Ongoing Projects

Fusion of data-driven approach and physical metallurgy

The goal of material design is to achieve inverse material design, the aim of which is to discover novel materials that have certain desired properties. The establishment of a process-structure-property linkage is indispensable for developing a general methodology for inverse material design and understanding the physical mechanisms behind material microstructure generation. For that purpose, we develop the general methodology for the characterization and analysis of random heterogeneous materials by applying the deep generative model. In the submitted paper, it is shown that our approach can be a basis for establishing a stochastic process-structure-property linkage. Recently, we have been working on interpreting abstract knowledge captured by the deep-learning framework from a physical or material perspective to extract explainable material knowledge.

  1. S. Noguchi and J. Inoue, “Stochastic characterization and reconstruction of material microstructures for establishment of process-structure-property linkage using the deep generative model”, Physical Review E, 104 (2021) 025302.
  2. S. Noguchi, H. Wang, J. Inoue, "Identification of microstructures critically affecting material properties using machine learning framework based on metallurgists' thinking process", Scientific Reports,

Recrystallization and evolution of a texture is the complex phenomenon affected by various kinds of factors, and the problem which is the main dominating factor of recrystallization has not been resolved. In this study, application of Data-driven method for recrystallization model was done and the dislocation density contained in the subgrain was estimated quantitatively. For the model of recrystallization mechanism, a multi phase-field model was adopted and the recrystallization has been assumed to be a subgrain growth affected by the grain boundary energy and mobility depending on misorientation, and driving force caused by dislocation density. As observation data, microstructure obtained by Electron Back Scattering Diffraction analysis was used. By repeating procedure that is consisted of annealing and analysis several times, the recrystallization process in the same view point was observed. Then the dislocation density was estimated by assimilating simulation data with observation data. By comparing estimated value with experimental value the validity of model can be discussed quantitatively.

Depending on alloy composition and process control, the microstructure of steels may consist of a range of different phases. If the microstructure consists of more than one phase, the properties of the material strongly depend on the type and distribution of the respective phases. Therefore, it is crucial to determine the type and amount of the different phases in order to assess the underlying structure-property relationship. In this study, an efficient deep learning method is presented for distinguishing microstructures of a low- carbon steel. There have been numerous endeavors to reproduce the human capability of perceptually classifying different textures using machine learning methods, but this is still very challenging owing to the need for a vast labeled image dataset. Therefore, we introduce an unsupervised machine learning technique based on convolutional neural networks and a superpixel algorithm for the segmentation of a low-carbon steel microstructure without the need for labeled images. The effectiveness of the method is demonstrated with optical microscopy images of steel microstructures having different patterns taken at different resolutions. In addition, several evaluation criteria for unsupervised segmentation results are investigated along with the hyperparameter optimization.

  1. H. Kim, J. Inoue, T. Kasuya. "Unsupervised microstructure segmentation by mimicking metallurgists' approach to pattern recognition." Scientific Reports ,
  2. H. Kim, Y. Arisato, J. Inoue, "Unsupervised segmentation of microstructural images of steel using data mining methods.", Computational Materials Science 210 (2022) 110855.

It is well-known that the key of almost everything in materials science relates to process-structure-property-performance (PSPP) relationships, which are far from being well-understood. When establishing PSPP relationships of a material using data driven techniques, two modeling routes can be adopted; forward modeling for prediction and inverse modeling for materials discovery. In general, the former is usually not difficult because its relationship is many to one. On the other hand, the latter whose relationship is one to many requires much more effort for at least two reasons: (i) the model parameters may have different values or not consistent with the data, and (ii) discovering the model parameters usually requires the exploration of a huge parameter space. In this study, we aim to solve inverse problems in materials research especially in case of various properties of metallic materials by applying sparse modeing.

