A learning method for vector field approximation by neural networks

Kuroe, Yasuaki; Mitsui, Masao; Kawakami, Hajimu; Mori, Takahide

doi:10.1109/ijcnn.1998.687220

Cited by 5 publications

(3 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To solve the learning problem as mentioned above, we require calculating 'rotA A' in the equation (3), ' e c ' and ' e c 2 ' in the equation (6), and the partial derivative of J with respect to the synaptic weight wij of the network for steepest descent method, wJ / ww ij . An efficient algorithm for computing those partial derivatives can be derived by using 'Adjoint Neural Network' (Kuroe et al, 1998). The air purifier has an outlet (620 mm wide, 150 mm high) at the upper part and an inlet (510 mm wide, 300 mm high) at the lower part with a HEPA (High Efficiency Particulate Air) filter.…”

Section: Neural-net Based Modelingmentioning

confidence: 99%

Neural-net based modeling of velocity and concentration fields

Kimura

Yoke

Kaga

et al. 2009

J Vis

Self Cite

View full text Add to dashboard Cite

This paper proposes a novel algorithm using an artificial neural network for modeling simultaneously both a 3-D flow velocity vector and a concentration field. The neural network is trained so that four outputted values of the network, three components of a 3-D velocity vector and a concentration of substances such as air pollutants or bacilli, agree with measured ones and additionally the continuity and diffusion equations are satisfied in the flow field. An approximate model for the velocity and concentration field can be constructed in the neural network from sparsely measured data. When any 3-D position, (x, y, z), is inputted to the neural network model, it outputs a 3-D velocity vector and a concentration at the position. The entire 3-D velocity vector and concentration field, therefore, can be easily estimated using the model. To validate the algorithm, the smoke concentration distribution estimated from a very limited set of measured data is compared with the measured one in which most of the data is unused for the modeling. Even from sparsely measured velocity vectors and smoke concentrations, the novel algorithm gives the entire concentration distribution whose flow characteristics are almost similar to the experimental result.

show abstract

Section: Neural-net Based Modelingmentioning

confidence: 99%

Neural-net based modeling of velocity and concentration fields

Kimura

Yoke

Kaga

et al. 2009

J Vis

Self Cite

View full text Add to dashboard Cite

show abstract

“…An efficient algorithm for computing those derivatives can be derived by using the adjoint neural network (Kuroe et al, 1998). …”

Section: Construction Ofneural Networkmentioning

confidence: 99%

Gradient-based PIV using neural networks

Kimura

Susaki

Kiyohara

et al. 2002

J Vis

Self Cite

View full text Add to dashboard Cite

This paper proposes a new gradient-based PIV using an artificial neural network for acquiring the characteristics of a two-dimensional flow field. The neural network can effectively realize an accurate approximation of the vector field by introducing some knowledge on the characteristic property. The neural network is trained by using spatial and temporal image gradients so that the basic equation of the gradient-based method is satisfied. Since the neural network itself learns the stream function, the continuity equation of flow is consequently satisfied in the measured velocity vector field. The new gradient-basedPIV can be applied to even partly lacking visualized images.

show abstract

“…This method has been extended to preserve the isotropic turbulence property and isotropic homogeneous turbulence property (Zhong et al, 1993). Kuroe et al (1998) proposed a method for approximating the vector field using neural networks.…”

Section: Previous Workmentioning

confidence: 99%

Visualization and approximation of post processed computational fluid dynamics data in a virtual environment

Shahnawaz¹

View full text Add to dashboard Cite

Computational fluid dynamics (CFD) data is obtained as a result of numerical analysis performed on a flow model. In order for this data to be accurately analyzed, it is necessary to be able to visualize it in 3-D space. While visualizing this data, engineers sometimes find it necessary to visualize changes in the characteristics of the flow with respect to changes in the value of certain input parameters. However, generating CFD data sets for several values of input parameters is a tedious procedure, and cannot be done interactively. This thesis proposes a method where approximations between data sets are used instead of resolving the entire CFD model so that interaction with the data can be performed interactively. There are three parts to this thesis. The first part describes the methodology and features of visualizing the data in a virtual environment through the interactive creation and manipulation of flow entities such as streamlines, cutting• planes, and isosurfaces. The second part of the thesis analyzes two methods of approximation on large data sets. The third part, describes the actual interactive implementation of the approximation methods in a full scale application. The methods used here are the Virtual knot technique of B-spline curve fitting and basic linear interpolation. The input parameters, that are being varied, are the geometrical attributes of the data set. The application was developed using OpenGL and the Visualization Toolkit (VTK) libraries for the rendering and visualization calculations. In addition, the C2 interface libraries were used to create the Virtual Reality environment.

show abstract

A learning method for vector field approximation by neural networks

Cited by 5 publications

References 3 publications

Neural-net based modeling of velocity and concentration fields

Neural-net based modeling of velocity and concentration fields

Gradient-based PIV using neural networks

Visualization and approximation of post processed computational fluid dynamics data in a virtual environment

Contact Info

Product

Resources

About