Structural damage detection in steel beams is a key challenge in Structural Health Monitoring (SHM), aiming to improve structural safety, reliability, and maintenance planning. This study presents a vibration-based damage assessment framework that integrates finite element (FE) modeling, feature extraction, machine learning, and transfer learning for damage detection, localization, and severity estimation in steel beams. A high-fidelity three-dimensional finite element model of an IPE160 steel beam was developed in ANSYS and validated against experimental measurements through comparison of natural frequencies and mode shapes. The validated model was subsequently used to generate a dataset comprising 259 damage scenarios with varying crack locations and severities.