Data-driven condition monitoring of mooring systems: A demonstration with the M4 multi-float wave energy converter

• Data-driven monitoring of mooring systems using time-delay neural networks. • Captures highly nonlinear mooring dynamics under varying sea conditions. • Detects subtle and sudden faults across different mooring configurations. • Robust fault detection under unseen and complex sea state conditions. Monitoring the condition of mooring systems is essential for the safe operation and timely maintenance of floating platforms including wave energy converters (WECs). However, mooring dynamics are highly nonlinear due to nonlinearities in wave forcing, marine growth effects on floater and mooring cables or chains, WEC dynamics, synthetic cable or chain nonlinearity, and floater-mooring coupling. These effects reduce the effectiveness of linear system identification methods and make accurate condition assessment challenging. In this study, we introduce a data-driven approach for condition monitoring that uses dynamic modelling to address these challenges. The method enables continuous monitoring, allowing gradual or sudden changes to be detected and faults addressed promptly. A time-delay neural network (TDNN) with Bayesian regularization backpropagation is employed to model the system’s behaviour. Its performance is comprehensively compared with four popular models. The approach is validated using two surrogate fault scenarios: (i) two different mooring configurations tested in the same basin, and (ii) nominally identical configurations tested in two different basins. Using six-degree-of-freedom (6-DOF) motion data, the approach is demonstrated to identify the fault condition for all sea states under both scenarios, demonstrating the robustness and sensitivity for subtle changes, for unseen and complex conditions.