Knowledge reasoning and strategy optimization for ship operation and maintenance based on digital twin and improved KD tree algorithm

Objective  With the continuous development of industrial technology, the intelligence of modern ship processes has been continuously advancing. The propulsion system, auxiliary power system, etc. of ships have become increasingly intelligent, and ship maintenance work has become ever more complex. Different from land equipment, the environment in which ships are located is more severe. When a problem occurs, it will not only affect the stability of the ship during operation, but also bring huge safety hazards.

Method This paper focuses on a knowledge reasoning method for ship operation and maintenance based on digital twin technology. Based on the physical entity of the ship, the ship operation and maintenance process is analyzed, and a digital twin model for ship operation and maintenance is constructed from the multi-dimensions of "geometry-physics-behavior-rule". Aiming at the early warning information in the ship operation and maintenance knowledge model, by using previous ship operation and maintenance cases, a ship operation and maintenance case database containing ship dynamic monitoring data and maintenance methods is established. Based on the database, a method for ship operation and maintenance knowledge reasoning and strategy generation using an improved KD tree algorithm is proposed. Neighboring cases are weighted using Gaussian distance weighting, and the whale optimization algorithm (WOA) is used to optimize the characteristic attributes of ship equipment to achieve accurate knowledge reasoning.

Results The experimental results show that the proposed improved KD tree algorithm (ω-KDtree-WOA) achieves an inference accuracy of 0.928 when the K value is 4 and the population size is 400, which is approximately 3.2% higher than that of the traditional KD tree algorithm under the same conditions. In addition, compared with the classification confidence weighted and distance weighted K-nearest neighbor algorithm (CCW-WKNN) and the smoothing weight distance to solve K-nearest neighbor (SDWKNN) algorithm, etc., the algorithm proposed in this paper has significant advantages in accuracy, recall, precision, and F1 score, especially showing stronger stability when the K value is larger.

Conclusion The proposed method can be effectively applied to the operation and maintenance process of ship gas turbines.