This work aims to compare the effectiveness of different machine-learning techniques for the image classification of steel microstructures. For this, we use a set of samples of hypoeutectoid steels subjected to three heat treatments: annealing, quenching and quenching with tempering. Logically, the samples contain the typical constituents expected, and these are different for each treatment. Images are obtained by optical microscopy at 400× magnification and from different low-carbon steels to generate the data with some heterogeneity. Learning models are created with an image dataset for classification into three classes based on the respective heat treatments. Likewise, we develop two kinds of models by using, on the one hand, classical machine-learning methods based on the “bag of features” technique and, on the other hand, convolutional neural networks (CNN) with a transfer-learning approach by using GoogLeNet and ResNet50. We demonstrate the superiority of deep-learning techniques (CNN) over classical machine-learning methods.

The utilization of convolutional neural networks (CNNs) for semantic segmentation has proven to be successful in various applications, such as autonomous vehicle environment analysis, medical imaging, and satellite imagery. In this study, we investigate the application of different segmentation networks, including Deeplabv3+, U-Net, and SegNet, each recognized for their effectiveness in semantic segmentation tasks. Additionally, in the case of Deeplabv3+, we leverage the use of pre-trained ResNet50, ResNet18 and MobileNetv2 as feature extractors for a comprehensive analysis of steel microstructures. Our specific focus is on distinguishing perlite and ferrite phases in micrographs of low-carbon steel specimens subjected to annealing heat treatment. The micrographs obtained using an optical microscope are manually segmented. Preprocessing techniques are then applied to create a dataset for building a supervised learning model. In the results section, we discuss in detail the performance of the obtained models and the metrics used. The models achieve a remarkable 95% to 98% accuracy in correctly labeling pixels for each phase. This underscores the effectiveness of our approach in differentiating perlite and ferrite phases within steel microstructures.

Characterizing the microstructures of steel subjected to heat treatments is crucial in the metallurgical industry for understanding and controlling their mechanical properties. In this study, we present a novel approach for generating images of steel microstructures that mimic those obtained with optical microscopy, using the deep learning technique of generative adversarial networks (GAN). The experiments were conducted using different hyperparameter configurations, evaluating the effect of these variations on the quality and fidelity of the generated images. The obtained results show that the images generated by artificial intelligence achieved a resolution of 512 × 512 pixels and closely resemble real microstructures observed through conventional microscopy techniques. A precise visual representation of the main microconstituents, such as pearlite and ferrite in annealed steels, was achieved. However, the performance of GANs in generating images of quenched steels with martensitic microstructures was less satisfactory, with the synthetic images not fully replicating the complex, needle-like features characteristic of martensite. This approach offers a promising tool for generating steel microstructure images, facilitating the visualization and analysis of metallurgical samples with high fidelity and efficiency.

The aim of this investigation is to analyze the performance of several supervised machine learning algorithms for solving the automatic classification problem of steel image microstructures. We conducted an experiment using a public-domain dataset of Ultra High Carbon Steel Micrographs (UHCSM). This image database consists of a collection of scanning electron micrographs (SEM) taken from samples of a commercial roll-mill casting with a nominal carbon of 2%. Heat treatments such as annealing, water quenching, air and furnace cooling were performed on steel samples so primary microconstituents could be found in micrographs. Each of these microconstituents defines each of the categories of classification to be accomplished by machine learning algorithms. The heat treatments brought about 4 usable classes (sets of images) of primary microconstituents: pearlite, spheroidite, proeutectoid cementite network, pearlite containing spheroidite. All labeled images are prepared to improve models' accuracy in a preprocessing stage so that the image dataset is ready for feature extraction. In order to develop classification models, we put to the test distinct machine learning approaches by working with Matlab's classification learner application where we perform automated training to search for the best classification model type, including Decision Trees, Support Vector Machines (SVM), Discriminant Analysis, Nearest Neighbors, Naive Bayes, Ensemble k-NN, and Neural Network classification. For obtaining the features of the images (feature extraction) we choose the method of Bag-of-features with 400 words for the first experiment, and 327 words by removing less important features for a second experiment. The experimented models reached very different accuracy values on training, with SVM as the best classifier which gets 91.6% accuracy. We can conclude that classic machine learning algorithms solve the classification, but an accuracy improvement can be reached by investigating deep learning techniques.