AI-Enabled Lifecycle Integrity Management of LPG Pressure Vessels: A Critical Review from Design and Inspection to Digital Twins and Risk Assessment

  • O. AKPOTU Federal University of Petroleum Resources, Effurun, Delta State
  • F. I. ASHIEDU Federal University of Petroleum Resources, Effurun, Delta State
  • E. EMAGBETERE Federal University of Petroleum Resources, Effurun, Delta State
Keywords: Machine learning, Predictive maintenance, Pressure vessel integrity, Regulatory qualification, Structural health monitoring

Abstract

Liquefied petroleum gas (LPG) pressure vessels require reliable integrity
management because degradation, fabrication defects, corrosion, fatigue,
overpressurisation, and mechanical damage can lead to loss of containment and
severe fire or explosion consequences. This review critically evaluates the
application of artificial intelligence (AI) across the lifecycle of metallic LPG
pressure vessels, from design and materials selection to manufacturing, inspection,
structural health monitoring, predictive maintenance, risk assessment, logistics,
and digital twins. A PRISMA 2020-based literature search was conducted across
Scopus, Web of Science, IEEE Xplore, ScienceDirect, SpringerLink, and Google
Scholar for publications from 2015 to 2026. The review distinguishes direct
evidence of metallic LPG vessels from evidence transferred from pressure-vessel
and other adjacent engineering applications. The literature shows that machine
learning, deep learning, physics-informed neural networks, Bayesian networks,
and digital-twin frameworks can support design optimisation, automated defect
detection, continuous condition monitoring, and degradation prediction, dynamic
risk updating, and accelerated consequence modelling. However, the evidence base
remains fragmented, with many studies relying on generic pressure vessels,
composite or hydrogen-storage systems, simulations, laboratory datasets, or
synthetic data rather than operational LPG vessels. Key barriers to deployment
include limited representative industrial datasets, class imbalance, model
interpretability, uncertainty quantification, sensor and data-quality issues,
cybersecurity, edge-computing constraints, and the absence of established
procedures for AI verification and regulatory qualification. The review identifies
physics-informed and explainable AI integrated with validated digital twins as a
promising pathway for lifecycle integrity management, while emphasizing the need
for LPG-specific datasets, independent field validation, and standardized
qualification frameworks.

Author Biographies

O. AKPOTU, Federal University of Petroleum Resources, Effurun, Delta State

Department of Mechanical Engineering, Federal University of Petroleum Resouces, Effurun,
Nigeria

F. I. ASHIEDU, Federal University of Petroleum Resources, Effurun, Delta State

Department of Mechanical Engineering, Federal University of Petroleum Resouces, Effurun,
Nigeria

E. EMAGBETERE, Federal University of Petroleum Resources, Effurun, Delta State

Department of Mechanical Engineering, Federal University of Petroleum Resouces, Effurun,
Nigeria

Published
Y-m-d