Machine Learning Modeling of Thermal System Components For BEVs

Abstract

Thermal-management components such as valves, fans and pumps are essential for maintaining the safety, efficiency, and reliability of Battery Electric Vehicles (BEVs). Their operation varies across a wide range of operating conditions, while progressive degradation can reduce thermal performance and increase the risk of unexpected maintenance. This thesis investigates a machine-learning-based framework for modelling the behaviour of selected thermal-system components and estimating their Remaining Useful Life (RUL) using condition-monitoring data. The methodology includes data analysis, generation and preparation of representative operating data, extraction of degradation-relevant features, construction of component health indicators, and RUL estimation. The developed framework demonstrates how data-driven models can support component health assessment when real run-to-failure data are limited. The work provides a basis for future validation using measured degradation data and for the development of condition-based maintenance strategies for BEV thermal systems.