φ tr is the transmitted component for a thickness x, of the initial flux Φ 0 , μ is the linear attenuation coefficient for a field comprised of γ rays, whilst it corresponds to the total macroscopic cross section (Σ tot ) in radiation fields constituted by neutrons 45 . The backscattered flux can be expressed as, sc abs t r 0 where Φ abs is the absorbed component of the radiation flux, for a thickness x. Thus, the scattered flux is given by, www.nature.com/scientificreports www.nature.com/scientificreports/ If the radiation encounters a combination of different materials (see example depicted in Fig. 4b), whilst the physical principle is the same, the mathematical model becomes more sophisticated because both scattered and absorbed components derive from the superposition of effects from each of the different compounds. Figure 4c presents the measured, experimental backscattered neutron flux as a function of the steel thickness measured in this research; the backscattered neutron flux was assumed to be isotropic but this is not the case for γ rays, since Compton scattering is not isotropic. These experimental data are compared with the mathematical model presented in Eq. (3). Experimental results are presented with both ±1σ and ±3σ standard deviation and demonstrate consistency with the model given in Eq. (3).
The flux of backscattered neutron and γ rays are presented in Fig. 5a, as a function of the steel thickness in the presence of a 1-cm thick layer of high-density polyethylene and 1-cm thick layer of concrete, to illustrate the effect of insulation on the technique. The reflected neutrons and γ rays were measured over a period of 20 minutes for each individual sample of steel, for a total of 1 hour per slab given the three cases as follows: bare steel; steel and polyethylene; and steel and concrete. The results from these measurements are compared qualitatively with the corresponding results from MCNP6 simulations (Fig. 5b) obtained prior to the experiment. In this case, uncertainties are presented as ±1σ and the fit to the data is a second-order polynomial function.
The possibility of γ rays arising from inelastic neutron reactions exists but this is anticipated to be small for the energy of the neutrons from 252 Cf and therefore this has not been accounted for in this research. If sources with harder neutron spectra are considered for this application, such as AmBe or D-T generators, then this influence would need to be quantified. combined n-γ backscattered tomography: a case study. For the case of neutron and γ-ray backscatter from a sample comprising steel and concrete (included by way of insulation), a relevant scenario is that of a steel pipeline of 25-mm wall thickness, insulated with concrete and which is to be scanned to identify regions of corrosion. Using the experimental data of Fig. 5a (red dotted curves) of such a 25-mm pipeline, an exemplar pipe tomography study has been conceived and is presented in Fig. 6. Two different regions of different thicknesses have been inserted deliberately, and positioned randomly in the pipeline wall to emulate regions of corrosion. The first region is 5 mm thick and the second region 20 mm (Fig. 6a, denoted by numbers 2 and 3). The experimental backscattered flux from each of these features has been reproduced and imaged, together with the backscattered flux of the original 25-mm thick pipe. Figures 6a-c show the fast neutron backscatter tomography (FN-BCT), the γ-BCT and the combined n-γ BCT, respectively, for this case. The 5-and 20-mm regions, that render the steel respectively 20-and 5-mm thick, are clearly discernible using fast neutrons, whereas the 5-mm pit-area is not easily-discernible scanning the pipe with γ rays in isolation. Combining the two different imaging modalities, it is still possible to identify the two areas of reduced thickness. However, this scenario can arise in reverse, that is, depending on which materials are scanned, the neutron tomography data that results might mislead, in contrast to the γ-ray case, as explained in previous work for transmission tomography 47,48 . . Pulse shape discrimination. Left: A scatter plot (peak amplitude versus discrimination amplitude) of the γ radiation generated by 137 Cs with an EJ-301 detector in this research. The peak and discrimination amplitudes are, respectively, the maximum signal amplitude and the signal magnitude measured after a fixed discrimination-time. The top-inner plot is the Compton spectrum when an EJ-301 is exposed to γ rays from 137 Cs, whereas the bottom-inner histogram is the discrimination value of the γ-event produced with 137 Cs. Right: A scatter plot obtained with an EJ-301 scintillation detector exposed to 252    www.nature.com/scientificreports www.nature.com/scientificreports/ wall thicknesses will have similarly-contrasting gradients or count rates. In this particular circumstance, after a given observation time t has elapsed, the error (σ) on the counted number of events (N) is the square root (σ = √N), assuming Poisson statistics. Equation 4 describes the minimum time needed to differentiate two different thicknesses of the same material, within a sensitivity of n σ = 1, 2, 3… standard deviations, where N 1 and N 2 are the counts, at a given time t, of the two thicknesses. An illustrative example with