Tomographic Reconstruction of Triaxial Strain Fields from Bragg-Edge Neutron Imaging

06/20/2019 ∙ by J. N. Hendriks, et al. ∙ 0

This paper presents the first triaxial reconstruction of strain in three-dimensions from Bragg-edge neutron imaging. Bragg-edge neutron transmission can provide high-resolution tomographic images of strain within a polycrystalline material. This poses an associated rich tomography problem which seeks to reconstruct the full triaxial strain field from these images. Successful techniques to perform this reconstruction could be used to study the residual strain and stress within engineering components. This paper uses a Gaussian process based approach that ensures the reconstruction satisfies equilibrium and known boundary conditions. This approach is demonstrated experimentally on a non-trivial steel sample with use of the RADEN instrument at the Japan Proton Accelerator Research Complex. Validation of the reconstruction is provided by comparison with conventional strain scans from the KOWARI constant-wavelength strain diffractometer at the Australian Nuclear Science and Technology Organisation and simulations via finite element analysis.



There are no comments yet.


page 4

page 5

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Bragg-edge neutron transmission techniques provide a means for obtaining lower dimensional (one- or two-dimensional) tomographic strain images from higher dimensional (two- or three-dimensional) strain fields within polycrystalline materials. The success of these techniques, and the development of associated instruments and detectors has prompted research into the tomographic reconstruction of strain — i.e. strain tomography. This research aims to provide methods analogous to conventional Computed Tomography (CT) whereby the complete triaxial strain distribution within a sample could be reconstructed from a sufficient set of strain images. As the strain field is a tensor, this is a more complex task than conventional scalar CT.

Tomographic reconstruction of strain fields from these images can be used to study the residual elastic strain (and hence stress) within engineering components. Residual stresses are those which remain after applied loads are removed (e.g. due to heat treatment, or plastic deformation), and may have a significant and unintended impact on a component’s effective strength and service life (in particular fatigue life). A full field analysis of these stress and strains could have a significant impact on several areas of experimental mechanics. In particular, it could be used to study the residual stress within additively manufactured, laser-clad, preened, welded, cast, forged and/or otherwise processed components. This full field analysis will have significant advantages over current destructive and semi-destructive techniques (Prime, 2001; ASTM Standard, ), and point-wise x-ray and neutron diffraction methods (Hauk, 1997; Noyan and Cohen, 1987; Fitzpatrick and Lodini, 2003).

Strain tomography falls into the class of ‘rich’ tomography problems where the tomographic strain image is related to an unknown tensor field. The acquisition and analysis of these strain images is described in detail elsewhere (Santisteban et al., 2002a, 2001, b). Briefly summarising the process: the relative transmission of neutron pulses with known wavelength-intensity spectra through the sample is measured at a pulsed neutron source (e.g. the Japan Proton Accelerator Research Complex (J-PARC) in Japan, ISIS in the United Kingdom, or the Spallation Neutron Source in the USA). For example, current state-of-the-art Micro-Channel Plate detectors (Tremsin et al., 2011) are capable of measuring the transmitted spectra over an array of pixels with a resolution of . From this data, the position of a given Bragg-edge (a sudden increase in transmitted intensity as a function of wavelength) is observed. The relative position of a Bragg-edge provides a measure of strain at each pixel of the form


where is the wavelength at which the Bragg-edge occurs, is the corresponding Bragg-edge wavelength in a stress-free sample, and the following applies:

  1. As with all diffraction measurements, only the elastic component of strain is measured.

  2. The measured strain is the normal component in the direction of the transmitted neutron beam.

  3. The measurement corresponds to a through-thickness average along the path of the corresponding ray.

The strain measured at each pixel can be related to the strain field using the Longitudinal Ray Transform (LRT) (Abbey et al., 2009; Lionheart and Withers, 2015). With the inclusion of measurement error this gives a measurement model as


where the LRT geometry is defined in Figure 1 and

is the measurement error term, which is assumed to be zero-mean Gaussian with standard deviation

. Estimating the strain field given a set of LRT measurements is made more complex as the LRT mapping is non-injective

(Lionheart and Withers, 2015; Sharafutdinov, 1994). This means that if the strain field components are considered independent, then infinitely many fields could produce the same set of measurements.