There have been various numerical models to explain phase transformation of steels suggested by researchers. The transformation process can be identified by fitting model parameters to experimentally obtained dilatometric curves. However, several issues still remain to be resolved. The first issue is related to the uncertainty in the kinetic models. It is widely accepted that models are imperfect since real phenomena are composed of complex physics taking place across different lengths and time scales. Even if the correct values of the model inputs are used in a simulation, the output of the model will differ owing to the incomplete understanding of the model parameters and insufficient knowledge of the underlying physics. The second issue is related to the uncertainty arising from the choice of a particular model class. In this study, a Bayesian approach is presented for clarifying the best kinetic model explaining the transformation kinetics of a low-carbon steel under different continuous cooling conditions only form dilatometric curves. To estimate kinetic parameters as well as the model plausibility of candidate kinetic models, the exchange Markov chain Monte Carlo method was used. The effectiveness of the proposed method was demonstrated by metallographic investigations of the ferrite formation in a Fe-0.15C-1.5Mn alloy. It is shown that the method is successfully applied for clarifying ferrite transformation kinetics, such as transformation start temperatures, formation mechanisms, and fractions of microstructures.

Deformation and phase transformation kinetics of structural materials

When steel austenite is continuously cooled from Ae3 temperature, plate-like or lath-like ferrite products usually form, which play important roles in determining the mechanical properties of low-carbon steels. For instance, coarse-grained WF or BF may form fracture initiation sites, reducing the fracture toughness of steels. In order to control the microstructures and mechanical properties of low-carbon steels, it is important to reveal the transformation mechanism of these microstructures. Examples of our in-situ observations of the transformation behavior can be found here.

Lath martensite has long been used as the primary phase for tool materials due to its exceptional combination of hardness and toughness, while recently it has been increasingly applied as one of the primary phases to enhance the strength of advanced high-strength steel (AHSS). This new interest in lath martensite has led to the need for better understanding of its plastic deformation behaviors, and a number of papers have reported its plastic deformation under different loading conditions. In our lab, various deformation behaviors of structural materials, not only steel but also Mg and Al alloys under different loading conditions has been analyzed through in-situ observation. Examples of such observation can be found here.

Previous Projects

Development of multilayered metallic composite

Multilayered steels have been developed to provide a novel route to achieving higher-performance steels by employing a high-strength steel and a high-ductility steel independently in the layer structure. It has been clarified that multilayered steels exhibit exceptional combinations of strength and ductility compared with existing monolithic steels and also excellent deformation behaviors under high-strain-rate deformation as well as good formability. Those improved performances of multilayered steel are achieved by increased interfacial toughness between the layers and decreased thickness of the brittle steel layers according to the fracture toughness of the brittle steel and the strength of the ductile steel. 

The concept of multilayered steels has been extended to Mg/steel multilayered composite. For the development of the Mg/steel multilayer, new bonding technique between Mg and steel was first developed, which can be found in this page. With the help of the new bonding technique, the excellent mechanical properties of Mg/steel multilayered was demonstrated. The improved strength of the multilayer samples was clarified due to the combined effect of surface crack prevention by the steel layer and the higher work-hardening rate caused by the possible increased activity of non-basal systems.

Stress partition measurement using high energy beams

Stress partitioning in multilayered steels consisting of martensitic and austenitic layers was measured during tensile deformation by in situ neutron diffraction measurements to investigate the mechanism of the improved strength–elongation balance. The deformation mode can be classified into three stages, and the results indicate that the applied stress is effectively transferred to the martensitic phase, because no stress concentration sites exist, owing to the multilayered structure. Hence, even as-quenched martensite can deform uniformly, resulting in improved strength–elongation balance in multilayered steels.

Mechanical properties of thin films

New technique to evaluate the fracture toughness of a brittle thin film on a elastoplastic substrate was developed. In this technique, the fracture toughness can be evaluated simply from the thickness of the brittle layer and the crack interval in the layer after applying uniaxial tensile deformation. The technique was applied to various kinds of Intermetallic compound layer developed on steel substrates to reveal the fracture behavior of brittle thin films.

Formation mechanism of bonding interface between dissimilar metals