We attempted to improve the strength–ductility balance in tempered martensite steel using multi-objective Bayesian optimization. The martensite steels were tempered at two stages, and fine and coarse cementite particles were mixed. Water-quenching rather than furnace-cooling between the first and second temper stages provided a better balance of strength and ductility. Moreover, the strength–ductility balance was also improved by tempering at a low temperature in the first stage and a high temperature in the second stage, rather than tempering at a high temperature in the first stage and a low temperature in the second stage. Based on these experimental results, multi-objective Bayesian optimization was used to further improve both tensile strength and total elongation. The strength–ductility balance that is better than the experimental results was achieved with a minimal number of optimization times. Additionally, machine learning suggested that it is crucial to control the average aspect ratio of cementite particles to less than 1.8 and the size of coarse cementite particles to less than 2.0 μm in order to improve the strength–ductility balance.

Creep deformation and precipitation behavior of 9Cr–1Mo–V–Nb steel with excess nitrogen introduced by solution nitriding were investigated. Precipitation of Cr₂N phase was confirmed in addition to M₂₃C₆ and MX phases in the tempered microstructure. The creep strength of the steel was significantly reduced by solution nitriding, while the creep rupture elongation was increased. To characterize the complex precipitation behavior of the nitrogen-added steel, a machine learning-based clustering method of the multidimensional scatter diagram of the X-ray intensity of the alloying elements in each pixel of a STEM-EDS map was developed. Reduced number density of precipitates and enhanced coarsening kinetics of both Cr₂N and MX were proposed as the mechanism of weakening caused by excess nitrogen.

The solid–liquid interface energy anisotropy of Zn alloys remains poorly understood. Recently, characteristic 14-arm dendritic growth has been observed using time-resolved X-ray computed tomography at SPring-8 during the solidification of a Zn–4mass%Al alloy. This study investigates the dependence of the dendrite morphologies of Zn alloys on solid–liquid interface energy anisotropy through systematic phase-field simulations of the growth of an isolated equiaxed dendrite. We also develop a data assimilation system to estimate the anisotropy parameters of solid–liquid interface energy and crystal orientation in Zn alloys and validate the system through twin experiments. This study provides insights into the solidification of Zn alloys and a powerful tool for their investigation.

Creep testing is time-consuming and costly, leading institutions to limit the number of tests conducted to the minimum necessary for their specific objectives. By pooling data from each institution, it is anticipated that predictive models can be developed for a wide range of materials, including welded joints and degraded materials exposed to service conditions. However, the data obtained by each institution is often highly confidential, making it challenging to share with others. Federated learning, a type of privacy-preserving computation technology, allows for learning while keeping data confidential. Utilizing this approach, it is possible to develop creep life prediction models by leveraging data from various institutions. In this paper, we constructed global deep neural network models for predicting the creep rupture life of heat-resistant ferritic steels in collaboration with eight institutions using the federated learning system we developed for this purpose. Each institution built a local model using only its own data for comparison. While these local models demonstrated good predictive accuracy for their respective datasets, their predictive performance declined when applied to data from other institutions. In contrast, the global model constructed using federated learning showed reasonably good predictive performance across all institutions. The distance between each institution's data was defined in the space of explanatory variables, with the NIMS data, which had the largest dataset, serving as the reference point. The global model maintained high predictive accuracy regardless of the distance from the NIMS data, whereas the predictive accuracy of the NIMS local model significantly decreased as the distance increased.

Using two types of image processing techniques without machine learning, edge extraction processing and keypoint extraction processing, progressively corroded regions under the rust layer from images of corroded steel surfaces was extracted. We found that there was a relatively good correlation between the keypoint strength obtained from the keypoint extraction processing for HSL transformed and histogram flattened corroded surface images and the corrosion depth after rust removal.

Corrosion products formed on steel in various environments were identified with image analysis, Raman spectroscopy, and X-ray diffraction (XRD). The results of Raman spectroscopy showed that γ-FeOOH exhibits bright yellow color and β-FeOOH exhibits gray color. Main corrosion products composition formed on the steels by XRD analysis was γ-FeOOH. Using proper threshold of HSV color space, makes it possible to identify corrosion products formed on steel. Comparison of corrosiveness of exposed environments and formed corrosion products, α-FeOOH is preferentially formed in mild corrosive environments, while β-FeOOH was formed in severe corrosive environments.

