Fatigue in Materials: Fundamentals, Multiscale Modeling and Prevention : Data-Driven Investigations of Fatigue
Sponsored by: TMS Structural Materials Division, TMS: Advanced Characterization, Testing, and Simulation Committee, TMS: Computational Materials Science and Engineering Committee, TMS: Mechanical Behavior of Materials Committee
Program Organizers: Ashley Spear, University of Utah; Jean-Briac le Graverend, Texas A&M University; Antonios Kontsos, Drexel University; Tongguang Zhai, University of Kentucky
Monday 8:30 AM
February 27, 2017
Location: San Diego Convention Ctr
Session Chair: Ashley Spear, University of Utah
Nondestructive Evaluation as a Link between Fatigue Diagnostics and Prognostics: Brian Wisner1; Ryan Whitmore1; Konstantinos Baxevanakis1; Antonios Kontsos1; 1Drexel University
To link diagnosis to prognosis of fatigue behavior of materials, a multiphysics nondestructive evaluation approach is presented. Specifically, Acoustic Emission and Digital Image Correlation were combined so that real time acquisition of nondestructive data related to microstructure-level damage can be fed into computational mechanics models. To achieve this goal, Acoustic Emission data were combined with Outlier Analysis to define damage laws by identifying the least amount of features that contained the most significant information about the behavior and failure of the tested materials. These data-driven damage laws were defined as a function of measured strain, which was computed by full field deformation maps. Once the damage law was defined, it was fed into computational finite elements models that include both plasticity and damage considerations. This model was then used in computational problems as the constitutive material law that takes into account progressive damage defined by multiphysics nondestructive data.
8:50 AM Invited
Linking Length Scales during Fatigue: Investigating the Effect of Microscale Strain Localization on Macroscopic Response: Samantha Daly1; 1University of California at Santa Barbara
This talk will present our recent work on exploring the connections between the macroscopic and microscopic length scales, and the role microscale strain localization plays in failure. This research utilized a combination of distortion-corrected digital image correlation and scanning electron microscopy to measure deformation fields at small length scales, including a creation of functionalized nanoparticles for sub-grain deformation tracking. These approaches enable us to glean critical insights into material behavior, including the impact of constituent architecture on damage accumulation in aerospace composites and the relationship between processing and performance in metallic alloys. Recent studies examining microscale damage accumulation in ceramic matrix composites under dwell fatigue at 800C will be discussed as an illustrative example of these emerging experimental approaches and associated analysis. The talk will also include a discussion of on-going work to develop unsupervised learning approaches that statistically link microscale deformation behavior and microstructural attributes.
Strain Field Mining of Cyclic Damage Accumulation in Nonwoven Fiber Composites: Yoon Joo Na1; James Collins1; Christopher Muhlstein1; 1Georgia Institute of Technology
Paper, the most representative nonwoven fiber composite, is an orthotropic fiber network material with performance that is controlled by inter-fiber bonding. Although paper’s mechanical properties have been studied for many decades, some researchers have maintained that it does not truly fatigue. In this presentation we will present results from cyclic deformation and fatigue crack growth experiments conducted on as-fabricated and notched paper specimens in two orthogonal orientations. Non-contact digital image correlation strain fields were mined using lineal path and point correlation functions to reveal in-plane damage evolution and accumulation trends in the material, and their relationships to cyclic and environmentally driven damage accumulation processes.
9:30 AM Invited
Correlation of Microstructural Configurations to Fatigue Indicator Parameters: Sushant Jha1; Robert Brockman2; Rebecca Hoffman2; Vikas Sinha3; William Porter2; Dennis Buchanan2; Adam Pilchak4; James Larsen4; Reji John4; 1US Air Force Research Laboratory/Universal Technology Corporation; 2University of Dayton Research Institute; 3UES, Inc.; 4US Air Force Research Laboratory
Fatigue cracks often initiate due to combinations of microstructural features. For the same nominal microstructure and remote stress level, there can be multiple possibilities for grain and neighborhood combinations that accumulate fatigue damage and produce a crack. Machine learning based approaches can be useful in exploring various microstructural combinations and elucidate their relationship to crack initiation. In this study, statistically representative instantiations of two microstructures of Ti-6Al-2Sn-4Zr-2Mo were generated using DREAM.3D. A Crystal Plasticity-Finite Element Method model was used to calculate various grain-level deformation metrics on basal, prism, and pyramidal slip systems. Principal Component Analysis and fuzzy C-means clustering algorithms were applied in order to determine independent combinations of grain and neighborhood metrics that correlate with strain-based Fatigue Indicator Parameters, as well as experimentally quantified crack-initiation neighborhood parameters. The analysis also provided insights into the probability of occurrence of life-limiting fatigue failures with respect to microstructure and stress level.
