WEB Data-driven modelling workflows for crack growth under complex loading scenarios
Increasing thermal and mechanical loads on components operating at high temperature (e.g. gas turbine blades) result in a growing demand for the implementation of damage tolerant design approaches. These approaches highly rely on accurate predictions of the crack growth under service conditions in order to define inspection intervals and ensure component reliability.
However, the experimental campaigns usually required to calibrate and validate such models are time consuming and cost intensive. Especially reliable measurements of crack growth rates under thermo-mechanical fatigue (TMF) loadings are scarce and difficult to obtain from an experimental point of view. The issue of transferability onto real component geometries is often disregarded in the context of TMF crack growth modelling due to the already high complexity of the subject. The introduction of additive manufacturing processes with faster innovation cycles and tailored material properties further complicate these challenges
It has been shown that linear-accumulative models are a very effective tool to describe TMF crack growth under service conditions and have a proven transferability onto components, but the calibration of the model parameters requires intensive test campaigns. For a number of high temperature alloys such as IN718, exists a literature data base containing creep crack growth, fatigue crack growth, and combined load cycle measurements at air and in vacuum. This data base should in theory contain all necessary information for the optimized derivation of accumulative crack growth laws, assuming fatigue, creep and oxidation as the most prominent drivers of crack growth.
In this paper, different workflows to deduct data-driven accumulation crack growth models based on optimization and data science methods are presented. More than 200 data set for isothermal fatigue crack growth, creep crack growth and crack growth under complex load cycles, all obtained on different versions of the “role-model” high-temperature alloy IN718, have been comprised in a machine-readable data repository as training data, also including meta-information such as load ratios or grain sizes. The presented workflows allow to quickly assess different model formulations, calibrate model parameters for specific load conditions or even deduct undiscovered correlations to generate new description variants by using AI methods.
|Category||Short file description||File description||File Size|
|Extended Abstract||Graphical Abstract||This is an extension of the abstract with a small graphic for visualization||536 KB||Download|