Multifidelity Model-Assisted Probability of Detection Via Cokriging


This work introduces multifidelity metamodeling for reliability analysis of nondestructive testing (NDT) systems. Specifically, the Cokriging metamodel is utilized to accelerate the uncertainty propagation within model-assisted probability of detection (MAPOD) analysis of ultrasonic testing (UT) systems. The Cokriging multifidelity metamodel fuses a limited amount of data obtained from high-fidelity (HF) physics-based UT models, which are accurate but time-consuming to evaluate, with a conservative to large amount of data from low-fidelity (LF) physics-based UT models, which are less accurate but faster to evaluate. The resulting Cokriging metamodel is fast to evaluate and yields an accurate estimate of the output of the HF model. The proposed approach is demonstrated on three benchmark UT MAPOD cases involving spherically-void and pill-box shaped defects in flat aluminum plates using planar and focused transducers. The results show that a two-level Cokriging metamodel is capable of yielding estimations of the HF model that are globally accurate within 1% of the standard deviation of the testing points. Furthermore, the result show that the Cokriging metamodeling approach needs around one order of magnitude fewer training data points when compared to the current state-of-the-art approaches that rely on metamodels constructed with Kriging.


Mechanical and Aerospace Engineering


Iowa State University, Grant None

Keywords and Phrases

Cokriging; Model-assisted probability of detection; Multifidelity modeling; Nondestructive testing; Ultrasonic testing simulation; Uncertainty propagation

International Standard Serial Number (ISSN)


Document Type

Article - Journal

Document Version


File Type





© 2023 Elsevier, All rights reserved.

Publication Date

01 Dec 2019