无监督学习在相图测定中识别合金三相平衡点的研究

Research on Unsupervised Learning for Identification of Three-Phase Equilibrium Point of Alloy in Phase Diagram Determination

传统的扫描电镜分析方法,合金物相的三相平衡点均为手工标记。针对扫描电镜金相分析中常碰到的物相分析问题,提出了一种物相三相平衡点的自动识别方法,实现平衡点的智能化标注,提升实验效率。该方法分为三相分割和三相平衡点识别两个阶段。在分割阶段提出一种无监督分割方法完成三种物相分割,在平衡点识别阶段设计识别算法对分割结果的平衡点区域进行识别。实验结果表明,分割方法在三元合金相图数据集上的平均交并比(Mean Intersection over Union,MIo U)达到49.86%,三相平衡点平均检出率达62%。

In traditional SEM(Scanning Electron Microscopy)analysis method,the three-phase equilibrium points of alloy phases are manually marked.In order to address common phase analysis problems in scanning electron microscopy metallographic analysis,this paper proposes a method for automatic identification of three-phase equilibrium points to achieve intelligent mark of equilibrium points and improve experimental efficiency.This method is divided into two stages:three-phase segmentation and three-phase equilibrium point recognition.In the segmentation stage,an unsupervised segmentation method is proposed to complete three-phase segmentation.In the equilibrium point recognition stage,a recognition algorithm is designed to identify the equilibrium point area of the segmentation results.The experimental results show that the average Mean Intersection over Union(MIoU)of the segmentation method on the ternary alloy phase diagram dataset reaches 49.86%,and the average detection rate of the three-phase equilibrium point reaches 62%.