Predictive maintenance under uncertainty in smart hospitals: Decision support with belief functions

Modern hospitals increasingly depend on interconnected
biomedical devices and Internet of Medical Things (IoMT)
infrastructures to guarantee patient safety and continuity of care. Yet, Predictive Maintenance within Prognostics and Health Management (PHM) remains difficult to deploy in healthcare because data are heterogeneous, incomplete and sometimes contradictory, and because safety-critical decision making requires explicit uncertainty handling and transparency. This paper presents a modular PHM framework grounded in the belief function theory, also known as Dempster-Shafer theory, that represents ignorance explicitly, discounts sources by reliability, fuses heterogeneous evidence under conflict, and produces rational decisions via pignistic probabilities. The framework is tailored to hospital contexts and integrates practical considerations of equipment criticality, data governance, and regulatory constraints. A numerical
proof-of-concept oriented to an intensive care ventilator
illustrates the approach using three information sources (two
sensors and an expert assessment), with reliability-aware fusion
and conflict management. The results highlight how evidential
reasoning can transform noisy and partially conflicting hospital
data into actionable maintenance decisions, paving the way
toward resilient, explainable and standards-aligned PHM in
smart hospitals.