Non-grain oriented electrical steel photomicrograph classification using transfer learning

Abstract

Among the many factors that contribute to achieving a sustainable and efficient economy is the raw material of machinery and equipment of strategic sectors. Non-grain oriented (NGO) electrical steel is used in the manufacturing of electric motors. Therefore, it is directly related to the electromagnetic efficiency of engines both in industry and in homes. This work aims to develop an intelligent 1.26% Si NGO electrical steel photomicrograph classification system to assist in the identification of better energy-efficient steel. The concept of Transfer Learning was used to apply Convolutional Neural Network architectures as feature extractors. Traditional machine learning classifiers are applied for coherent categorization of material efficiency. From the results, it is noted that the combination of the InceptionV3 architecture with the k-nearest neighbors classifier reached 100% accuracy and F1-Score. The average extraction time and test time were approximately 15 and 0.920μs, respectively. Given these results, the literature on this application is surpassed. The best extractor-classifier combination is available in an Internet of Things (IoT) system. Therefore, a professional can freely make use of the proposed approach to assist them in identifying low magnetic loss electrical steel.