Stage 3 & 4 - Analytics and AI Digital Twins

Highlights

• The digital twin concept was introduced by Michael Grieves in 2003 at the University of Michigan, later adopted by NASA in 2012, with the human digital twin formally defined in 2021 and brought into the maritime context through the Mariner 4.0 framework in 2022
• A human digital twin in maritime is described as a real-time digital replica of a seafarer, updated continuously across physiological, behavioral, and environmental data, capable of simulation and prediction
• Three distinguishing capabilities are outlined: simulation of a specific body's response using finite element methods, personalization through individual baselines, and prediction that flags risk before symptoms emerge
• In classical machine learning, decision trees reach 82.6% on fatigue detection and naive Bayes 85.5% on stress, while deep learning and hybrid models perform higher, such as LSTM-based fatigue detection reporting 99.9%
• Caution is urged that very high accuracy figures may reflect overfitting on small datasets, and that accuracy does not equal generalization across populations, conditions, and real-world variation