From Uncertainty to Decision: An Intelligent Predictive Maintenance Approach Leveraging Fuzzy Bayesian Networks for Critical IT Components

Abstract

Information Technology (IT) systems are essential for ensuring the continuous availability of digital services, yet they remain vulnerable to unexpected failures that can disrupt operations and generate significant costs. To address these risks, predictive maintenance (PdM) strategies are required to anticipate failures and preserve reliable performance. This study develops a PdM approach for critical IT systems. A hybrid methodology is introduced, combining Bayesian networks (BNs) to model intricate causal dependencies among various system parameters and fuzzy logic (FL) to calculate the conditional probabilities, which are then integrated into the BNs to accurately predict potential system failures. Model validation was performed through the verification of three axioms. Six degradation scenarios are then generated, followed by two sensitivity analyses: a partial analysis to identify key parameters influencing individual components, and a global analysis to determine parameters affecting overall system degradation. Finally, an event tree analysis is applied to evaluate the effectiveness of different maintenance strategies, providing insights to minimize failure risks and improve system reliability. The results indicate that the system enters a critical state once a major component fails, and the risk of total system failure becomes almost certain when multiple components degrade simultaneously.