| dc.description.abstract | The human fatigue is a complex and has no universal standard to quantify or to record it. Several studies have attempted to measure with the use of machine learning techniques like Bayes Network or Deep Neural Networks. These algorithms assert the fatigue by mapping a subjective scale, i.e. the Karolinska Sleepiness Scale, with objective data like the EOG, ECG, etc. The maritime industries have also researched the understanding and quantification of tiredness, though its focus has been more on the risk assessment problem. A few researchers have dedicated themselves to uncover the causes of fatigue in the offshore operations, however, to prevent it effectively a reliable method to measure sleepiness is necessary. This work has attempted to design a general framework where fatigue could be measured and recorded, allowing operators in the maritime industries to be monitored over time in a reliable and high accurate way. Using methods from data fusion theory and machine deep learning techniques two data fusion architectures were proposed and studied: A centralized and a distributed architecture. Though the centralized was able to achieve a high precision in the prediction of the fatigue scale KSS(over 98%), the structure that was redeemed better suited for the offshore operation was the decision fusion. The decision fusion had achieved a precision of 96.08% while losing less than 10% of accuracy if certain sensors were missing. These properties are ideal for offshore operations where the driving task of each operator may change over time, allowing the decision of which sensor to use be done by the crew and management. Besides the environment of the operation is hostile and, therefore, damage to the equipment is inevitable so the architecture should be resilient to it. | nb_NO |
