Resilience Engineering Framework for Crude Oil Refineries

Abstract

Crude oil refineries generate vast, multidimensional data streams that are critical for maintaining real-time operational control and ensuring compliance with environmental regulations. Distillation columns, in particular, produce intricate and evolving datasets that create significant challenges for early failure detection, escalating the risk of environmentally harmful incidents. To effec- tively tackle this pressing issue, we advocate for a robust resilience-engineering framework that seamlessly integrates the Lagrangian Support Vector Machine (LSVM) with the Functional Resonance Analysis Method (FRAM) within cutting-edge environmental infor- matics architecture. The LSVM component rigorously analyzes refinery operational data — such as pressure, temperature, flow rates, and reflux ratios using an 80:20 train-test split and consistently achieves a remarkable 96% accuracy rate in classifying failure events. Beyond merely identifying anomalies, this environmental informatics approach leverages additional data sources, including alarm logs and weather information, to enable proactive predictive analytics for forecasting environmental risks. The FRAM operator then deci- sively evaluates the anomalies identified by the LSVM through functional resonance modeling, which effectively traces system-level variability and uncovers potential cascading failure pathways, such as tray flooding, over-pressurization, or off-spec discharges. This in- tegrative approach not only enhances interpretability and facilitates early intervention, but it also fortifies environmental protection by linking data-driven fault detection with actionable decision-making insights for proactive risk mitigation. Ultimately, this proposed framework firmly advances operational resilience and reinforces sustainability objectives while ensuring unwavering regulatory com- pliance within refinery processes.

Keywords: failure analysis, functional resonance analysis method, distillation column, support vector machine