A universal visualization method for promoting system representation and performance prediction of underwater gliders based on deep learning
Abstract
Artificial intelligence is revolutionizing the process of product design to make it smarter and more efficient, which provides a promising technique for constant development and optimization of underwater gliders (UGs). In this case, this paper proposes a universal visualization method for system representation and performance prediction of UGs based on deep learning, which can achieve intelligent and rapid performance evaluation by learning from serialized design schemes. Dataset is built via co-simulation and image representation based on mass properties and layout forms. The architecture of the performance prediction model is optimized and its prediction proficiency is improved through dataset statistical analysis and hyperparameter analysis. The performance and interpretability of the trained model are evaluated by comparison with other methods and feature map visualization. Finally, the effectiveness and applicability of the proposed method are evaluated with a practical engineering prototype, and the results show that it can realize performance prediction with a remarkable speed while maintaining considerable efficiency. Compared with the traditional parameter-based prediction methods, the proposed method innovatively involves visualized design schemes of multiple models and multiple series, which can be applied to diversified products, broadening the application range of deep learning in intelligent design of underwater equipment.