Against the backdrop of widespread application of large-scale equipment, online monitoring of equipment structural health status has become of paramount importance. Structural health monitoring (SHM) methods based on active guided waves have found applications in the field of damage diagnosis due to their characteristics and advantages such as high sensitivity to damage and the ability to propagate over long distances. However, the random and irregular vibrations generated by large-scale equipment during operation can affect the propagation characteristics of guided wave signals. Severe vibrations may even obscure the guided wave signals in the structure, hinder the extraction of these signals, and reduce the accuracy of SHM. To address this issue, this study proposes a guided wave-Gaussian process (GW-GP) damage prediction model. The model integrates active guided wave-based SHM technology with Gaussian process machine learning algorithms. It constructs a nonlinear mapping relationship between damage indices and crack length using damage indices such as root mean square deviation and normalized cross-correlation moment, and optimizes hyperparameters via the conjugate gradient method. Results from aluminum plate crack propagation experiments show that the maximum absolute error between the model-predicted crack length and the true value is 1.52 mm, and the root mean square error is 0.72 mm. This effectively enables quantitative diagnosis and prediction of structural damage under vibration conditions, providing a new technical pathway for structural health monitoring of large-scale equipment.