Choose timezone
Your profile timezone:
In a single crystal diffraction experiment at BioMAX a small crystal sample must be carefully positioned within the photon beam. The problem can be handled manually or by beam scanning a large spatial area. The latter can imply a radiation damage to the sample. LTH students will present a machine learning-based method which does not rely on either brute force or manual supervision where to aim the beam. The students collected a dataset of visible camera images of crystal samples fixed in the instrument goniometer and labelled them using X-ray diffraction at BioMAX. Then they trained neural networks on that dataset. Two different approaches were used. In the first one a deep neural network was trained to classify images in respect to a crystal being present in them or not. With the second approach hidden correlations between visible camera pictures of crystal samples and diffraction quality criteria were searched by the machine learning regression method. The results proves that deep neural networks have good capability to learn where the crystal is. Further research might improve on this and guess on a more specific point on the crystal sample to give optimal results in the diffraction experiment.
Balasubramanian Thiagarajan et al.
R&D organisers on behalf of Zdnek Matej
