Time series classification models for anomalous transport phenomena

This thesis investigates time series classification through the application of various algorithms on two distinct datasets, developed in collaboration with Eltek S.p.A. The research aims to assess the performance of these algorithms on both artificially generated and experimentally obtained data. The first case study involves an artificially generated dataset created via simulations to analyze the heat transport of a sensor immersed in a fluid under an applied voltage. The simulations were divided into four groups: one with normal Fourier heat transport and three with different models of anomalous transport, the primary objective was to recognize the four heat transport models using temperature time series. The second case study utilized experimentally obtained data from capacitor discharge in a fluid, generating time series of the capacitor’s voltage. Each experiment was classified by assigning classes from 0 to 15, the objective was to evaluate the performance of various time series classification models on both real and artificially generated datasets. The study's results highlighted the strengths and limitations of different algorithms in controlled and real-world settings, providing insights into their generalizability and practical industrial applications. This research demonstrates the feasibility of using time series classification algorithms to differentiate between heat transport models and classify experimental data based on transport parameters. The collaboration with Eltek S.p.A underscores the industrial relevance of these techniques, suggesting future exploration of more sophisticated models and experimental conditions to enhance accuracy and applicability in various domains.