Figure 1: A two-dimensional representation of the Longitudinal Ray Transform. A ray of direction enters the sample at and has a through-thickness length of . The ray represents the path taken through the sample by neutrons arriving at the pixel with which the measurement is associated.

This problem has been overcome in a number of ways, allowing the successful reconstruction of two-dimensional strain fields. Several special cases have been considered including axisymmetric systems (Abbey et al., 2009, 2012; Kirkwood et al., 2015; Gregg et al., 2017) and granular systems (Wensrich et al., 2016a). More arbitrary strain fields caused by in-situ loadings have been successfully reconstructed by assuming compatibility (Wensrich et al., 2016b; Hendriks et al., 2017). Methods for reconstructing a broader class of strain fields (i.e. residual strains) have been developed by ensuring the strain field satisfies equilibrium (Gregg et al., 2018; Jidling et al., 2018; Hendriks et al., in press).

Recently, an approach suitable for modelling and estimating three-dimensional strain fields has been presented for simulated data (Hendriks et al., 2019). While this method deals with the reconstruction of strain from high-energy x-ray diffraction measurements, it requires only limited modification to be applied to the problem presented here.

1.1 Contribution

This paper details the first experimental demonstration of tri-axial strain field reconstruction using Bragg-edge neutron transmission strain imaging. These strain images were acquired during an experiment at J-PARC (January 2019) of a hollow EN26 steel cube sample loaded by an interference fit with a titanium plug. The reconstruction approach follows the method presented in Hendriks et al. (2019) which was previously demonstrated on simulated high-energy x-ray strain measurements. This method is adapted for the LRT measurement model and to include boundary conditions. The resulting reconstruction is compared with conventional constant-wavelength strain measurements performed with the KOWARI engineering diffractometer at the Australian Nuclear Science and Technology Organisation (ANSTO) as well as a Finite Element Analysis (FEA) model.

2 Reconstruction Approach

In this paper, we present the first practical demonstration of a method capable of reconstructing the full tri-axial strain field from Bragg-edge neutron transmission measurements. The reconstruction approach is modified from the method presented by Hendriks et al. (2019); which was used to reconstruct 3D strain field from simulated high-energy x-ray measurements. This approach models the strain field by a Gaussian process (Rasmussen and Williams (2006) provides a good introduction), and ensures that the reconstructed strain field satisfies the physical properties of equilibrium; assuming the sample to be linearly elastic and isotropic (i.e. without texture). Ensuring the strain field satisfies equilibrium is critical as the LRT mapping (2) has a non-trivial null space (Lionheart and Withers, 2015) (i.e. without these constraints a unique solution to the inverse problem does not exist). By enforcing equilibrium the null space is reduced to contain only the trivial field, allowing a unique solution to be (Hendriks et al., in press).

Gaussian processes are suitable for the modelling and estimation of spatially correlated phenomena. The use of Gaussian processes to model and estimate strain fields was pioneered in Jidling et al. (2018). By modelling the Airy stress function (a scalar potential field) by a Gaussian process a solution to the two-dimensional stress (and hence strain) could be given that satisfies equilibrium in the absence of body forces. This method can be extended to three dimensions for which the Beltrami stress functions are used instead of the Airy stress function. The Beltrami stress functions consist of six unique potential fields from which a complete solution to the equilibrium equations in three-dimensions can be given (Sadd, 2009). These potential fields are each modelled by a Gaussian process allowing a tri-axial strain field satisfying equilibrium to be reconstructed.

