A fast artificial neural network approach for dynamic light scattering time series processing

Abstract

This paper presents a simple alternative to dynamic light scattering (DLS) time series processing by using an artificial neural network. A simple experiment for recording a DLS time series is presented. The reference method for DLS time series processing consisted of fitting the analytical form of the Lorentzian line to the frequency spectrum of the recorded scattered light intensity. An artificial neural network with one hidden layer was designed and trained. The training data consisted of a big set of autocorrelations of simulated time series for monodispersed spherical particles with diameters in the range 10–1200 µm. The neural network output precision was tested both on simulated and on experimental time series recorded on fluids containing nanoparticles and microparticles. The errors of the artificial neural network output relative to the reference diameters were small enough and the data processing procedure was three orders of magnitude faster, proving that, in spite of the simplicity, the artificial neural networks approach can be a faster alternative for DLS time series processing.

A Study of Milk Particles Size Variation with pH Change using Dynamic Light Scattering

“…1 and consists of a laser source, which was a Laser diode, 15 mW, with the wavelength of 633 nm, working in continuous regime, a circular glass tube for the liquid sample, a detector, a preamplifier with a linear response in the audio frequency range 20 -20000 Hz, a PC for recording and later on processing the time series. The scattering angle was chosen to be 90 o , which is typical for DLS experiments [6], [7], [9], [11][12][13]. The data acquisition rate was 16 KHz, which was sufficient for the particles range covered in this work.…”

“…( ) ( ) ( ) ( ) (1) If the suspended particles are considered to be monodispersed, the autocorrelation of the time series (ACR hereafter) has a simplified form [6], [11][12][13], [14]:…”

“…Firstly, constant B is subtracted, which makes ACR to be zero-based. Secondly, an experimental adjustment is made in such a manner that the detector will cover a single speckle, as described in detail in [11][12][13] and [14], making the spatial coherence factor, included in A for this particular case, to be equal to 1. After normalization, the ACR has a simple form depending on a single parameter, which is D:…”

“…A Matlab code was written and used to record the DLS time series [11][12][13]. The time length of the time series and the delay between consecutive recordings are adjustable parameters.…”

Comparison of neural networks and regression time series in predicting export from Czech Republic into People´s Republic of China

“…Since the first computers were created, the best programmes have been striving for creating an algorithm that would be able to imitate the activity of human brain. The result of these efforts is later referred to as artificial intelligence, or artificial neural networks [1].…”