The utility of hyperspectral measurement was assessed as a means of predicting the corrosion risk of steel materials based on surface information. Carbon steels exposed to outdoor conditions in Choshi and Miyakojima were used as the test specimens. Exposure tests were conducted every six months for a duration of two and a half years. Corrosion loss was calculated by comparing the weight of specimens before exposure test and after removing corrosion products from the surface. Hyperspectral measurements were conducted on these specimens, with corrosion products identified through SAM (Spectral Angle Mapper) analysis. α-FeOOH, β-FeOOH, γ-FeOOH, and Fe₃O₄ were employed as reference data for SAM analysis. In both Choshi and Miyakojima test sites, γ-FeOOH was predominantly detected on the specimens after ordinary exposure tests, whereas Fe₃O₄ was prevalent on the specimens exposed to sheltered environments. The correlation between the proportion of each corrosion product identified through SAM analysis and the amount of corrosion change for one year was explored. α-FeOOH exhibited a positive correlation with the amount of corrosion change, whereas the amount of corrosion change tended to decrease with an increase in β-FeOOH fraction.

Analysis of the corrosion distribution and composition of corrosion products on steel surfaces using supervised machine learning of optical microscopic images was investigated. The accuracy of the artificial intelligence in evaluating the composition of iron compound reference samples was affected by the illumination intensity and surface roughness during image capture. The evaluation accuracy was high for compounds with a wide distribution of R value such as Fe₂O₃ and FeOOH, but low for compounds with a narrow distribution such as Fe₃O₄. The results of wet-dry cycling tests on weathering steel with NaCl particles on the surface showed that the transition of corrosion products during the corrosion progress can be analyzed from optical microscope images.

In material design, the establishment of process–structure–property relationship is crucial for analyzing and controlling material microstructures. For the establishment of process–structure–property relationship, a central problem is the analysis, characterization, and control of microstructures, since microstructures are highly sensitive to material processing and critically affect material’s properties. Therefore, accurately estimating the morphology of material microstructures plays a significant role in understanding the process–structure–property relationship. In this paper, we propose a deep-learning framework for estimating material microstructures under specific process conditions. The framework utilizes two deep learning networks: Vector Quantized Variational Autoencoder (VQVAE) and Pixel Convolutional Neural Network (PixelCNN). The framework can predict material micrographs from the transformation behavior given by some physical model. In this sense, the framework is consistent with the physical knowledge accumulated in the field of material science. Importantly our study demonstrates qualitative and quantitative evidences that incorporating physical models enhances the accuracy of microstructure estimation by deep learning models. These results highlight the importance of appropriately integrating field-specific knowledge when applying data-driven frameworks to materials design. Consequently, our results provide a foundation for integrating data-driven methods with the accumulated knowledge in the field. This integration holds great potential for advancing material design through deep learning.

The distribution of alloying elements was investigated to clarify latent features of the microstructure formation in TRIP-aided Multi-phase steels by using an unsupervised machine learning technique. The cold-rolled specimens with a composition of Fe-0.20mass%C-1.5mass%Si-1.5mass%Mn were heat-treated under various conditions. The FE-EPMA (field emission electron probe microanalysis) method was used to accurately measure the distribution of C, Si and Mn. The two-phase microstructure of ferrite and austenite was formed during intercritical annealing at 800 °C for 120 s. C and Mn were distributed in the austenite region while Si was distributed in the ferrite region. Part of the austenite transformed to bainite by subsequent heat-treatment of isothermal holding at 400 °C for 1200 s. The C concentration in the austenite region increased due to bainite formation. The microstructure formation factor was estimated by primary component analysis of big data on the elemental distribution. The cumulative contribution ratio of the second principal component was 0.85. Therefore, most of the information was contained in the first and second principal components. In the first principal component, the coefficients for the standardized C and Mn concentrations possessed positive values, while that for the standardized Si concentration possessed negative value. The tendency of each coefficient corresponds to the C, Si and Mn distribution caused by intercritical annealing. Moreover, the absolute value of the coefficient for the standardized C concentration was significantly larger than those of the other coefficients in the second principal component. The component represents the C enrichment of austenite due to bainite transformation.

Materials Integration is the concept of accelerating materials development by linking processing, structure, property, and performance (PSPP) on a computer using any types of models such as theoretical, empirical, numerical-simulation, and machine learning models. In the first and second phases of Cross-ministerial Strategic Innovation Promotion Program in Cabinet Office, Japan, we have developed a system called MInt (Materials Integration for Network Technology), which links PSPP with computational workflows that combine modules implemented, in order to realize the concept of Materials Integration. MInt is equipped with an application programming interface (API) that can be called from various algorithms in the artificial intelligence (AI) field and one can use MInt-API together with the AI algorithms to inversely design materials and processes from desired performance. The target material systems have expanded to steel, aluminum alloys, nickel alloys, and titanium alloys, and the target processes have also expanded to welding, heat treatment, 3D additive manufacturing, and powder metallurgy. MInt is more than just software for materials design; it is designed to serve as a digital platform for industry-academia collaboration. The Materials Integration Consortium has been established with MInt as its core technology, based on the philosophy of sharing tools such as modules and workflows, while competing on how to use them. In materials research and development, which has traditionally been regarded as a competitive area, we hope that a digital collaborative area will be formed and that investment efficiency will be drastically improved.

