Search

We propose a method to construct a reduced order model with machine learning for unsteady flows. The present machine-learned reduced order model (ML-ROM) is constructed by combining a convolutional neural network autoencoder (CNN-AE) and a long short-term memory (LSTM), which are trained in a sequential manner. First, the CNN-AE is trained using direct numerical simulation (DNS) data so as to map the high-dimensional flow data into low-dimensional latent space. Then, the LSTM is utilized to establish a temporal prediction system for the low-dimensionalized vectors obtained by CNN-AE. As a test case, we consider flows around a bluff body whose shape is defined using a combination of trigonometric functions with random amplitudes. The present ML-ROMs are trained on a set of 80 bluff body shapes and tested on a different set of 20 bluff body shapes not used for training, with both training and test shapes chosen from the same random distribution. The flow fields are confirmed to be well reproduced by the present ML-ROM in terms of various statistics. We also focus on the influence of two main parameters: (1) the latent vector size in the CNN-AE, and (2) the time step size between the mapped vectors used for the LSTM. The present results show that the ML-ROM works well even for unseen shapes of bluff bodies when these parameters are properly chosen, which implies great potential for the present type of ML-ROM to be applied to more complex flows. [ABSTRACT FROM AUTHOR]

Abstract In order to understand the electrostatic charging state of atmospheric hundreds nanometer particles that has particular impact on health, we have observed the atmospheric particles charge distributions intermittently for one year. To easily estimate particle charge distribution, first of all, we have developed a simple instrument and calculation method. We confirmed by comparison with stationary charge distributions that the distributions obtained by our method can be used to easily determine the relative variation of particle charging states, although the proportion of particles with a small charge number was quantitatively underestimated. The charge distributions of the atmospheric particles observed from April 2017 to February 2018 were different from the generally accepted stationary charge distribution. Moreover, the observed distributions changed significantly throughout the year. This variation was found to correlate with seasonal variations in temperature, relative humidity, volumetric humidity, and air mass trajectories. Present result suggests that the seasonal variation of the charging state is possibly due to the seasonal variation of the atmospheric ion number concentrations. Thus, it is important to consider the variation in the charging state of atmospheric particles due to dynamic changes of atmospheric conditions, as the charging state can dramatically influence the deposition of particles in the lungs. Highlights • The charging state of particles is important for the adverse effect on human health. • We observed the charging distributions of the atmospheric particles for one year. • The observed charging state were different from the stationary charging state. • The particles charging states at atmosphere has a seasonal variation. • These variations were correlated with variations in the atmospheric conditions. [ABSTRACT FROM AUTHOR]

For adverse-pressure-gradient turbulent boundary layers, the study of integral skin-friction contributions still poses significant challenges. Beyond questions related to the integration boundaries and the derivation procedure, which have been thoroughly investigated in the literature, an important issue is how different terms should be aggregated. The nature of these flows, which exhibit significant in-homogeneity in the streamwise direction, usually results in cancellation between several contributions with high absolute values. We propose a formulation of the identity derived by Fukagata et al. (2002), which we obtained from the convective form of the governing equations. A new skin-friction contribution is defined, considering wall-tangential convection and pressure gradient together. This contribution is related to the evolution of the dynamic pressure in the mean flow. The results of the decomposition are examined for a broad range of pressure-gradient conditions and different flow-control strategies. We found that the new formulation of the identity allows to readily identify the different regimes of near-equilibrium conditions and approaching separation. It also provides a more effective description of control effects. A similar aggregation between convection and pressure-gradient terms is also possible for any other decomposition where in-homogeneity contributions are considered explicitly. • Pressure gradient and convection shall be considered together in skin-friction contributions. • Near-equilibrium and near-separation boundary layers differ in the balance of contributions. • A dynamic-pressure contribution assists to describe pressure-gradient and control effects. [ABSTRACT FROM AUTHOR]

We use Gaussian stochastic weight averaging (SWAG) to assess the epistemic uncertainty associated with neural-network-based function approximation relevant to fluid flows. SWAG approximates a posterior Gaussian distribution of each weight, given training data, and a constant learning rate. Having access to this distribution, it is able to create multiple models with various combinations of sampled weights, which can be used to obtain ensemble predictions. The average of such an ensemble can be regarded as the 'mean estimation', whereas its standard deviation can be used to construct 'confidence intervals', which enable us to perform uncertainty quantification (UQ) with regard to the training process of neural networks. We utilize representative neural-network-based function approximation tasks for the following cases: (i) a two-dimensional circular-cylinder wake; (ii) the DayMET dataset (maximum daily temperature in North America); (iii) a three-dimensional square-cylinder wake; and (iv) urban flow, to assess the generalizability of the present idea for a wide range of complex datasets. SWAG-based UQ can be applied regardless of the network architecture, and therefore, we demonstrate the applicability of the method for two types of neural networks: (i) global field reconstruction from sparse sensors by combining convolutional neural network (CNN) and multi-layer perceptron (MLP); and (ii) far-field state estimation from sectional data with two-dimensional CNN. We find that SWAG can obtain physically-interpretable confidence-interval estimates from the perspective of epistemic uncertainty. This capability supports its use for a wide range of problems in science and engineering. • Weight sampling during convergence quantifies epistemic uncertainty of deep learning. • We fit a multivariate Gaussian distribution of sampled weights for resampling. • Samples from Gaussian are propagated for ensemble predictions. • Ensemble predictions provide confidence intervals of epistemic uncertainty. [ABSTRACT FROM AUTHOR]

We fabricate multimode polymer optical waveguides with circular graded-index (GI) cores which are aligned in parallel at desired positions using the Mosquito method. In the Mosquito method, three-dimensional wiring patterns can be formed with a simple process. However, the core position is likely to deviate from the designed position because of multiple fabrication factors. Hence, in this paper, the dominant parameters to influence on the core height in the cladding are investigated both theoretically and experimentally. In particular, a linear relationship between the core height and the needle-tip height is confirmed with theoretical fluid analysis. Using this relationship, we succeeded in fabricating a waveguide in which the maximum variation of the core height from the designed value is controlled to be less than 10 µm.