Ship propulsion shafting assembly detection: a model-data-driven approach

The ship propulsion shafting assembly is one of the keys in building a ship power system, which directly affects the power performance and navigation safety. In this paper, a model-data-driven approach called semi-supervised domain-adversarial neural networks (SDANN) is proposed to solve the lack of actual training samples during ship propulsion shafting assembly. In terms of the model, a finite element model (FEM) for ship propulsion shafting considering misalignment is built in ANSYS. In terms of data, a ship propulsion shafting comprehensive test platform is set up in order to collect the actual data in the laboratory. Furthermore, a small amount of the unlabelled actual data is used to train semi-supervised networks, which can identify four different bearing elevations. The research results illustrate that the proposed method outperformed classical deep learning methods in the ship propulsion shafting assembly detection under three types of cases, with an average accuracy of 95.61% +/- 0.45%.