Slot-die coating is a precision-controlled film deposition technique widely used in lithium ion battery manufacturing and flexible electronics. Despite its advantages in uniformity and material efficiency, the process is highly sensitive to variationsin flow rate, gap height, viscosity, and tension, which can lead to defects such as edge waves, ribbing, and voids.

This thesis presents a hybrid modeling framework that combines analytical modeling, synthetic data generation, machine learning, and CFD simulations to predict and optimize coating uniformity in real-time.

A mathematical model based on lubrication theory was used to generate a synthetic dataset that captures coating thickness and uniformity across a range of input parameters.

This dataset was used to train and evaluate machine learning models capable of predicting film behavior and classifying defects. Among several algorithms, multilayer perceptrons and random forests showed strong predictive performance.Selected configurations were validated using CFD simulations performed in SimScale (OpenFOAM), revealing strong agreement with model predictions in terms of flow structure and interface stability.

The main conclusion of this work is that data-driven modeling, when grounded in physical principles, can effectively predict and prevent coating defects, even in the absence of real-time sensor data. Furthermore, the framework lays the foundation for a digital twin prototype that could support adaptive control in automated liquid manufacturing environments.

These results demonstrate that a simulation-informed, machine learning-guided strategy can improve process reliability, minimize material waste, and enhance product quality in thin-film production systems.