In addition to equilibrium, boundary conditions can also be included following the work by Hendriks et al. (in press). Knowledge of unloaded surfaces for which the distribution of applied forces (known as tractions) is known to be zero can be incorporated. This can be done by including artificial measurements of zero traction at points on the surface not subject to an applied load.

A detailed description of the method and its implementation is given by Hendriks et al. (2019). This requires only minor modification to the measurement model, and the inclusion of traction measurements. The measurement model requires a slight modification due to the difference between high-energy x-ray and neutron transmission strain measurements. For high-energy x-ray strain measurements, the measured strain direction, denoted , is almost perpendicular to the ray direction . Whereas, for neutron transmission strain measurements the direction of measured strain is aligned with the ray, and so . Additionally, a large variation in strain measurement uncertainty is observed and therefore the method is modified so that each measurement can be assigned its own standard deviation. In essence this weights the importance of each measurement according to it’s confidence given by the strain imaging process.

In this paper, this approach is used to reconstruct the strain field within an EN26 steel sample from a set of strain images measured at J-PARC. The sample, load case, and strain image acquisition are described in Section 3.1. The resulting strain field is validated by comparison with conventional diffraction strain scans and FEA results, which are described in Section 3.2. The reconstructed strain field and a comparison with the validation data is given in Section 3.3; potential sources of error are also discussed in this section.

3 Demonstration

3.1 Sample Design and Strain Imaging

The method is applied to a real world example using a data set collected on the RADEN energy-resolved-neutron-imaging instrument at J-PARC (Shinohara and Kai, 2015; Shinohara et al., 2016). The experiment tomographically imaged the strain within an EN26 steel (medium carbon, low alloy) sample. The sample consisted of a steel cube with a precision ground hole of diameter along the diagonal. A load was applied by a interference fit (i.e. shrink fit) with a titanium plug. The sample and plug are shown in Figure 2.

[width=0.225]sample_plot_coords.pdf [width=0.225]plane1_diagram.jpg [width=0.225]plane2_diagram.jpg [width=0.225]plane3_diagram.jpg

Figure 2: Left: The sample (dark grey) and plug (light grey) assembly, as well as the coordinate systems used. Middle Left: Cross section plane 1 used for validation with KOWARI and FEA. Middle Right: Cross section plane 2 used for validation against KOWARI and FEA. Right: Cross section plane plane 3 used for validation against FEA. For section planes one and two, the location of the KOWARI measurements are shown in orange. Note that gauge volume orientation shown is indicatively only, as it varies for each component of strain measured.

Prior to assembling, the sample was heat treated to relieve stress and provide a uniform tempered-martensite (i.e. ferritic) structure with minimal texture, and with a final hardness of . The sample was assembled by first inserting the plug into a cylinder with an interference fit of , after which a cube with sides was milled from the cylinder and plug.

This sample and loading set-up was designed to provide a smooth three-dimensional strain field suitable for the first demonstration of three-dimensional strain tomography. This has additional advantages for validation as FEA of strain resulting from interference fits is a more straight forward process than FEA of strain fields resulting from plastic deformation.

To this end, a titanium plug was chosen as strain within the plug was ‘invisible’ to the strain imaging process. This is because titanium does not have a Bragg-edge near to the steel Bragg-edge that was chosen for analysis. As a result, the titanium plug can be ignored during the reconstruction process and considered as imparting an in-situ load on the interior face of the hollow cube.

Strain images were measured on the RADEN instrument using a micro-channel plate detector ( pixels, per pixel)(Tremsin et al., 2011) at a distance of from the source. At the time of the experiment (January 2019) the source power was . Pixels were grouped together into macro pixels of giving sufficient neutron counts to provide a reasonable edge fit; giving a final strain image of macro pixels, each with a resolution of . It is important to note that this does not directly correspond to the resolution of the final reconstruction. The Gaussian process method used reconstructs a continuous (smooth) strain field where the maximum rate of change is automatically adapted to suit the available data.

