Manifold-based Conditional Bayesian network for aging pipe yield strength estimation with non-destructive measurements

Qiongfang Zhang, Nan Xu, Daniel Ersoy, Yongming Liu

Research output: Contribution to journalArticlepeer-review

8 Scopus citations


Deterioration of an aging pipeline network is a significant concern for the integrity and safety of the energy transportation infrastructure. Bulk yield strength is one of the most critical mechanical parameters for the risk assessment of aging pipes. Destructive testing for bulk yield strength is prohibitively expensive, and non-destructive testing with surface-only measurements is ideal for health management. A novel Manifold-based Conditional Bayesian network (MCBN) approach is presented to automatically classify steel type and predict yield strength for the aging pipe with non-destructive measurements. First, Uniform Manifold Approximation and Projection enhanced k-Nearest Neighbors algorithm is proposed as the first stage of MCBN to achieve dimension reduction, clustering visualization, and steel type classification. Following this, a Conditional Bayesian network (CBN) is learned and constructed in the second stage to interpret causal relationships and predict yield strength. Numerical experiments are conducted on experimental and synthetic data to verify and validate the proposed MCBN. Comparison with several existing methods is presented for the performance evaluation in steel type classification and yield strength prediction, which provides a foundation for the future reliability and risk assessment of aging pipes.

Original languageEnglish (US)
Article number108447
JournalReliability Engineering and System Safety
StatePublished - Jul 2022


  • Aging pipe
  • Bayesian network
  • Manifold learning
  • Reliability
  • Strength estimation

ASJC Scopus subject areas

  • Safety, Risk, Reliability and Quality
  • Industrial and Manufacturing Engineering


Dive into the research topics of 'Manifold-based Conditional Bayesian network for aging pipe yield strength estimation with non-destructive measurements'. Together they form a unique fingerprint.

Cite this