respect to this sensitivity formulism is given in Fig. 7a: Here, Eq. 4 has been used to construct a sensitivity matrix for neutrons and γ rays, for the system tested with bare steel (Fig. 7b), steel with polyethylene (Fig. 7c) and steel with concrete (Fig. 7d). The left side of the matrices in Fig. 7b-d corresponds to γ rays and the right to neutrons. Two different colour maps for γ rays and neutrons have been used to separate the sensitivity of each, and a 900-second cut-off on the minimum measurement time necessary to distinguish  www.nature.com/scientificreports www.nature.com/scientificreports/ two thicknesses has been set deliberately, on the basis of what is anticipated to be an acceptable limit in the field. Generally speaking, times lower than this value allow the discrimination of difference in thicknesses from 5 mm and upwards. For the identification of possible pits of less than 5 mm depth, the measurement time required increases exponentially, at which point this technique starts to become impractical. The sensitivity matrix presented here has been calculated with the experimental data presented in Fig. 5a, using a 252 Cf source with an emission rate of 8.7 × 10 6 neutrons/second into 4π. The sensitivity can be improved and thus the experimental exposure time reduced using a source of higher activity. Alternatively, shielding the detectors in order to reduce false-negative scattering events due to background and cross-talk between detectors could also improve the sensitivity.

Discussion
The results presented in this paper demonstrate that it is possible to discern different thicknesses of steel slab with a combination of fast neutron and γ-ray backscattering. Our research was carried out using organic liquid scintillators, although a diversity of organic scintillators exists which could be similarly applied; for example stilbene might constitute a valid alternative to the EJ-301 used in this research if a liquid scintillant is not desirable in the application field. Organic scintillators, coupled with the real-time PSD system used in this research, are particularly suitable because their detection efficiency falls sharply for energies below ~1 MeV for neutrons and below 200 keV for γ rays. The elastic scattering cross sections for neutrons in the range 1-10 MeV, are of the same order of magnitude for the majority of elements (between 1 barn and 10 barns). However, since the neutron energy loss after an elastic collision is far greater for low-atomic number materials, incident fast neutrons can fall below the detection energy threshold of the scintillator detectors when scattered. This fact, for example, explains the difference in slope in the responses for steel and steel-polyethylene for neutrons in Fig. 5a,b, since HDPE is rich in hydrogen relative to concrete, and thus moderates neutrons more effectively. For small thicknesses (i.e., ≤15 mm), the polyethylene-induced backscattering is higher than that measured with bare steel. As the steel thickness is increased, the number of elastic neutron collisions also increases; therefore, before reaching the detector, the backscattered neutrons pass through an extra thickness of polyethylene, accruing a higher probability of falling below the energy threshold for detection, and thus reducing the proportion of the backscattered component that is detected. Conversely, this does not occur for γ rays because Compton scattering depends on the atomic number of the material, which is relatively low for polyethylene compared to that of steel, and consequently, the gradient of the γ-ray backscatter count dependence with thickness is similar for all three sample arrangements. Our results are consistent with what is predicted on a qualitative basis by the MCNP6 simulations for both neutrons and γ rays. The mathematical relationship elaborated for neutrons overlaps the neutron experimental data for the case of bare-steel, in the range 6-25 mm, when results are presented within ±3σ from the mean.
The novelty of this research lies, in primis, in the demonstration of a non-destructive imaging alternative to the widespread modality of X-ray computed tomography. Moreover, this technique comprises the parallel application of both neutrons and γ rays, leading to three different final illustrations of a given sampler (i.e., n-, γand n-γ-BCT). Read together, these yield a more comprehensive and faster representation of the inner structure of steel and possibly, other materials. Our research highlights and confirms the potential of combining different imaging modalities. Not only does this technique have applications in an engineering context, but it may also have potential in wider materials science applications such as for quality assessments of metals and materials, and also in a wide range of different scenarios, ranging from the medical field (as proposed by 49 ) to safety and security inspections, and particularly where in situ examinations are required. This research not only highlights the benefit of combining the effects of contrasting reflection phenomena for technological requirements, such as non-intrusive corrosion assessment; it also illustrates the significant potential that can accrue from our primitive mimicry of sensing modalities that have evolved for analogous requirements in the natural world.
