Bayesian Network‐Based Uncertainty Quantification of Ship Isolation System Using Multi‐Level Data

The shock response of ship isolation systems is directly related to the safety of gas turbines and associated equipment. However, due to the high complexity of system‐level shock testing, existing experimental data are sparse and highly scattered, posing significant challenges for reliability assessment. To address this issue, this study proposes a multi‐level shock response analysis method to quantify the uncertainty in system outputs. An isolation system composed of three vibration isolators and one mechanical stopper was designed based on the actual layout of a naval vessel. For the first time, a vertical dual‐pulse shock testing machine was employed to conduct the experiments. The isolation system was decomposed into three hierarchical levels: single‐impact tests on the first‐level isolators, dual‐pulse shock tests on the second‐level isolators, and system‐level dual‐pulse shock tests. A Bayesian network model was developed to propagate the calibrated parameter distributions from the lower levels to the system level, while accounting for correlations among multiple output variables. Model validation indicates that multi‐level calibration improves the reliability estimation accuracy by 17.64% and 6.05%, respectively, compared to single‐level models. Furthermore, a novel method for quantifying shock inputs was introduced and integrated into a global sensitivity analysis. The predicted system‐level responses based on the integrated model show good agreement with the experimental data. Although the algorithms used are standard methods, their integrated application to multi‐layered structural systems in the context of naval engineering is a novel contribution. This approach provides a new perspective for uncertainty analysis in shock testing of isolation systems.