Wyniki wyszukiwania - BazTech

1

Performance of artificial neural networks in an inverse problem of laser beam diagnostics

Pietrak Karol, Muszyński Radosław, Marek Adam, Łapka Piotr

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2022

|

Vol. 70, nr 1

art. no. e140100

EN

The presented results are for the numerical verification of a method devised to identify an unknown spatio-temporal distribution of heat flux that occurs at the surface of a thin aluminum plate, as a result of pulsed laser beam excitation. The presented identification of boundary heat flux function is a part of the newly proposed laser beam profiling method and utilizes artificial neural networks trained on temperature distributions generated with the ANSYS Fluent solver. The paper focuses on the selection of the most effective neural network hyperparameters and compares the results of neural network identification with the Levenberg–Marquardt method used earlier and discussed in previous articles. For the levels of noise measured in physical experiments (0.25–0.5 K), the accuracy of the current parameter estimation method is between 5 and 10%. Design changes that may increase its accuracy are thoroughly discussed.

2

Machine learning based prediction of regularization parameters for seismic inverse problems

Liu Shihuan, Zhang Jiashu

Acta Geophysica

|

2021

|

Vol. 69, no. 3

809--820

EN

Regularization parameter selection (RPS) is one of the most important tasks in solving inverse problems. The most common approaches seek the optimal regularization parameter (ORP) from a sequence of candidate values. However, these methods are often time-consuming because they need to conduct the estimation process on all candidate values, and they are always restricted to solve certain problem types. In this paper, we propose a novel machine learning-based prediction framework (MLBP) for the RPS problem. The MLBP frst generates a large number of synthetic data by varying the inputs with diferent noise conditions. Then, MLBP extracts some pre-defned features to represent the input data and computes the ORP of each synthetic example by using true models. The pairs of ORP and extracted features construct a training set, which is used to train a regression model to describe the relationship between the ORP and input data. Therefore, for newly practical inverse problems, MLBP can predict their ORPs directly with the pre-trained regression model, avoiding wasting computational resources on improper regularization parameters. The numerical results also show that MLBP requires signifcantly less computing time and provides more accurate solutions for diferent tasks than traditional methods. Especially, even though the MLBP trains the regression model on synthetic data, it can also achieve satisfying performance when directly applied to feld data.

3

Chlorophyll profile estimation in ocean waters by a set of artificial neural networks

Cortivo F. D., Chalhoub E. S., Campos Velho H. F., Kampel M.

Computer Assisted Methods in Engineering and Science

|

2015

|

Vol. 22, no. 1

63--88

EN

In this work, we propose a methodology to estimate the profile of chlorophyll concentration from the upwelling radiation at the ocean surface, using a system of artificial neural networks (ANNs). The input patterns to train the networks are obtained from the resolution of the radiative transfer equation, where the absorption and scattering coefficients are represented by bio-optical models, with the profile of chlorophyll concentrations based on a shifted-Gaussian model. In the performed analysis, we used 14 720 profiles of chlorophyll that were generated by attributing two values to the biomass quantity, and by considering two sets of wavelengths and three sets containing the directions in which the radiation emitted at the surface is measured. To be able to recover the chlorophyll profile, we need to use a system of networks that works in a “cascade mode”. The first one performs an analysis on the features of the chlorophyll profile from the upwelling radiation and determines which profiles can be recovered. The second and third ANNs act only on those profiles that can be recovered. The second ANN performs estimation of the standard deviation from the upwelling radiation and the chlorophyll concentration at the surface. Finally, the third ANN performs an estimation of the peak depth from the upwelling radiation, the chlorophyll concentration at the surface and the standard deviation estimated by second network. The stopping criteria we adopted was the cross-validation process. The obtained results show that the proposed methodology is quite promising.