Yuki SAKAMOTO

At the European Spallation Source (ESS) ERIC, the liquid hydrogen moderator development is undertaken for its predominantly high parahydrogen fraction, which helps to attain a higher neutron intensity at a very high brightness. The Cryogenic Moderator System (CMS) is equipped with a catalyst to convert hydrogen from the ortho to the parastate to keep desirably high parahydrogen fractions of more than 99.5% in the cold moderators, which is required to deliver high brightness cold neutron beams to the neutron instruments. An in-situ measurement system for the ortho and para fractions of liquid hydrogen (OPMS) has been developed using a Raman spectroscopy to detect any undesirable shift towards a high orthohydrogen fraction caused by neutron scattering driven para-to-ortho back conversion. A Raman optics system was installed into a mock-up OPMS vacuum chamber and its performance evaluation tests have been conducted by flowing liquid hydrogen. It was verified that the developed Raman optics system succeeded in measuring the parahydrogen fraction with an accuracy of 0.1%, which met the requirement.

Cryogenic fluids such as liquid hydrogen and liquid oxygen used as the rocket propellants easily evaporate and form the gas-liquid two-phase flow. The control of the two-phase flow is difficult due to the large fluctuation of the density. For high-precision control, it is necessary simultaneously to understand flow regimes which represent a gas-liquid distribution pattern. Therefore, in this paper, we developed a classifier that uses Bidirectional LSTM networks, which is part of a family of deep learning methods, with a measured value of a void fraction meter as input and gas-liquid flow rate conditions as output, in order to realize a flow regime classifier in the future. The classifier succeeded in classifying with more than 80% accuracy. In addition, in order to verify what features of the input data the classifier captures, a test data-set of which frequency was artificially changed was classified. As a result, it was confirmed that the classifier would use the frequency component of the input data as one of the basis for classification.