Refractive index determination of dynamic droplets in a flow by analyzing light scattering signals with a machine learning approach

This container provides the measurement data, Python resources, and trained model weights associated with the scientific work that will be presented in 2025 at the Turbulence, Heat and Mass Transfer 11 conference in Tokyo. The following data files are included: Dataset_40_4ch1234.rar(unpacked: Dataset_40_4ch1234.pth) M1_SegmentsTHR40.csv SegmentsTHR40.rar(unpacked: M1_SegmentsTHR40.csv … M55_SegmentsTHR40.csv) Model_weights_4ch1234.pth File descriptions Dataset_40_4ch1234.pthPyTorch file containing a ready-to-use dataset of four-channel light-scattering signals, preprocessed for direct use in the accompanying Python scripts. M1_SegmentsTHR40.csvExample file illustrating the format used to store and load light-scattering signals of individual droplets together with auxiliary information. The columns are: MID – measurement ID FID – frame ID SID – signal ID CID – channel ID NOP – number of parts PNM – part number TCH – trigger channel TLE – trigger level TID – trigger ID CON – label used for training SegmentsTHR40.rarArchive containing multiple .csv files (M1_SegmentsTHR40.csv … M55_SegmentsTHR40.csv) in the same format as described above. Together, these files form the full segmented signal dataset. Model_weights_4ch1234.pthPyTorch file containing the trained weights of the neural network model based on data from all four detector channels. External code repository The corresponding code repository for this dataset is hosted on Azure DevOps:https://dev.azure.com/ai-quanton/PBa202 This repository contains the Python script used to build, train, and evaluate a neural network that determines the refractive index of single droplets by analyzing light-scattering signals generated as the droplets pass through a Gaussian laser beam. The model is designed to predict refractive indices from time-resolved light-scattering data in dynamic spray environments with high accuracy.