Identification of Electromagnetic Steel Sheets for Motors by Material Structure Characteristics

[Introduction] As part of a growing effort toward a sustainable, recycling-oriented society, more countries have been actively improving energy efficiency and accelerating decarbonization. Consequently, the electrification of vehicles and autonomous driving have been progressing as well. At the same time, due to the globalization and diversification of suppliers in recent years, the visibility of the supply chain is becoming more easily lowered. For these reasons, quality requirements for drive-motor parts are becoming stricter, so a technique is needed for enhancing the reliability of the parts. In particular, suppressing silent changes, which replaces parts with low-quality ones that differ from the ordering specifications of product manufacturers in the supply chain, also contributes to the construction of a safe and secure society through quality control. Thus, we developed a method that combines machine learning with simple measurement methods for high-precision identification of electromagnetic steel sheets for motors (non-directional electromagnetic steel sheets).<br/>[Methods] Magnetic properties (iron loss), mechanical strength, and degradation resistance are performance indices (objective variables) for electromagnetic steel sheets. These descriptors can be used to identify materials, but the differences between the materials may not be evident due to their macro characteristics. However, because characteristics derived from material structures differ depending on the material, they can be useful descriptors for identifying the materials when they can be appropriately selected. However, it is difficult to use most analytical methods to select descriptors for identifying the materials among the many extracted characteristics. Therefore, by using material knowledge, we selected analysis methods that are expected to be highly correlated with the objective variables and determined the optimal descriptors using a data drive approach. The electromagnetic steel sheets are extremely advanced functional materials, where the size and orientation of crystal grains are controlled by introducing inhibitors. Stress is controlled by covering the materials with an insulating film for degradation resistance, and electrical resistance and deterioration due to magnetic aftereffects are controlled by adjusting the composition of the materials. Based on these control mechanisms, XRD, ICP, optical microscope, and EBSD were selected as analytical methods. Subsequently, a total of 48 characteristics were extracted as explanatory variables (i.e., diffraction peak characteristics from the XRD, components from the ICP, average particle size and dispersion from the optical microscope, and orientation degree from EBSD).<br/>[Results] We found 36 explanatory variables that highly correlated with the objective variables. In particular, the characteristics from XRD correlated with all of the objective variables except for the high frequency component of vortex loss in iron loss. On the other hand, only the amount of Mn from ICP showed a high correlation with the high frequency component of vortex loss. Thus, we concluded that the combination of XRD measurement and Mn measurement from ICP can substitute the evaluation of magnetic and mechanical properties. This alternative can reduce analytical costs by 70%. When the random forest method using only the results of XRD was used to identify 16 materials, an accuracy of 94% was attained. The results of cluster analyses will also be presented at the meeting.