Application of Gaussian process regression models for capturing the evolution of microstructure statistics in aging of nickel-based superalloys

Yabansu, Yuksel C.; Iskakov, Almambet; Kapustina, Anna; Rajagopalan, Sudhir; Kalidindi, Surya R.

doi:10.1016/j.actamat.2019.07.048

Cited by 71 publications

(28 citation statements)

References 88 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The three models considered include a fully connected NN, random forest (RF), and Gaussian process regression (GPR). The first two have shown good performance as property prediction models in previous studies on structure property prediction of energetic molecules [41,34,24] whereas GPR has been shown to work well on small data sets [43] . All known models to date have used a direct mapping between physicochemical features and material properties, in contrast to the present work where the mapping is directly from the latent vector.…”

Section: Methodsmentioning

confidence: 81%

Locally Optimizable Joint Embedding Framework to Design Nitrogen‐rich Molecules that are Similar but Improved

Balakrishnan

VanGessel

Boukouvalas

et al. 2021

Molecular Informatics

View full text Add to dashboard Cite

Deep learning has shown great potential for generating molecules with desired properties. But the cost and time required to obtain relevant property data have limited study to only a few classes of materials for which extensive data have already been collected. We develop a deep learning method that combines a generative model with a property prediction model to fuse small data of one class of molecules with larger data in another class. Common low‐level physicochemical properties are jointly embedded into a latent space that can be used to design molecules in the smaller class. The chemical space around the molecules in the training set is explored through local gradient ascent optimization. Based on nine molecules from the original training set, nine new molecules are found to have improved properties while remaining structurally similar to the training molecules thereby easing requirements for entirely new synthesis routes. Validation is performed using an equilibrium thermochemistry code to verify the molecules and target properties. A specific example targeting the Chapman‐Jouguet velocity and small data for nitrogen‐rich molecules is shown. Despite the relative lack of nitrogen‐rich molecule data, the results demonstrate that fusing and joint embedding with plentiful low nitrogen molecular data can produce higher generative performance than using the scarce data alone.

show abstract

Section: Methodsmentioning

confidence: 81%

Locally Optimizable Joint Embedding Framework to Design Nitrogen‐rich Molecules that are Similar but Improved

Balakrishnan

VanGessel

Boukouvalas

et al. 2021

Molecular Informatics

View full text Add to dashboard Cite

show abstract

“…14,15 The efficacy of the MKS framework in producing high-fidelity reduced-order PSP linkages has been demonstrated in prior work. [16][17][18] There exist many opportunities for the further development of the MKS framework. Of particular relevance to this work are the approaches used to build the reduced-order models after the feature engineering step (i.e., after the low-dimensional representations of the material structure are obtained employing n-point spatial correlations and PCA).…”

Section: Introductionmentioning

confidence: 99%

“…26 GPR was used by Hoang et al 27 to predict the compressive strength of concrete samples based on composition and process variables. Yabansu et al 18 used GPR models to predict the microstructural evolution of Ni-based superalloys in a thermal aging process. Tallman et al 28 used a GPR-based approach to drive data collection for a homogenization model in order to study deformation of a-Ti.…”

Section: Introductionmentioning

confidence: 99%

Autonomous Development of a Machine-Learning Model for the Plastic Response of Two-Phase Composites from Micromechanical Finite Element Models

Marshall

Kalidindi

2021

JOM

Self Cite

View full text Add to dashboard Cite

We demonstrate a novel strategy for the autonomous development of a machine-learning model for predicting the equivalent stress-equivalent plastic strain response of a two-phase composite calibrated to micromechanical finite element models. A unique feature of the model is that it takes a user-defined three-dimensional, two-phase microstructure along with user-defined hardening laws for each constituent phase, and outputs the equivalent stressplastic strain response of the microstructure modeled using J 2 -based isotropic plasticity theory for each constituent phase. Previously, this task was addressed using linear regression approaches on a large training dataset. In this work, it is demonstrated that the use of Gaussian process regression together with a Bayesian sequential design of experiments can lead to autonomous protocols for optimal generation of the training dataset and the development of the model. It is shown that this strategy dramatically reduces the time and effort expended in generating the training set.

show abstract

“…In this paper, we demonstrate novel workflows that extend significantly the previously demonstrated assays in multiple research directions: (i) the prototyping of a much larger library of AM Ti–Mn alloys employing intentionally induced compositional gradients coupled with different post-build heat treatments, and (ii) the use of data-driven model-building strategies such as Gaussian process regression (GPR) [ 40 , 41 , 42 , 43 , 44 , 45 , 46 ] for extracting practically useful correlations from experimental datasets. GPR offers many potential advantages compared to other machine learning approaches, including the ability to utilize smaller data sets (i.e., smaller numbers of data points) [ 42 , 44 ], rigorous treatment of uncertainty [ 47 , 48 ] and dynamic selection of new experiments that maximize the expected information gain [ 49 , 50 , 51 ]. This work explores and demonstrates a framework for high-throughput experimental assays to facilitate the efficient exploration of the AM process space as well as statistical analyses of the accumulated data using GPR approaches.…”

Section: Introductionmentioning

confidence: 99%

Evaluation of Ti–Mn Alloys for Additive Manufacturing Using High-Throughput Experimental Assays and Gaussian Process Regression

et al. 2020

Self Cite

View full text Add to dashboard Cite

Compositionally graded cylinders of Ti–Mn alloys were produced using the Laser Engineered Net Shaping (LENS™) technique, with Mn content varying from 0 to 12 wt.% along the cylinder axis. The cylinders were subjected to different post-build heat treatments to produce a large sample library of a–b microstructures. The microstructures in the sample library were studied using back-scattered electron (BSE) imaging in a scanning electron microscope (SEM), and their mechanical properties were evaluated using spherical indentation stress–strain protocols. These protocols revealed that the microstructures exhibited features with averaged chord lengths in the range of 0.17–1.78 mm, and beta content in the range of 20–83 vol.%. The estimated values of the Young’s moduli and tensile yield strengths from spherical indentation were found to vary in the ranges of 97–130 GPa and 828–1864 MPa, respectively. The combined use of the LENS technique along with the spherical indentation protocols was found to facilitate the rapid exploration of material and process spaces. Analyses of the correlations between the process conditions, several key microstructural features, and the measured material properties were performed via Gaussian process regression (GPR). These data-driven statistical models provided valuable insights into the underlying correlations between these variables.

show abstract

Application of Gaussian process regression models for capturing the evolution of microstructure statistics in aging of nickel-based superalloys

Cited by 71 publications

References 88 publications

Locally Optimizable Joint Embedding Framework to Design Nitrogen‐rich Molecules that are Similar but Improved

Locally Optimizable Joint Embedding Framework to Design Nitrogen‐rich Molecules that are Similar but Improved

Autonomous Development of a Machine-Learning Model for the Plastic Response of Two-Phase Composites from Micromechanical Finite Element Models

Evaluation of Ti–Mn Alloys for Additive Manufacturing Using High-Throughput Experimental Assays and Gaussian Process Regression

Contact Info

Product

Resources

About