In structural engineering, offshore structures are undoubtedly among the most exposed to the effects of harsh environmental conditions. The external conditions of these semi-immersed systems involve complex combinations of wave and wind loads. The operating conditions are also unique because oil production platforms are subjected to repeated loading and unloading cycles of the extracted material, which continuously alter their mass. These characteristics make the definition of a structural health monitoring (SHM) protocol highly challenging but necessary to avoid environmental disasters. In this regard, this study discusses an SHM method that can be applied to offshore structures under realistic wave and wind loads. This approach combines anomaly detection, frequency domain decomposition, and a cointegration strategy. Two machine learning regression algorithms were tested to define a cointegration relationship: the support vector machine and the relevance vector machine. The effectiveness of the overall method was evaluated on time-domain signals generated from a finite-element model of a fixed steel platform, on which the Davenport and JONSWAP spectra were used to simulate wind and wave forces. The results show that this damage detection strategy is effective in supervising the health conditions in the analyzed scenario.