What question did this study set out to answer?

The aim is to enhance construction health and safety by predicting the severity of ship-bridge pier collisions under various conditions.

April 12, 2026Open Access

Real‐Time Machine Learning Ship and Bridge Pier Collision Prediction to Enhance Construction Health and Safety

Key Points

The aim is to enhance construction health and safety by predicting the severity of ship-bridge pier collisions under various conditions.
Developed a hybrid finite element-machine learning framework for prediction.
Conducted nonlinear finite element simulations based on parameters like ship tonnage, speed, and collision angle.
Trained machine learning models (random forest and neural networks) using finite element datasets for rapid prediction.
Evaluation included an 80/20 train-test split and fivefold cross-validation on a large synthesized dataset.
Increased ship speed from 5 to 20 m/s led to a fourfold increase in pier displacement.
Peak impact forces increased significantly from 0.8 MN to 3.5 MN with the same tonnage range.
Large tonnage collisions generated forces up to 5.7 MN, indicating significant stress concentrations.
Energy absorption increased from 180 kJ to 650 kJ under varying collision conditions.
Machine learning models achieved high predictive accuracy, with R² values close to 1.0.

Abstract

Ship–bridge pier collisions signify a serious risk to structural integrity and construction health and safety during bridge construction phases where structural redundancy is limited. This study develops a hybrid finite element–machine learning (FE–ML) replacement framework for rapid prediction of collision consequence severity under different maritime and environmental conditions. Nonlinear finite element (FE) simulations were conducted across a parametric domain defined by ship tonnage, velocity, collision angle and pier geometric characteristics to quantify structural responses. It includes displacement, peak impact force, stress distribution, energy absorption and acceleration behaviour. FE results demonstrate strong nonlinear dependence of structural response on kinetic energy transfer. The increased ship speed from 5 to 20 m/s produced approximately fourfold growth in pier displacement from 0.08 to 0.30 m for a 16,000‐ton vessel. Moreover, the peak impact force increased from 0.8 MN to 3.5 MN under the same tonnage range. The large tonnage collision scenarios (50,000 tons) generated forces up to 5.7 MN along with stress concentrations approached 90 MPa at the pier base. Energy absorption capacity increased substantially from 180 kJ for moderate impacts to 650 kJ under severe collision conditions. This confirms the dominance of velocity and vessel mass in governing structural damage mechanisms. ML models, random forest regression and feedforward neural networks (NNs) were trained using FE‐generated datasets to enable rapid consequence prediction. Baseline evaluation using an 80/20 train–test split yielded strong predictive capability with coefficients of determination of 0.93 (random forest regressor RFR) and 0.95 (NN) during training (0.89) and testing (0.91). Expanded fivefold cross‐validation on a synthesized dataset ( N = 50,000) produced near‐unity regression accuracy and achieved mean R 2 values of 0.9980 for RFR and 0.9988 for NN with minimal prediction error dispersion. The proposed FE–ML framework enables near real‐time estimation of ship collision consequence severity and establishes a direct linkage between navigation parameters and structural response demand. The results support implementation within construction health and safety management systems for rapid hazard screening, protective design optimization and proactive maritime traffic control.

Real‐Time Machine Learning Ship and Bridge Pier Collision Prediction to Enhance Construction Health and Safety

Key Points

Abstract

Cite This Study