PSI - Issue 28

Adrian Loghin et al. / Procedia Structural Integrity 28 (2020) 2304–2311

2305

2

A. Loghin et al. / Structural Integrity Procedia 00 (2020) 000–000

deterministic predictions for life assessment with minimal e ff ort and to generate an accurate solution base for training machine learning models that can be used more e ffi ciently (low runtime) in probabilistic type assessments. Three di ff erent mixed mode (mode I and mode II) experimental datasets were considered by Loghin and Ismonov (2020) to validate this approach and, in the study presented herein, a fourth experimental measurement (Berrios and Franco (2018)) is considered. 3D FEA coupled with surrogate modeling was also employed by Leser et al. (2017) to demonstrate a path for development of more e ffi cient simulation capabilities suited for Digital Twin - Structural Health Management appli cations. Similar to the procedure presented herein, Leser et al. (2017) made use of mixed mode experimental data for validation purposes which is an important addition to any modeling development. Usage of di ff erent surrogate model types reported in the literature is discussed by Shantz. (2010) along with employing FEM solutions in training of such models. Gaussian Process is a surrogate model type that was employed by Shantz. (2010) and Loghin and Ismonov (2019) for planar crack growth predictions and, by Leser et al. (2017) for mixed mode conditions (mode I and mode II). RBF response surface modeling was considered in the work of Loghin and Ismonov (2020) to predict out-of-plane crack path and RUL and further validate it against experimental data sets.

Nomenclature

RUL Remaining Useful Life FEA Finite Element Analysis FEM Finite Element Modeling 3D Three-dimensional CAD Computer Aided Design RBF Radial Basis Function RS Response Surface R Cyclic stress ratio DOE Design of Experiment GP Gaussian Process K 1

Mode I Stress Intensity Factor

2. Modeling Procedure

2.1. 3D FEA Procedure

A three-dimensional finite element representation of the specimen used in the test procedure (sample #2) provided by Berrios and Franco (2018) was used to explicitly simulate crack propagation process. The application used in this development is SimModeler Crack, a tool designed to perform fatigue crack growth simulations using component level CAD models. Details of this development are provided in Loghin (2018) and not repeated herein. The finite element simulation was performed using rigid body motion constraints, isotropic linear elasticity (E = 20.6E4 MPa, ν = 0.3) and a constant amplitude loading (R = 0). Since the initial edge crack was not planar, the 2D measurement of the initial precrack provided by Berrios and Franco (2018) was used to create a representative replica of the initial edge crack in the 3D model. Using the physical measurement from the sample to set up an accurate initial crack in the model is very important in this assessment since orientation of the initial crack can be a source of crack path variability. The 3D geometry of the specimen containing the initial crack along with a typical mesh used in the crack growth simulation is provided in Figure 1. The post-processing step is performed in SimModeler where stress intensity factors are computed using displacement correlation technique (Ingra ff ea and Manu (1980)) and maximum tangential stress criteria is used to determine crack growth direction (Erdogan (1963)). Using nominal geometry, two di ff erent modes for applying loading on the pin holes are considered for initial assessment: using rigid region - constraint equations (CERIG) in Ansys following similar model setup described in Berrios and Franco (2018) and the second, uniform loading along the pin holes where the pins engage the specimen.