The analysis of ship structural strength based on machine learning algorithms

  Objectives  In the consideration that the traditional FEA method has problems of time-consuming, cost of computational resources, etc., a regression prediction model based on machine learning algorithms is presented for assessment of ship structural strength.

  Methods  Specifically, this model takes the data obtained by traditional FEM analysis for a local support structure of crane post under external loads on the main post of vessel, and is characterized by external loads and structural scantlings, with the stress and deformation as its targets. The model takes the existing finite element numerical calculation data as samples. Four traditional and four ensemble machine learning algorithms were introduced to predict and analyze the local structure response.

  Results  The experimental results show that the machine learning algorithms can provide solutions of high accuracy with a significant improvement of computational efficiency when compared with the traditional FEA method. Among these algorithms, the Light Gradient Boosting Machine(LightGBM) has the best performance with respect to the accuracy and efficiency.

  Conclusions  Furthermore, current study provides a feasible and efficient technical approach for further study of design of more complex hull structures.