Bridging tacit knowledge loss and data scarcity of coke drum integrity management using a multi-stage hybrid decision-analytic framework

Maintaining the structural integrity of coke drums in delayed coking units remains a critical challenge for refineries due to severe cyclic thermal and mechanical loading that promotes complex damage mechanisms, including thermomechanical fatigue, shell bulging, and weld cracking. Effective integrity management is further complicated by limited inspection data, fragmented historical records, high uncertainty, and the gradual loss of experienced subject-matter expertise. Conventional data-driven artificial intelligence approaches often require extensive datasets and are therefore unsuitable for such data-constrained environments. To address these challenges, the research was structured in two complementary parts: Part I focused on the development of a novel Integrity Management with Validation and Verification (IMV 2 ) methodology, while Part II concentrates on its application, validation, and verification through an industrial coke drum case study. This study presents and validates an integrated Integrity Management with Validation and Verification (IMV 2 ) framework designed to support integrity management decision-making under conditions of data scarcity. The framework combines the Analytic Hierarchy Process (AHP), Fuzzy AHP (FAHP), Decision-Making Trial and Evaluation Laboratory (DEMATEL), Bayesian Networks (BNs), ranked-node probability modelling, and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) within a unified decision-support architecture. The methodology systematically captures tacit expert knowledge, quantifies causal relationships among degradation drivers, transforms qualitative judgments into probabilistic representations, and ranks integrity management strategies through a transparent and auditable process. The proposed framework was validated through a real refinery coke drum case study, demonstrating its capability to integrate sparse inspection information with expert judgment while effectively handling uncertainty and missing data. Results show that the model successfully captures the complex interdependencies among failure modes and produces strategy rankings consistent with established engineering practice. Sensitivity, robustness, and statistical analyses confirm the stability of BN-based outputs despite significant data limitations. Furthermore, convergence between probabilistic assessment, sensitivity analysis, and Minkowski-based mathematical verification demonstrates the framework’s global robustness. Observed rank variations were limited to alternatives that were mathematically and statistically indistinguishable, indicating that such variations arise from the underlying data structure rather than methodological deficiencies. Overall, the IMV 2 framework provides a reliable, interpretable, and defensible approach for high-risk refinery asset management. By enabling dynamic updates, preserving organisational knowledge, and establishing a rigorous audit trail for decision-making, it offers a practical blueprint for integrity management in environments characterised by uncertainty, limited data availability, and complex failure mechanisms.