Degradation in the mooring system of floating offshore wind turbines (FOWTs) leads to changes in the moored platform’s dynamic responses. A robust health monitoring system able to continuously monitor different mooring systems for FOWTs should rely on a fast model able to generate these dynamic responses under diverse health, operational, and metocean conditions. To this purpose, we propose a hierarchical variational autoencoder (HVAE) designed based on diffusion probabilistic architecture. This training occurs through a pretrain-finetune procedure aimed at learning the nonlinear relationship between healthy and minority-damaged responses belonging to a mooring system, considered as the source domain under different sea states. Following this, by utilising the healthy data from the target mooring system, the HVAE can effectively generate real-scale damaged responses of the system under various operational and environmental conditions. Owing to the lack of real measurements, the training and evaluation datasets are simulated using the open-source software OpenFast, with the OC4-DeepCWind semisubmersible platform serving as the basis for the use case. A thorough examination of the similarity between the simulated records from OpenFast and those generated by the HVAE architecture is conducted through visual, statistical, and behavioural analyses. To investigate the impact of random excitation, the testing data encompass various wave seed values. The generated records for unobserved sea states exhibit behaviours closely resembling real ones during downstream binary classification, underscoring the efficacy and adaptability of HVAE.