Each macro pixel provides a strain measurement of the form (1) where the Bragg-edge position was found following the procedure given by Santisteban et al. (2001) applied to the (110) Bragg-edge. The undeformed location was determined from a stress-free sample. The resulting strain measurements had, on average, an uncertainty of standard deviation . This uncertainty is higher than previously achieved in two-dimensional strain tomography experiments (Hendriks et al., 2017; Gregg et al., 2018), however further increasing the macro pixel size did not provide a sufficient decrease in uncertainty to warrant the loss in strain image resolution. A systematic bias in the edge fit as a function of the measurement path length was observed. This effect was previously described in Vogel (2000) and Gregg et al. (2018); although the exact mechanism is yet to be established. Following Gregg et al. (2018), a linear correction was applied to .

In total 70 tomographic strain images were collected. For these images the sample was positioned using a two axis goniometer, which allowed rotation in azimuth and elevation. Limitations of the experimental set-up restricted the achievable elevation angles to the range from to . Therefore, in order to cover the full range of measurement directions, the sample was repositioned by rotating about the -axis for the final 11 strain images.


(a) Sample positioning


(b) Bragg-edge height
Figure 3: (fig:sample_positioning) Sample, detector, goniometer, and beam direction are shown. The geniometer enabled rotation in azimuth and elevation. (fig:edge_height) Bragg-edge height map for a projection that aligned the beam direction and the plug. The -axis of the sample was aligned with the beam and the -axis aligned with vertical. This illustrates that the titanium plug does not contribute to the Bragg-edge chosen for analysis.

The exact geometry of each neutron ray passing through the sample before reaching the detector is required to model the measurements by the LRT (2). In addition to designing a sample holder to carefully position the sample, an optimisation routine was run to determine the remaining orientation offsets and the offsets between the centre of rotations. The optimisation maximised the sum of Bragg-edge heights associated with rays that would pass through the sample for a given choice of offsets.Gregg et al. (2018)

Strain fields were reconstructed from this set of strain images using the Gaussian process method described earlier with the inclusion of 400 measurements of zero traction evenly distributed on each of the exterior faces.

3.2 Validation Data

Validation relies on comparison with conventional strain scans (Kisi and Howard, 2012; Fitzpatrick and Lodini, 2003; Noyan and Cohen, 1987) from the KOWARI constant-wavelength strain diffractometer at the Australian Nuclear Science and Technology Organisation (Kirstein et al., 2009; Brule and Kirstein, 2006)

and Finite Element Analysis (FEA). The strain scans provide measurements of the 6 components of strain on two section planes (33 points on plane 1 and 45 points on plane 2). As with all diffraction methods these measurements correspond to the average strain inside gauge volumes. These gauge volume locations were chosen on two section planes that were expected to exhibit strong skew symmetry and therefore help to validate a larger region of the reconstruction. The section planes and gauge volume locations are shown in Figure 


These measurements were based on the relative shift of the (211) diffraction peak measured with neutrons of wavelength ( geometry) and a gauge volume. The and lattice planes have effectively the same diffraction elastic constants (Daymond and Priesmeyer, 2002), therefore the results from the transmission and diffraction experiments can be directly compared without rescaling or recalculation to stress. Sampling times with the KOWARI diffractometer were based on providing uncertainty in strain around : which required 60 hours. Together with sample set-up and alignment, a total of 4 days of beam time was required. The long sampling times required for the comparatively small gauge volumes meant that only a portion of each section plane could be measured.

While only a small amount of the strain field can be verified using the strain scanning measurements, the full reconstructed strain field can be compared to FEA results to provide further validation. Comparison of the reconstruction to the FEA results is made for the three section planes shown in Figure 2.

3.3 Results and Discussion

The reconstructed strain field is shown together with strain scans made on the KOWARI diffractometer and FEA results in Figures 4 and 5. As the KOWARI strain scans cover only a region of section plane 1 and 2, comparison between all three sets is made in these regions only. Example Matlab code to perform this reconstruction and produce the strain plots shown in this paper is available on Github (Hendriks, 2019).


