No abstract

This paper presents the results of statistical analyses on the difference between measured and calculated fatigue rates for an oceangoing container vessel. Data from a hull monitoring system installed onboard this ship provides time-series of observed fatigue rates in actual operation, sea-and weather conditions. In addition, a hydrodynamic model has been used to calculate fatigue rates for selected locations in the mid-ship section due to vertical bending moment only, corresponding to experienced weather conditions over the same time period. The measured fatigue rates are then compared to the calculated fatigue rates and statistical regression models are established to explain the differences as a function of selected explanatory variables. This can then be used to correct for biases in the numerical models. Overall, the results indicate that the models perform reasonably well and are able to describe much of the variation in the difference. However, the models are found to not generalize very well and it may be challenging to find models for a whole fleet of ships.

Combined hydrodynamic and structural models are used to simulate structural responses on ship hulls in a seaway for design and risk assessment purposes. From a safety and inspection perspective, there is demand for continuously monitoring the ship hull conditions to estimate the structural utilization in real time. However, setting up the computer models and running the analysis are time consuming and costly, preventing such models from being used operationally. We developed a statistical model that approximates the wave bending moment output from the computer model, by linking the wave bending moment to ship design parameters and environmental information. This statistical model is computationally cheap and much faster to run than a hydrodynamic model, and may thus act as a virtual indicator sensor for structural condition monitoring. The approximated wave bending moment can also be used to compute fatigue damage for a ship as an indicator for crack risk. Although somewhat sensitive to the training dataset, a validation study reveals that the statistical model performs decently well. For fatigue rates, relative errors range from1.4% to 60% for out-of-sample results with weighted least squares, which is deemed acceptable for an indicator model used for screening a fleet of vessels.