The TM14RE2B-based phases (TM = transition metal, RE = rare earth metal; hereafter called 14:2:1) enable permanent magnets with outstanding magnetic properties. Novel chemical compositions that represent new 14:2:1 phases necessitate that they do not demagnetize at application-specific operating temperatures. Therefore, an accurate knowledge of the Curie temperature (Tc) is important. For magnetic 14:2:1 phases, we present a machine learning model that predicts Tc by using merely chemical compositional features. Hyperparameter tuning on bagging and boosting models, as well as averaging predictions from individual models using the voting regressor, enables a low mean-absolute-error of 16 K on an unseen test set. The training set and a test set have been constructed by randomly splitting, in an 80:20 ratio, of a database that contains 449 phases (270 compositionally unique) mapped with their Tc, taken from distinct publications. The model correctly identifies the relative importance of key substitutional elements that influence Tc, especially in an Fe base such as Co, Mn, and Al. This paper is expected to serve as a basis for accurate Curie temperature predictions in the sought-after 14:2:1 permanent magnet family, particularly for transition metal substitution of within 20% in an Fe or Co base.
Materials microstructure analysis is a crucial task for understanding the relationship between a material’s structure and its properties. The quality of the acquired images is vital, as blurring can result in the loss of essential information needed for microstructure analysis. This issue becomes more apparent with higher magnification and the corresponding decrease in depth of field (D.O.F.). To address this, new AI-based approaches are being explored to automatically improve image quality. Experts are testing innovative loss functions for training AI models to improve the image deblurring performance and thus a better image reconstruction. In the research field of image deblurring, it is known that pixel-based loss functions badly correspond with human perception of image quality. Therefore, a perceptual loss term is used for better visual reconstruction results. However, it has been shown that a perceptual loss term can lead to visible artifacts. Current literature has not yet sufficiently investigated why these artifacts are generated. In this work, we tackle this lack of research and provide new insights into the generation of such artifacts. We not only show that artifact generation is caused by the feature map used for the perceptual loss, but we also show that the level of distortion in the input impacts the extent of the artifact generation. Furthermore, we show that common metrics are insensitive to these artifacts, and we propose a method to reduce the extent of the generated artifacts.
Deep learning-based image super resolution (SR) is an image processing technique designed to enhance the resolution of digital images. With the continuous improvement of methods and the growing availability of large real-world datasets, this technology has gained significant importance in a wide variety of research fields in recent years. In this paper, we present a comprehensive review of promising developments in deep learning-based image super resolution. First, we give an overview of contributions outside the field of microscopy before focusing on the specific application areas of light optical microscopy, fluorescence microscopy and scanning electron microscopy. Using selected examples, we demonstrate how the application of deep learning-based image super resolution techniques has resulted in substantial improvements to specific use cases. Additionally, we provide a structured analysis of the architectures used, evaluation metrics, error functions, and more. Finally, we discuss current trends, existing challenges, and offer guidance for selecting suitable methods.
14:2:1 phases enable permanent magnets with excellent magnetic properties. From an application viewpoint, saturation polarization, Curie temperature, and anisotropy constant are important parameters for the magnetic 14:2:1 phases. Novel chemical compositions that represent new 14:2:1 phases require especially maximum saturation magnetization values at application-specific operating temperatures to provide maximum values for the remanence and the maximum energy density in permanent magnets. Therefore, accurate knowledge of the saturation magnetization Ms is important. Ms gets affected by chemical composition in a twofold way, with chemical composition significantly influencing both magnetic moments and crystal structure parameters. Therefore, for magnetic 14:2:1 phases, we have developed a regression model with the aim to predict the saturation magnetization in [µB/f.u.] at room temperature directly from the chemical composition as input features. The dataset for the training and testing of the model is very diverse, with literature data of 143 unique phases and 55 entries of repeated phases belonging to the ternary, quaternary, quinary, and senary alloy systems. Substitutionally dissolved elements are heavy and light rare earth elements, transition metals, and additional elements. The trained model is a voting regressor model with different weights assigned to four base regressors and has generalized well, resulting in a low mean absolute error of 0.8 [µB/f.u.] on the unseen test set of 52 phases. This paper could serve as the basis for developing novel magnetic 14:2:1 phases from chemical composition.
Training of semantic segmentation models for material analysis requires micrographs as the inputs and their corresponding masks. In this scenario, it is quite unlikely that perfect masks will be drawn, especially at the edges of objects, and sometimes the amount of data that can be obtained is small, since only a few samples are available. These aspects make it very problematic to train a robust model. Therefore, we demonstrate in this work an easy-to-apply workflow for the improvement of semantic segmentation models of micrographs through the generation of synthetic microstructural images in conjunction with masks. The workflow only requires joining a few micrographs with their respective masks to create the input for a Vector Quantised-Variational AutoEncoder (VQ-VAE) model that includes an embedding space, which is trained such that a generative model (PixelCNN) learns the distribution of each input, transformed into discrete codes, and can be used to sample new codes. The latter will eventually be decoded by VQ-VAE to generate images alongside corresponding masks for semantic segmentation. To evaluate the quality of the generated synthetic data, we have trained U-Net models with different amounts of these synthetic data in conjunction with real data. These models were then evaluated using real microscopic images only. Additionally, we introduce a customized metric derived from the mean Intersection over Union (mIoU) that excludes the classes that are not part of the ground-truth mask when calculating the mIoU of all the classes. The proposed metric prevents a few falsely predicted pixels from greatly reducing the value of the mIoU. With the implemented workflow, we were able to achieve a time reduction in sample preparation and acquisition, as well as in image processing and labeling tasks. The idea behind this work is that the approach could be generalized to various types of image data such that it serves as a user-friendly solution for training models with a smaller number of real images.
FastCAR is a novel task consolidation approach in Multi-Task Learning (MTL) for a classification and a regression task, despite the non-triviality of task heterogeneity with only a subtle correlation. The approach addresses the classification of a detected object (occupying the entire image frame) and regression for modeling a continuous property variable (for instances of an object class), a crucial use case in science and engineering. FastCAR involves a label transformation approach that is amenable for use with only a single-task regression network architecture. FastCAR outperforms traditional MTL model families, parametrized in the landscape of architecture and loss weighting schemes, when learning both tasks are collectively considered (classification accuracy of 99.54%, regression mean absolute percentage error of 2.4%). The experiments performed used "Advanced Steel Property Dataset" contributed by us https://github.com/fastcandr/AdvancedSteel-Property-Dataset. The dataset comprises 4536 images of 224x224 pixels, annotated with discrete object classes and its hardness property that can take continuous values. Our proposed FastCAR approach for task consolidation achieves training time efficiency (2.52x quicker) and reduced inference latency (55% faster) than benchmark MTL networks.