Concrete is a heterogeneous quasi-brittle material, whose failure behavior, when subjected to external loading, is characterized by diffuse microcracking, which eventually coalesce into localized cracking and ultimately results in structural failure. Early detection of damage is vital for reducing maintenance costs of concrete structures. Monitoring using diffuse ultrasonic waves (coda waves) have proven highly effective for this purpose. This study presents a virtual testing platform for early damage assessment in concrete structures using coda wave analysis. It integrates multiscale damage modeling, ultrasonic wave simulation, and machine learning-based damage classification. At the mortar scale, microcracking is captured through a hybrid modeling approach that combines continuum micromechanics with linear elastic fracture mechanics. This model is then incorporated into a reduced-order mesoscale model, enabling efficient simulations of concrete damage at multiple scales. Using this multiscale damage model, synthetic concrete specimens are generated to represent various states of degradation. Ultrasonic wave monitoring within these specimens is simulated using a rotated staggered-grid finite-difference method, offering a detailed analysis of wave propagation through damaged material. Finally, a transformer encoder processes the resulting coda wave data, providing an efficient classification of damage levels.