A method for measuring the dynamic state of heavy-duty shipbuilding engineering vehicles based on multi-source sensor data and a dual expert framework

In heavy-duty shipbuilding, the efficient transfer of ship blocks directly affects the overall construction schedule, while reliable identification of the lifting and dropping states of transport vehicles is essential for updating block locations. However, vehicle vibration, road disturbances, and local sensor anomalies can cause substantial fluctuations in multi-source measurement signals, limiting the recognition accuracy of conventional hard-threshold methods. To address this issue, a knowledge-driven dual-expert machine learning framework with physical gating is proposed. The standard deviation of vehicle speed is used as the physical gating indicator to partition and route data under different operating conditions. A 16-dimensional domain-informed hybrid feature space is then constructed, and LightGBM and CatBoost are employed as the steady-state and dynamic experts, respectively. Experiments on real-world engineering data show that the proposed method achieves a Macro-F1 of 0.9938 and an Accuracy of 0.9972. Measurement-noise enhancement experiments and tests on unseen vehicles with different rated tonnages further demonstrate its robustness and cross-vehicle generalization capability. t-SNE and SHAP analyses reveal distinct feature dependencies between the two experts, which are generally consistent with the physical mechanisms of lifting operations and driving disturbances. The proposed method provides an effective approach for reliable operational-state measurement of heavy-duty engineering vehicles under complex operating conditions.