Comparison of various uncertainty models with experimental investigations regarding the failure of plates with holes. (November 2020)

• Record Type: Journal Article
• Authors: Drieschner, Martin; Petryna, Yuri; Gruhlke, Robert; Eigel, Martin; Hömberg, Dietmar
• Is Part Of: Reliability engineering & system safety. Volume 203(2020)
• Publication Date: 2020-11
• Subjects: experimental and numerical investigations -- monomorphic uncertainty modeling -- polymorphic uncertainty modeling -- uncertainty quantification -- artificial neural networks -- failure analysis; 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.2020.107106
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: Geometrical and experimental inaccuracies lead to different failure mechanisms. Uncertainties have been considered in monomorphic and polymorphic uncertainty models. Simultaneous information about probabilities and possibilities can be extracted. A surrogate model based on Artificial Neural Networks (ANN) has been constructed. The same surrogate model is applicable independently of the uncertainty model. Abstract: Predicting ultimate limit states and failure mechanisms in real technical systems is essential. This task is a real challenge because of a variety of uncertainties. The result often depends very much on the way these uncertainties have been described. This contribution is a study of how the uncertainty models can affect such predictions and how well a prediction matches the behavior of a real system. Three uncertainty models are compared to a series of experiments on Plexiglas® plates with holes under uniaxial tension regarding the failure mechanism and the associated ultimate load. A plate with holes can represent many technical applications, such as the behavior of adhesive bonds in fiber composite structures with air voids. A stochastic model, a model based on fuzzy-set theory and a polymorphic uncertainty model are applied to point out the individual usefulness and the informative value of the resulting numerical predictions. The comparison shows that the polymorphic uncertainty model is more costly but simultaneously contains both information of the monomorphic uncertainty models. In order to overcome the computational costly uncertainty propagations, a surrogate model based on Artificial Neural Networks (ANN) is constructed independently of the uncertainty model.