Figure 4: Comparison of the KOWARI strain scans, the reconstruction from RADEN strain images, and FEA results. Shown is the region on section plane 1 for which KOWARI strain scans were made. The KOWARI strain scans correspond to measured averages within the gauge volumes and so are shown as a constant value within representative gauges to best reflect this.


Figure 5: Comparison of the KOWARI strain scans, the reconstruction from RADEN strain images, and FEA results. Shown is the region on section plane 2 for which KOWARI strain scans were made. The KOWARI strain scans correspond to measured averages within the gauge volumes and so are shown as a constant value within representative gauges to best reflect this.

Comparison of the KOWARI strain scans, the RADEN reconstruction and the FEA results shows good agreement in general. However, some specific differences can be noted:

  • In the component of plane 1, the reconstruction shows a region of tension in the bottom right that is also present in the strain scans but is not seen in the FEA.

  • In the and components of plane 1, the reconstructed strain has the same shape but a greater magnitude than the FEA strain and it is not clear whether this is supported in the strain scans.

  • In the component of plane 2, the reconstructed strain field has a region of compression also present in the FEA but not seen in the strain scans.

When making these comparison it is important to remember that the KOWARI strain scans are themselves measurements which are relatively noisy ( standard deviation) and are averages over the gauge volumes; this in some cases makes comparison difficult. In particular, the and components of plane 1, and the and components of plane 2 appear particularly noisy.

The FEA strain fields and the reconstructed strain fields are available for the entire sample allowing comparison over the entire section planes 1, 2, and 3; these are shown in Figure 6, Figure 7 and Figure 8 respectively. A selection of strain components is shown for each section plane so that all components are shown at least once. This comparison shows that the reconstruction shows close agreement with the FEA strain fields on plane 2 and plane 3. The shape of the reconstructed strain field is very similar with the main difference being noted in slightly reduced magnitudes and the peak strains are less concentrated. There is a slightly greater difference between the reconstruction and FEA results observable in plane 1; particularly in the and components. However, at least for the component some of this difference is supported by the KOWARI strain scans which also showed the region of tension present in the bottom right. Additionally, as the strain fields are skew symmetric this would lend some support to the region of tension present in the top left of this component.


Figure 6: The strain field reconstructed from the RADEN strain images and the FEA strain field for section plane 1.


Figure 7: The strain field reconstructed from the RADEN strain images and the FEA strain field for section plane 2.


Figure 8: The strain field reconstructed from the RADEN strain images and the FEA strain field for section plane 3.

From these results a quantitative assessment of the discrepancies between the the KOWARI strain scan measurements and the reconstruction as well as between the reconstruction and the FEA results was carried out. In both cases the differences are mean zero and Gaussian, implying that there is no systematic error or bias resulting from the reconstruction technique. The differences between the KOWARI strain scans and the reconstruction are calculated by taking gauge volume style averages of the reconstructed strain field and comparing these to the KOWARI measurements. The resulting differences have an average magnitude of . The discrepancies between the FEA and reconstructed strain fields have an average magnitude of .

Although the reconstruction shows, in general, good agreement with the FEA results and KOWARI strain scans some differences have been noted and these differences may be attributed to several sources. Firstly, the number and quality of strain measurements acquired is less than has been previously achieved in two-dimensional strain tomography experiments. In total, 14000 strain measurements were acquired with an average standard deviation of compared to approximately with a standard deviation of in previous two-dimensional experiments (Hendriks et al., 2017; Gregg et al., 2018).

This is in part due to the trade off between macro pixel size and uncertainty in the Bragg-edge fits. While smaller macro pixels ( pixels) would give better resolution in the strain image, the measurement standard deviations would be increased to around . Conversely, larger macro pixels () could be used to decrease the measurement standard deviation to around , however the resolution in the strain image would be made worse. This is in contrast to two-dimensional geometry where the assumption of no out of plane strain variation meant the pixels could be binned into columns without affecting the resolution of the resulting one-dimensional strain image.

