Model discrepancy calibration across experimental settings. (August 2020)

• Record Type: Journal Article
• Authors: Maupin, Kathryn A.; Swiler, Laura P.
• Is Part Of: Reliability engineering & system safety. Volume 200(2020)
• Publication Date: 2020-08
• Subjects: Computational modeling -- Model inadequacy -- Model form error -- Model discrepancy -- Inverse problems -- Bayes' Rule; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2020.106818
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: An approach for calculating the discrepancy between a model and data is presented. The discrepancy is a function of experimental setting and the spatiotemporal field. This discrepancy model uses an additive bias correction and a Gaussian process model. The evaluation of the discrepancy is performed in a two step process. Two case studies illustrate the feasibility of the proposed approach. Abstract: Despite continuing advances in the reliability of computational modeling and simulation, model inadequacy remains a pervasive concern across scientific disciplines. Further challenges are introduced into the already complex problem of "correcting" an inadequate model when experimental data is collected at varying experimental settings. This paper introduces a general approach to calibrating a model discrepancy function when the model is expected to perform for multiple experimental configurations and give predictions as a function of temporal and/or spatial coordinates.