Machine learning and bayesian network based on fuzzy AHP framework for risk assessment in process units

Area of Science:

• Chemical Engineering
• Process Safety
• Artificial Intelligence

Background:

• Risk assessment is vital for process unit safety.
• Artificial intelligence (AI) offers advanced capabilities for risk prediction and assessment.
• Integrating AI with traditional methods can improve risk assessment precision.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning algorithms in process risk assessment.
• To compare conventional statistical methods with advanced AI techniques.
• To explore the synergistic potential of AI with Bayesian networks and Multi-Criteria Decision Making (MCDM) for risk prioritization.

Main Methods:

• Utilized a dataset of 160 deviations identified via Hazard and Operability (HAZOP) studies.
• Employed a diverse range of algorithms: ensemble methods (Random Forest, Hist Gradient Boosting, XGBoost, CatBoost) and traditional methods (Logistic Regression, KNN, SVM, CNN).
• Applied a fusion of Bayesian networks and MCDM for risk option prioritization.

Main Results:

• Random Forest, XGBoost, and CatBoost demonstrated superior performance, achieving near-perfect AUC scores and accuracy.
• The combined approach of Bayesian networks and MCDM identified "Corrosion in Electrolysis Cells" and "Damage and Explosion of Cells" as high-priority risks.
• Machine learning models significantly outperformed traditional methods in accuracy and predictive power.

Conclusions:

• Machine learning techniques are highly effective tools for process risk assessment.
• The integration of Bayesian networks and MCDM provides a robust framework for prioritizing risks.
• These methodologies enable the implementation of targeted control and preventive measures for enhanced industrial safety.