Methods the mechanical rig. The rig frame is made from an assembly of 20 × 20 mm aluminium extrusions, it has dimensions 600 × 400 × 340 mm (L × W × H). The collimator sits within the aluminium frame and is mounted to four guide rails, two in both the X-and Y-axis, respectively. This gives the collimator the freedom of movement in the X-Y plane, actuated using a stepper motor and pulley system. A symmetrical array of detectors are mounted to a cylindrical assembly using Go-Pro arms. The assembly is made from eight aluminium extrusions around 360° with aluminium plates top and bottom; the cylindrical components of the collimator itself sit flush within the extrusion assembly. The entire assembly has a height of 311 mm and a diameter of 140 mm. On the top plate of the assembly, a bespoke 3D-printed component is mounted to house the isotopic source directly above the collimator void. The rig is controlled using an Arduino ® microcontroller board which interfaces with the user's device via USB. All electronic components on the rig are controlled by the Arduino ® which receives commands from the user. The user specifies coordinates relating to a position in the X-Y plane, the Arduino ® then handles the calculation necessary to get to the desired position. Limit switches at the end of each axis are used to calibrate the positioning of the collimator assembly as well as a fail-safe to prevent it from driving off the rails.
control system and counter. The acquisition system consists of a printed circuit board (PCB) that contains an Intel Cyclone V FPGA/ARM Processor system. A set of sixteen 32-bit transistor-transistor-logic (TTL) compatible counters were configured on the FPGA. The system uses a 50 MHz clock signal and Phase-locked Loop (PLL) logic allowing pulses of less than 1 μs width to be detected. The 4-channel MFA produces TTL signals from separate output ports dependent on whether a neutron or a γ ray has been detected. The 4 γ-ray outputs and the 4 neutron outputs from the MFA were connected up to the aforementioned 16-channel counter (thus up to 8 detectors can be used: 8 channels for neutron detection and 8 channels for γ-ray detection). The Cyclone V ARM Processor system runs an embedded version of Linux capable of interfacing to the described logic in the Detector stability, flux evaluations and background level. The whole experimental set-up was assembled approximately 20 hours prior to the start of the experimental measurements. Detectors, mixed field analyzer, electronics, embedded hardware and control software were turned on and their functioning monitored and verified. The stability of the detectors, with and without neutron and γ irradiation, was also demonstrated both in the laboratory at Lancaster University and at the National Physical Laboratory. Within 48 hours, during the preliminary tests carried out at Lancaster University, the counting rate of the detectors were observed to be constant over elapsed time. The baseline level of neutron and γ-ray background at the low-scatter facility was measured, as well as the background level induced by the 252 Cf source in the room. Albeit being a low-scatter facility, a low-level of background is present due to the interaction of the radiation with air, room walls, laboratory and experimental components. This background was subtracted from the readings of the detectors when performing the data analysis. Finally, the mixed radiation flux from the collimator (Φ 0 , see Eq. 3) was evaluated carefully for each individual detector. This value has been 10 times higher than the value of the background induced by the californium source and measured by the detectors when placed in the position hidden by the collimator (see, for instance, Figs. 1b and 2a), confirming experimentally the validity of the simulations and collimator function.
MCNP6 simulation. The final design of the collimator is the result of a detailed MCNP6 simulation study that has been performed in order to optimise the system geometry, materials and system characteristics. The collimator is a cylinder of 10-cm diameter and 30-cm length. It has a 1-cm diameter pinhole which allows the passage of the mixed-field radiation produced by the aforementioned radiation source. Three layers of lead (respectively of 3 cm, 1 cm and 2 cm) and two layers of high-density polyethylene (12 cm thickness each) shield the detectors from the radiation emitted by the source. The experiment carried out has been modelled as accurately as possible with MCNP6 simulations. Six steel thicknesses (from 5 mm to 30 mm) of a pipeline have been reproduced in this way. Neutrons and γ rays for scattered events from the pipeline wall have been tallied simulating the EJ-301 detectors and its liquid scintillant. The collimator and 252 Cf source have also been modelled in the experiment-simulation. The steel (0.3% carbon component), concrete (Hanford type, dry) and polyethylene model details are listed in the Radiation Portal Monitor Project, Compendium of Material Composition Data for Radiation Transport Modelling 51 .

Data availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.