The effect of this is particularly significant in regions near the sample boundary – i.e. in the corners where the plug and sample intersect. These regions contain a small amount of material and so the averaging effect of the LRT means that the strains are poorly sampled. Additionally, the smaller amount of material means that the Bragg-edge height of any strain measurements passing predominately through these regions is reduced, resulting in poorer measurement confidence. When combined with the macro pixel averaging, this results in the peak strains in these regions being somewhat obscured. This issue may be somewhat alleviated in future experiments at J-PARC with an expected source power increase to over the next few years.

Differences between reality and the FEA model may also account for some of the observed discrepancies. The sample was milled into a cube from a typical ring and plug. During this process it was not possible to ensure the plug was perfectly on the diagonal of the cube. This would account for the measured strain fields (KOWARI and RADEN) not being entirely skew symmetric and for some of the difference between these fields and the FEA strain field. Additionally, the milling process itself may have introduced residual stresses not accounted for in the FEA model. Finally, the peak stresses are around the yield strength of the material which could result in effects such as hardening and account for differences between the measured strain fields and the FEA strain field.

4 Conclusion and Future Work

This paper has presented the first three-dimensional full-field strain reconstruction from neutron transmission strain images. Strain images were collected using the RADEN energy-resolved-neutron-imaging instrument at J-PARC. The reconstructed strain field was validated by comparison with conventional strain scans from the KOWARI diffractometer and FEA results and shows very promising results.

The reconstruction was performed using a Gaussian process based method that ensures the resulting strain field satisfies equilibrium. This is achieved by using the Beltrami stress functions to provide a complete solution to the stress (and strain) fields in three-dimensions. Such an approach provides a continuous strain field throughout the entire sample, whereas conventional strain scanning methods may struggle to cover a reasonable portion of a sample considering the implication of small gauge volumes required for three-dimensional geometry.

The reconstructed strain field was developed within a hollowed EN26 cube by an ‘in-situ’ loading created by interference fitting a titanium plug. Although this strain field is compatible, the method is applicable to a broader class of problems (i.e. residual strains) as it makes no assumption of compatibility. To this end, future work involves the planning of a three-dimensional residual strain experiment. This could also involve adapting the Gaussian process model to be more suitable for strain fields exhibiting rapid changes or discontinuities.

Additionally, it was noted that the strain measurements were fewer and of poorer quality than what has been previously achieved and that this may have affected the accuracy of the reconstruction. Although this issue may be somewhat alleviated by an increase in source power, future work will also investigate full pattern fitting methods (Luzin et al., 2011; Sato et al., 2013, 2017) which could provide better measurement statistics by analysing multiple Bragg-edges. Full pattern fitting may also provide a path to extending the method to samples that contain significant texture.

Finally, future work should investigate methods for validating the results when other data sets are not available; such as cross validation (Devijver and Kittler, 1982).


This work was supported by the Australian Research Council through a Discovery Project Grant No. DP170102324. Access to the RADEN and KOWARI instruments was made possible through the respective user-access programs of J-PARC and ANSTO (J-PARC Long Term Proposal No. 2017L0101 and ANSTO Program Proposal No. PP6050). The authors would also like to thank AINSE Limited for providing financial assistance (PGRA) and support to enable work on this project.


