The objective of this study was to automate the phase analysis-unwrapping procedure using machine learning. We obtained phase changes Δϕexp using the existing rule-based procedure from the time series of intensity of the interference fringe I(t)exp derived from the temperature change in a homogenized solution with a temperature difference at both ends, which was observed using an interferometer onboard the International Space Station. Machine learning models were then trained using the least-square (LS) boosting method with a part of I(t)exp and the corresponding Δϕexp. As a result, the phase changes Δϕpred(exp) for untrained I(t)exp in the same fringes could be predicted with high accuracy by the trained models. In addition, training the spatial distributions of the fringe intensities depending on the optical set-up was found to be critical. The models were then trained with artificially generated time series of intensity I(t)art processed through the arbitrary spatial-temporal shifting, height scaling and noising to simulate I(t)exp. As a result, it was revealed that the machine learning models, whose pre-training time is about 20000 s at most in this study, can predict the phase changes Δϕpred(art) with high accuracy and quite high speed (about 1.8 s) without the existing rule-based methods (about 900 s).

抄録全体を表示

The crystal structure of DyMnO3 (Ln: lanthanide) in the equilibrium state has been reported to be orthorhombic when La3+ to Dy3+, and hexagonal when Ho3+ to Lu3+, are used as Ln3+. However, Kumar et al. reported a two-phase structure of orthorhombic and hexagonal phases is formed in DyMnO3 when it was solidified from an undercooled melt under containerless conditions. To investigate the reason for the formation of this two-phase structure in DyMnO3, the nucleation temperature Tn and the post-recalescence temperature Tpr were measured for the Dy2O3-Mn2O3 system as a function of the substitution ratio for Mn and Dy, x = NMn/(NDy+NMn). As a result, the phase constitution was found to depend on Tn and the amount of undercooling ΔT(=Tpr - Tn) ; h-DyMnO3 is formed when ΔT is small, and o-DyMnO3 is formed when ΔT is large. Furthermore, when Tn is higher than 1530 K, which is the equilibrium temperature at oxygen partial pressure of atmospheric gas at 105 Pa for the chemical reaction of DyMnO3 + 1/3Mn3O4 + 1/3O2 = DyMn2O5, h-DyMnO3 is formed at x(0.5, while h-DyMnO3 is suppressed at x)0.5, irrespective of Tn.

抄録全体を表示

In dusty plasmas, dust particles are negatively charged and massive so as to be affected by gravity. In microgravity experiments with a cylindrical discharge, the dust particles move their equilibrium position with changing their arrangements from an isotropic closed-packed structure to anisotropic linear chains. The Coulomb potential dominant in the isotropic structure was evaluated based on mesurement of plasma parameters of ion density and electron temperature. It was found that the Coulomb coupling parameters of the ratio of the potential to thermal energy were large enough to solidify Coulomb crystals of the dust particles. Neverthless the Coulomb potential can not overcome the potential forming the linear chains which appares in a microgravity condition, so-called wake potential induced by ion stream along the axis of the cylindrical discharge.

抄録全体を表示