JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Multi-objective global sensitivity analysis of shipborne equipment based on AGPSO-CNN-BiLSTM-attention

  • 0School of Mechanical Engineering, Shenyang University of Technology, Shenyang, 110870, China.
Scientific Reports +

|

November 20, 2025

|

November 20, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces an optimized deep learning model (CNN-BiLSTM-Attention) for efficient global sensitivity analysis of complex shipborne equipment. The approach significantly reduces computational cost while accurately identifying critical design parameters.

Area Of Science

  • Computational mechanics
  • Structural engineering
  • Machine learning applications

Background

  • Global sensitivity analysis of complex structural systems, like shipborne equipment, is computationally expensive.
  • Efficient evaluation of multi-input multi-output (MIMO) systems is crucial for structural design and optimization.
  • Existing methods often struggle with high dimensionality and computational burden.

Purpose Of The Study

  • To develop a computationally efficient method for global sensitivity analysis of large shipborne equipment.
  • To optimize hyperparameters of a deep learning model using an advanced metaheuristic algorithm.
  • To establish a framework for MIMO sensitivity evaluation using integrated deep learning and established sensitivity analysis methods.

Main Methods

  • A CNN-BiLSTM-Attention model was developed and optimized using adaptive Genetic Algorithm-Particle Swarm Optimization (AGPSO) for hyperparameter tuning (learning rate, batch size, L2 regularization).
  • The optimized model was integrated with Sobol' global sensitivity analysis to create a MIMO sensitivity framework.
  • The proposed approach was applied to a gas turbine isolation system to evaluate its effectiveness.

Main Results

  • The AGPSO-CNN-BiLSTM-Attention model achieved high training accuracies (99.49% for relative displacement, 99.12% for absolute acceleration).
  • The integrated framework efficiently quantified parameter influence in the MIMO system.
  • Critical isolator and limiter parameters affecting shock response in the gas turbine isolation system were successfully identified.

Conclusions

  • The proposed AGPSO-CNN-BiLSTM-Attention model significantly enhances prediction performance and computational efficiency for sensitivity analysis.
  • The integrated Sobol' method and deep learning framework provides a valuable tool for structural design and optimization of marine equipment.
  • This approach offers improved efficiency and generalization capabilities for complex system analysis.

Keywords: