Abstract
The commercial viability of Wave Energy Converters (WECs) depends heavily on reducing operation and maintenance costs by accurately detecting mechanical degradation, such as Power Take-Off (PTO) failures. A major challenge in the current state of the art is distinguishing true mechanical faults from false alarms triggered by a typical metocean conditions. This paper presents a test site case study proposing a three-phase data-driven methodology for the performance analysis of a WEC fleet. Due to early-stage development sensitivity, the study utilizes a semi-empirical dataset where both nominal metocean conditions and WEC power generation are extracted directly from real offshore telemetry, while environmental penalties and mechanical faults are introduced computationally to evaluate algorithmic robustness. The methodology integrates two key performance analysis models: a supervised learning approach and a parametric model. First, an Extreme Gradient Boosting (XGBoost) algorithm acts as a regressor to establish an absolute power generation baseline based on wave inputs. Second, a Cobb-Douglas Stochastic Frontier Analysis (SFA) evaluates the relative technical efficiency of the assets. The SFA decomposes the composite residual, isolating one-sided mechanical inefficiencies from symmetric environmental noise, such as spectral spreading. Finally, a deterministic decision engine merges the absolute power deficits with the relative SFA scores. Results demonstrate that this architecture effectively mitigates environmental false positives and identifies critical mechanical faults, providing a promising framework for offshore maintenance dispatch. To promote reproducibility and open science within the marine energy sector, the complete codebase developed for this study is made publicly available.