From Low-Cycle Fatigue Lifetime Data to its Probabilistic Uncertainty Quantification

Vorschau

22 Min. Untertitel (CC)

Towards a better understanding and better lifetime prediction of low-cycle fatigue data, sophisticated lifetime and damage assessment methods are continuously developed [1]. However, a single lifetime estimation inherits the aleatoric and epistemic uncertainties from the material data, the measurement process and the model imperfections. Suitable probabilistic methods overcome these issues and provide uncertainty quantification and the reliability assessment of the lifetime prediction while introducing a huge but automatable computational effort. The digitized workflow starts with measuring the material response of strain-controlled low-cycle fatigue experiments at high-temperature conditions and different loading scenarios. Further, the usage of appropriate ontologies, the collection of metadata and measurement data into the hierarchical data format (HDF), and the corresponding difficulties arising from proprietary data formats are shown.

The presented lifetime prediction is based on the continuum damage mechanics approach and adequate differential equations that govern the damage evolution [2]. Methods of statistical inference determine the corresponding model parameters. The parameter uncertainties are statistically estimated by bootstrap sampling over blocks [2], while a single block represents a specific interval of low-cycle fatigue data. The necessary computational effort increases drastically with an increasing number of experiments and the duration of a single experiment exposed to high-temperature conditions.

According to this procedure, many hundreds of parameter optimizations are necessary to provide the desired probabilistic information. This task is tackled using a high-throughput-computing workflow using the open-source software HTCondor [3]. This software allows the submission and organization of simulations tasks to larger subsequent computing workflows designed as directed acyclic graphs (DAG). Finally, the applied DAG and the desired output is explained: a probabilistic lifetime assessment considering component-near loading situations.

References

[1] X. Cai, P. Steinmann, X. Yao, and J. Wang, Thermodynamic formulation of a unified multi-mechanism continuum viscoplastic damage model with application to high-Cr steels. International Journal of Plasticity 114, 15–39 (2019)

[2] F. Kölzow, M. Saadi, C. Kontermann, H. Gottschalk, and M. Oechsner, AVIF-No. A316 Probabilistic Lifetime Model Comparison – Creep-Fatigue, Interim Report, FVV, Forschungsvereinigung der Arbeitsgemeinschaft der Eisen und Metall verarbeitenden Industrie e.V. (AVIF) (2022).

[3] D. Thain, T. Tannenbaum, and M. Livny, Distributed computing in practice: the Condor experience. Concurrency - Practice and Experience 17, 2005, 323–356 (2005)

Abstract

Erwerben Sie einen Zugang, um dieses Dokument anzusehen.

Dr.-Ing. Felix Kölzow