04 | Motion sickness among offshore wind technicians on smaller working vessels

Ramic E, Lutzen M | Institute of Mechanical and Electrical Engineering, University of Southern Denmark, Denmark

Background. As offshore wind farms expand in both number and scale,¹ so too does the need for skilled technicians and smaller working vessels to conduct maintenance operations.² Given the hazardous working conditions, maintaining the well-being and focus of technicians, especially upon arrival at the wind farm, is crucial. Motion Sickness Incidence (MSI) is the most used metric for quantifying motion sickness,³ but its focus on vertical motion and emesis as the primary symptom limits its applicability in this context. Smaller vessels experience more complex motion, and technicians may suffer from symptoms that impact performance without necessarily vomiting. This project aims to develop a machine learning tool that predicts the occurrence and severity of motion sickness symptoms based on sea and operational conditions.

Methods. Due to the interdisciplinary nature of the topic, multiple data sources are used. Motion data is collected from a subject vessel equipped with an onboard accelerometers measuring motion in several degrees of freedom. Sea state data is obtained through online forecast models. An onboard display allows technicians to self-report motion sickness symptoms and corresponding intensities. The data is used in a chained prediction system: first, vessel motion is estimated based on weather and operational inputs. Second, symptom severity is predicted from the motion data. To address uncertainties, Bayesian methods are integrated into the machine learning models for more informed predictions.

Results. Preliminary symptom data show low occurrences of nausea and emesis, but a notable presence of fatigue, emphasizing the importance of considering a broader range of symptoms. There are practical limitations to the collected data that may influence the modeling. Such a limitation could be that the subject vessel primarily operates in mild conditions, which could contribute to the generally healthy registrations. This dominance of healthy data affects model performance, a limitation to address in future work. Nonetheless, preliminary versions of the model have demonstrated that predicting symptom intensities based on vessel motion is feasible.