  • Abbey et al. (2009) Brian Abbey, Shu Yan Zhang, Wim JJ Vorster, and Alexander M Korsunsky. Feasibility study of neutron strain tomography. Procedia Engineering, 1(1):185–188, 2009.
  • Abbey et al. (2012) Brian Abbey, Shu Yan Zhang, Wim Vorster, and Alexander M Korsunsky. Reconstruction of axisymmetric strain distributions via neutron strain tomography. Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 270:28–35, 2012.
  • (3) ASTM Standard. Standard test method for determining residual stresses by the hole-drilling strain-gage method,”. E837, 2002.
  • Brule and Kirstein (2006) Alain Brule and Oliver Kirstein. Residual stress diffractometer kowari at the australian research reactor opal: status of the project. Physica B: Condensed Matter, 385:1040–1042, 2006.
  • Daymond and Priesmeyer (2002) MR Daymond and HG Priesmeyer. Elastoplastic deformation of ferritic steel and cementite studied by neutron diffraction and self-consistent modelling. Acta materialia, 50(6):1613–1626, 2002.
  • Devijver and Kittler (1982) Pierre A Devijver and Josef Kittler. Pattern recognition: A statistical approach. Prentice hall, 1982.
  • Fitzpatrick and Lodini (2003) Michael E Fitzpatrick and Alain Lodini. Analysis of residual stress by diffraction using neutron and synchrotron radiation. CRC Press, 2003.
  • Gregg et al. (2017) AWT Gregg, JN Hendriks, CM Wensrich, and MH Meylan. Tomographic reconstruction of residual strain in axisymmetric systems from bragg-edge neutron imaging. Mechanics Research Communications, 85:96–103, 2017.
  • Gregg et al. (2018) AWT Gregg, JN Hendriks, CM Wensrich, Adrian Wills, AS Tremsin, Vladimir Luzin, Takenao Shinohara, Oliver Kirstein, MH Meylan, and EH Kisi. Tomographic reconstruction of two-dimensional residual strain fields from bragg-edge neutron imaging. Physical Review Applied, 10(6):064034, 2018.
  • Hauk (1997) Viktor Hauk. Structural and residual stress analysis by nondestructive methods: Evaluation-Application-Assessment. Elsevier, 1997.
  • Hendriks (2019) J.N. Hendriks. Three dimensional strain tomography example code. Github, 2019. URL
  • Hendriks et al. (2017) JN Hendriks, AWT Gregg, CM Wensrich, AS Tremsin, T Shinohara, M Meylan, EH Kisi, V Luzin, and O Kirsten. Bragg-edge elastic strain tomography for in situ systems from energy-resolved neutron transmission imaging. Physical Review Materials, 1(5):053802, 2017.
  • Hendriks et al. (2019) JN Hendriks, CM Wensrich, and A Wills. A bayesian approach to triaxial strain tomography from high-energy x-ray diffraction. arXiv preprint arXiv:1903.02158, 2019.
  • Hendriks et al. (in press) J.N. Hendriks, A.W.T. Gregg, C.M. Wensrich, and A. Wills. Implementation of Traction Constraints in Bragg-edge Neutron Transmission Strain Tomography. Strain, in press.
  • Jidling et al. (2018) Carl Jidling, Johannes Hendriks, Niklas Wahlström, Alexander Gregg, Thomas B. Schön, Christopher Wensrich, and Adrian Wills. Probabilistic modelling and reconstruction of strain. Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 436:141 – 155, 2018. ISSN 0168-583X. doi: URL
  • Kirkwood et al. (2015) Henry J Kirkwood, Shu Yan Zhang, Anton S Tremsin, Alexander M Korsunsky, Nikolaos Baimpas, and Brian Abbey. Neutron strain tomography using the radon transform. Materials Today: Proceedings, 2:S414–S423, 2015.
  • Kirstein et al. (2009) O Kirstein, V Luzin, and U Garbe. The strain-scanning diffractometer kowari. Neutron News, 20(4):34–36, 2009.
  • Kisi and Howard (2012) Erich H Kisi and Christopher J Howard. Applications of neutron powder diffraction, volume 15. Oxford University Press, 2012.
  • Lionheart and Withers (2015) William RB Lionheart and Philip J Withers. Diffraction tomography of strain. Inverse Problems, 31(4):045005, 2015.
  • Luzin et al. (2011) V Luzin, K Spencer, and M-X Zhang. Residual stress and thermo-mechanical properties of cold spray metal coatings. Acta Materialia, 59(3):1259–1270, 2011.
  • Noyan and Cohen (1987) Ismail C Noyan and Jerome B Cohen. Determination of strain and stress fields by diffraction methods. In Residual Stress, pages 117–163. Springer, 1987.
  • Prime (2001) Michael B Prime. Cross-sectional mapping of residual stresses by measuring the surface contour after a cut. Transactions-American society of mechanical engineers journal of engineering materials and technology, 123(2):162–168, 2001.
  • Rasmussen and Williams (2006) Carl Edward Rasmussen and Christopher KI Williams.

