Department of Computer Science: MSc Thesis Presentations

Predicting the Concentration Values of Individual Chemical Components in Ternary Chemical Mixtures

Author: Ngoc Nguyen
Supervisor: Juho Rousu
Site of research: Vaisala Oyj
Time: Tuesday 21 May at 14:00-14:30
Place: online (zoom) https://aalto.zoom.us/j/69779858738

Abstract: In the realm of optical sensors for predicting the concentration of ternary solutions, such as a mixture of water, sugar and salt, challenges arise due to the inability to provide specific measurements for each individual component. This thesis proposes a novel approach to address these challenges by employing a machine learning technique. This technique integrates refractive index measurements with other parameters such as conductivity or density, aiming to accurately predict a variety of different ternary chemical compounds.

The objectives encompass the identification of a suitable machine learning model, its training and testing on various chemical compounds, and the validation of its performance on Linux Embedded devices with limited computing power. The research methodology includes stages of data collection, model selection, black box engineering, data splitting, model training and evaluation, and performance testing on Raspberry Pi. The anticipated outcome of this research is to enhance the precision of measurements in ternary solutions, thereby contributing to advancements in chemical analysis.

The machine learning models introduced in this thesis, employing support vector regression and neural network models, have shown promising initial results. They exhibit an improvement in prediction accuracy when compared to the pre-existing 3𝑟𝑑 degree polynomial models. Furthermore, these developed models are capable of operating on embedded systems with restricted computational resources.