Advancing sustainable shipbuilding through a hybrid risk prioritization model: integrating FMEA and machine learning

Effective risk management is critical in modern shipbuilding, where production complexity continues to grow. Traditional Failure Modes and Effects Analysis (FMEA), despite its widespread use, often lacks objectivity and scalability. This study introduces a novel hybrid methodology that integrates FMEA with machine learning (ML), specifically, Random Forest (RF) regression, to enhance failure prediction and defect prioritization within a real-world shipyard context. The model was trained on 489 documented defects from the construction of an LNG-powered fishing vessel at Cemre Shipyard. Each defect was assessed using four risk factors: cost, time, frequency, and stage. Risk Priority Numbers (RPNs) were computed accordingly and used as target values in model training. The framework also incorporates Pareto analysis and feature importance evaluations to identify dominant risk contributors. The ML model achieved high predictive accuracy (Coefficient of Determination (R²) = 0.9738; Mean Absolute Error (MAE) = 1.3470) under current operational conditions. Deformation, inappropriate production, and defective part usage were identified as the most critical categories. Time loss and frequency emerged as the most significant features influencing RPNs. Improvement scenarios revealed the model’s robustness and capacity to estimate risk reduction potential for high-priority failure modes. This hybrid approach bridges expert judgment with data-driven intelligence and offers a scalable, objective framework for real-time quality control. Its potential for integration with enterprise systems suggests broader industrial applications, including automated risk monitoring and continuous improvement. The results demonstrate that combining FMEA with ML can significantly advance predictive defect management in maritime manufacturing.