    Gaussian processes for machine learning

    , volume 1.
    MIT press Cambridge, 2006.
  • Sadd (2009) Martin H Sadd. Elasticity: theory, applications, and numerics. Academic Press, 2009.
  • Santisteban et al. (2001) JR Santisteban, L Edwards, A Steuwer, and PJ Withers. Time-of-flight neutron transmission diffraction. Journal of applied crystallography, 34(3):289–297, 2001.
  • Santisteban et al. (2002a) JR Santisteban, L Edwards, ME Fitzpatrick, A Steuwer, PJ Withers, MR Daymond, MW Johnson, N Rhodes, and EM Schooneveld. Strain imaging by bragg edge neutron transmission. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 481(1):765–768, 2002a.
  • Santisteban et al. (2002b) JR Santisteban, L Edwards, ME Fizpatrick, A Steuwer, and PJ Withers. Engineering applications of Bragg-edge neutron transmission. Applied Physics A, 74(1):s1433–s1436, 2002b.
  • Sato et al. (2013) H Sato, T Shinohara, R Kiyanagi, K Aizawa, M Ooi, M Harada, K Oikawa, F Maekawa, K Iwase, T Kamiyama, et al. Upgrade of bragg edge analysis techniques of the rits code for crystalline structural information imaging. Physics Procedia, 43:186–195, 2013.
  • Sato et al. (2017) H Sato, K Watanabe, K Kiyokawa, R Kiyanagi, KY Hara, T Kamiyama, M Furusaka, T Shinohara, and Y Kiyanagi. Further improvement of the rits code for pulsed neutron bragg-edge transmission imaging. Physics Procedia, 88:322–330, 2017.
  • Sharafutdinov (1994) Vladimir A Sharafutdinov. Integral geometry of tensor fields, volume 1. Walter de Gruyter, 1994.
  • Shinohara et al. (2016) T Shinohara, T Kai, K Oikawa, M Segawa, M Harada, T Nakatani, M Ooi, K Aizawa, H Sato, T Kamiyama, et al. Final design of the energy-resolved neutron imaging system “raden” at j-parc. In Journal of Physics: Conference Series, volume 746, page 012007. IOP Publishing, 2016.
  • Shinohara and Kai (2015) Takenao Shinohara and Tetsuya Kai. Commissioning start of energy-resolved neutron imaging system, raden in j-parc. Neutron news, 26(2):11–14, 2015.
  • Tremsin et al. (2011) AS Tremsin, JB McPhate, W Kockelmann, JV Vallerga, OHW Siegmund, and WB Feller. High resolution Bragg edge transmission spectroscopy at pulsed neutron sources: Proof of principle experiments with a neutron counting MCP detector. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 633:S235–S238, 2011.
  • Vogel (2000) Sven Vogel. A Rietveld-Approach for the Analysis of Neutron Time-Of-Flight Transmission Data. PhD thesis, Christian-Albrechts-Universität, 2000.
  • Wensrich et al. (2016a) Christopher M Wensrich, Johannes Hendriks, and Michael H Meylan. Bragg edge neutron transmission strain tomography in granular systems. Strain, 52(1):80–87, 2016a.
  • Wensrich et al. (2016b) CM Wensrich, JN Hendriks, A Gregg, MH Meylan, V Luzin, and AS Tremsin. Bragg-edge neutron transmission strain tomography for in situ loadings. Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 383:52–58, 2016b.