Investigation of Neighborhood Effects on Crack Initiation Sites in Different Ti Microstructures: Vahid Tari1; Michael Groeber2; Adam Pilchak2; Anthony Rollett1; 1Carnegie Mellon University; 2Air Force Research Laboratory (AFRL/RXCM)
Fatigue is a common and important failure mode in aerospace components. Many of these components are made of Titanium alloys because of their excellent mechanical properties. Their fatigue properties depend on the control of microstructure including grain size and morphology. To investigate the effect of varying the microstructure, we used a spectral full field elasto-viscoplastic method based on FFT to simulate the micro-mechanical response. A large number of three-dimensional synthetic input microstructures were generated based on experimental evidence. The results show that critical neighborhoods can be identified based on analysis of crack initiation sites.
10:10 AM Break
Toward the Use of Machine Learning to Understand and Predict Microstructurally Small Fatigue-Crack Evolution: Nathan Wilkinson1; Brian Phung1; Jacob Hochhalter1; Ashley Spear1; 1University of Utah
Improving structural prognosis capabilities for fatigue-critical components requires accurate rules for predicting growth of microstructurally small fatigue cracks (MSFCs). With simultaneous advancements in techniques for measuring and modeling material deformation and failure in 3D comes the challenge of post-processing, combining, and contextualizing large amounts of data from both experimental observations and numerical simulations. This talk will describe recent efforts to correlate, and to predict relationships among, intrinsic microstructural features, micromechanical fields, and crack-growth behavior within local neighborhoods along 3D crack fronts in a polycrystalline Al-Mg-Si alloy. Supervised machine learning is employed to map microstructural and micromechanical parameters to observed fatigue-crack growth rate and direction for discrete points along measured crack fronts. A non-trivial portion of the work involves parameterizing the descriptions of both microstructural and micromechanical neighborhoods as well as crack growth in three dimensions.
10:50 AM Invited
Correlating Experiments and Simulations to Develop Predicative Capabilities for Fatigue Crack Initiation in Ni-based Superalloys: Jun Jiang1; Fionn Dunne1; T Ben Britton1; 1Department of Materials, Imperial College
In many components, fatigue crack initiation represents one third of component life. Developing a predictive approach for microstructurally sensitive fatigue crack initiation is therefore of significant value. In this talk we present an overview of recent work which draws together high fidelity experiments (predominantly deformation and inspection with HR-EBSD & HR-DIC) and physically based crystal plasticity modelling with this goal in mind. Our results build up through a series of carefully constrained geometries, focussing on single crystal, oligocrystal, polycrystal and polycrystal with non-metallic inclusions based systematic monotonic and fatigue tests and full deformation field comparison. Coupling models and experiments in this programme of work has enabled us not only to calibrate and validate our modelling approach, but also to improve our understanding of experimental results leading to new experiment designs.
A General Probabilistic Framework Combining Experiments and Simulations to Identify the Small Crack Driving Force: Andrea Rovinelli1; Michael Sangid1; Ricardo Lebensohn2; Wolfgang Ludwig3; Yoann Guilhem4; Henry Proudhon5; 1Purdue University; 2Los Alamos National Lab; 3ESRF; 4ENS de Cachan; 5MINES ParisTech
Identifying the small crack (SC) driving force of polycrystalline engineering alloys is critical to correlate the inherent microstructure variability and the intrinsic SC growth rate’s scatter observed during stage I. Due to recent experimental advancements, “cycle-by-cycle” data of a SC propagating through a beta-metastable titanium alloy are available via phase and diffraction contrast tomography. To compute micromechanical fields not available from the experiment, FFT-based crystal plasticity simulations are utilized. Results of the experiment and simulations are combined in a single dataset and sampled utilizing non-local mining technique. Sampled data are analyzed into a machine learning Bayesian Network framework to identify statically relevant correlations between state variables, microstructure features, location of the crack front, and experimentally observed growth rate, in order to postulate a data-driven, non-parametric SC driving force. Results are presented with a particular focus on the correlation between well-established SC driving forces and the proposed new driving force metric.
11:30 AM Cancelled
Cloud-based Data-driven Modeling for Fatigue Life Prediction in Ti-6Al-4V: Ayman Salem1; Joshua Shaffer1; Richard Kublik1; Daniel Satko1; 1Materials Resources LLC
To predict fatigue life in titanium alloys, it is crucial to account for the underlying microstructure (including chemistry, morphology, and crystallography) and spatial distribution of constituents which results from standard thermomechanical processing protocols. Despite the wealth of experimental results on both high-cycle and low-cycle fatigue of various titanium alloys, there have been limited studies on ways to analyze these findings. Some of the main reasons include: (i) the absence of repositories for all generated data (locally and globally), (ii) lack of data analytics tools that enable researchers to analyze published data which causes wasted research money to reproduce published data. In this talk, we will present a cloud-based data-driven modeling using data curation and non-linear interpolation models based on Bayesian neural networks (BNN) for predicting fatigue performance from High Cycle Fatigue (HCF) program and NIMS database under strain-controlled and load-controlled testing conditions. The associated web application will also be presented.