In recent years, remarkable advances have been made in statistical analyses based on deep learning techniques. Applied studies of deep learning have been reported in various industrial fields, and those of the iron and steel industry are no exception. The production of iron and steel requires a variety of processes, such as processing of ingredients, iron making, casting, and rolling. As a result, the data acquired from them are diverse, and various tasks exist for which deep learning algorithms can assist. Hence, providing a summary of the application is helpful not only for researchers specializing in information science to grasp the current trend of applied studies on deep learning techniques, but also for researchers specializing in each field of the iron and steel making industry to understand what types of deep learning techniques are being utilized in other specialized fields. Therefore, in this paper, we summarize current studies on the application of deep learning in the iron and steel making field by organizing them into several categories of processes and analytical methodologies. Furthermore, based on the results, we discuss future perspectives on the development of deep learning techniques in this field.

We have attempted to predict creep rupture time for a wide range of ferritic heat resistant steels with machine learning methods using the NIMS Creep Data Sheet (CDS). The dataset consisted of commercial steel data from 27 sheets in the CDS, covering various grades of carbon steels, low alloy steels, and high Cr steels. The prediction models were constructed using three methods, support vector regression (SVR), random forest, and gradient tree boosting with 5132 training data in order to predict log rupture time from chemical composition (19 elements), test temperature, and stress. Evaluation with 451 test data proved that all three models exhibited high predictivity of creep rupture time; in particular, the performance of the SVR model was the highest with a root mean squared error as low as 0.14 over the log rupture time, which value means that, on average, the prediction error was factor 1.38 (=10^0.14). The high predictivity achieved with no use of information on microstructure was presumably because the data used was for commercial steels in which there should be a correlation between the composition and the microstructure. We confirmed that the prediction did not work well exceptionally for two heats having the same composition but different microstructures with and without stress relief annealing. The predictivity could be drastically improved by adding the 0.2% proof stress at the creep test temperature as one of the explanatory variables. As a use case of the prediction model, the effect of elements was evaluated for modified 9Cr 1Mo steels.

A method to automatically search brittle fracture initiation point is proposed. The method calculates flow paths from each edge of an image to opposite side along flow potential. The potential is derived from the direction of river-pattern and tear ridge. The most concentrated position of flow paths is determined as a fracture initiation point. The method achieves 99% precision and high speed analysis for low magnification images of 206 thick plate samples. Furthermore, discriminating ductile fracture by machine learning and excluding noise potential obtained from it, and considering local feature of river-pattern that spread radially near a fracture initiation point, the method precision is improved for high magnification images. Using fractal feature of brittle fracture and repeatedly applying the method, a fracture initiation point is determined within the size of a single grain level.

In steel making processes, influence of an equipment fault on production operation is significant. It is strongly required to detect an equipment fault at an early stage and to prevent the damage. Therefore, fault detection technique for steel making facilities based on data science is developed as an online-monitoring system. One of main features of the developed system is hierarchical monitoring consisting of three levels such as an entire process, facilities and sensors. Another is display of heat-mapping according to the degree of anomaly for huge number of monitoring items. Some anomaly signs at the hot rolling process where the system has been developed are successfully detected.

Heat transfer analyses of casting processes offer an effective way of understanding solidification processes which is important for prevention of formation of casting defects and for control of segregations. The accuracy of the simulations depends on the accuracy of input parameters such as thermal conductivity and heat transfer coefficient. It is not straightforward to evaluate such parameters, especially thermal conductivities of bulks and heat transfer coefficient with high accuracy. It is very important to develop a reliable method for estimation of these parameters. In this study, particle filter, which is a method of data assimilation, is applied to estimation of thermal conductivity of melt and heat transfer coefficient during alloy solidification in mold casting and its applicability is systematically investigated on the basis of twin experiments. It is shown that a constant heat transfer coefficient can be accurately estimated in this method by utilizing a single cooling curve measured in the mold or in the melt near the mold, while the thermal conductivity can be accurately estimated from a single cooling curve measured in the melt. In addition, this method can be utilized for estimation of time-dependent heat transfer coefficient without any assumption and/or approximation on its time dependence. Importantly, the thermal conductivity and time-dependent heat transfer coefficient can be estimated simultaneously with high accuracy.

Screening important microstructure factors dominantly controlling a stress-strain curve of a dual phase steel was studied by three kinds of sparse modelling approach; Sensitive analysis, data transformation & variable selection, and Bayesian inference. In addition, an effect of data noise on descriptor screening in sparse modelling was also investigated.

Two kinds of advanced image processing were applied to multi-phase microstructures. One is evolutional image processing where optimized filter set was suggested by genetic programing. Another is trainable WEKA segmentation where features are extracted by many kinds of filters, followed by machine learning for classification. Once an optimized filter set is determined, efficiency of image processing for new data set is improved remarkably in comparison with a case of manual image processing.