Topographical Reconstruction of Proton Radiographs

Lead Research Organisation: University of Oxford
Department Name: Oxford Physics


Proton radiography is a technique extensively used to resolve magnetic field structures in high energy density plasmas, revealing a whole variety of interesting phenomena such as magnetic reconnection, dynamo and collisionless shocks found in laboratory analogues to astrophysical systems. The process works by energetic protons, generated by petawatt-class laser interactions with thin foil targets, traversing the main plasma and being deflected by self-generated electric and magnetic fields therein. Post-processing of an image formed on a screen placed some distance away in principle allows the fields to be reconstructed but, in practice, this is extremely challenging.

Over the past two years, my team has developed novel 2-D and 3-D reconstruction methods using computational geometries1 and artificial neural networks2-4. The latter are based on the feed-forward and convolved artificial neural network methods that have the potential to provide quantitative results for the first time. They expand an arbitrary magnetic field into a linear combination of basis fields, which are deliberately chosen to minimise the data needed to train the neural network. We have recently demonstrated that this method is successful; has mean reconstruction errors of less than 5 percent, even after introducing noise; and that, over the long term, is more computationally efficient than other techniques. Its advantages are that it can overcome existing limitations that occur in the non-linear regime. It has applications not only to proton radiography, but also to quantitative optical shadowgraphy and, with that, a multitude of applications across high energy density plasma physics.

We highlight the need for a tomographic approach that the student, Benjamin Spiers, will investigate as the core element of his PhD because (i) certain field structures cannot be reconstructed from a single radiograph and (ii) the errors can be further reduced when reconstruction is performed on radiographs generated by proton beams fired in different directions. Our preliminary investigations are very promising. One generates batches of protons and propagates each batch onto a screen using a standard algorithm; the magnetic field values used are treated as weights of a neural network. By constructing a network that carries out this standard algorithm as a differentiable process, any discrepancies between the generated image and the desired radiograph image can then be minimised by a gradient descent variant - the field is iteratively updated until the difference between the generated and target images is negligible. This approach to tomographic reconstruction will provide an exquisite new tool for dense plasma physics.

Indeed, a number of experiments have been fielded on ORION since its commissioning that would benefit from this proton radiography toolkit. At first, the student will help complete the data analysis of the SCAP experiment for channelling, the modelling for which has yielded exquisite simulated proton radiographs. They formed central elements of his and Mr Angus Brayne's MPhys project reports. Mr Spiers will continue to work on that data, and will prepare the publications for that experiment, as well as assist with the interpretation of face-on proton radiographs using the Kasim approach1.

1. M. Kasim et al., Phys Rev E 95, 023306 (2017)
2. N. Chen et al., Phys. Rev. E 95, 043305 (2017)
3. A. Brayne, MPhys Project Report (2017)
4. B. Spiers, MPhys Project Report (2017)


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/R512333/1 01/10/2017 30/09/2021
1950656 Studentship EP/R512333/1 01/10/2017 31/03/2021 Benjamin Spiers