AI 中文总结
GRADAR针对ECCI需逐个晶粒调整双束条件的问题，提出取向图驱动的晶粒检测排序方法，可精准预测可触及双束条件的晶粒，排序精度高且验证结果良好。
AI 中文摘要
电子通道衬度成像（ECCI）可解析块状样品中的单个位错，但前提是成像晶粒需处于双束衍射条件下，且角度偏差小于半度。在多晶样品中，需逐个晶粒重新建立该条件，目前所有已发表的工作流程均先选择晶粒，再针对该晶粒调整样品台。本文研究逆问题：指定一个反射族和样品台包络，确定EBSD映射的多晶中哪些晶粒可达到该反射族的双束条件，该问题自2019年提出后一直未解决。在固定倾斜角t时，样品台旋转会使入射束在各晶粒的晶体坐标系中扫过半顶角为t的圆锥，当某晶面法线与样品表面的夹角在(t + θ_B)范围内时，该晶面族（布拉格角为θ_B）可被精确触及。在20 kV下测得的163个晶粒的奥氏体不锈钢映射图中，{111}晶面族的可触及率从7°倾斜时的38.7%升至15°倾斜时的77.3%；{220}和{311}晶面族在7°倾斜时已分别达到79.8%和91.4%。仅按预测暗度对候选晶粒排序是错误的：同一家族的候选晶粒在精确布拉格入射时对自身衍射带的贡献等效，且最暗的晶粒往往最接近竞争带的布拉格条件。因此GRADAR采用“暗度-洁净度”排序，对每个候选晶粒设置与所有竞争带的最小间隙阈值，最终返回每个选定晶粒的可直接操作的样品台移动参数。由于在7°倾斜时，样品台每旋转1度，光束仅移动约0.12度，普通旋转台可将选定条件设置至优于0.1度的精度。将预测扫描与AstroECP数据集已发表的硅进动系列进行验证：模拟与实测轨迹的相关系数r=0.84，且在16个预测的双束边缘处，实测强度均为暗态。单晶粒多晶验证将推迟至配套研究中开展。
英文摘要
Electron channeling contrast imaging (ECCI) resolves individual dislocations in bulk samples, but only when the imaged grain is held in a two-beam diffraction condition to well under half a degree. In a polycrystal it must be re-established grain by grain, and every published workflow we are aware of chooses the grain first and solves the stage for that orientation. This paper treats the inverse problem: nominate one reflection family and a stage envelope, and determine which grains of an EBSD-mapped polycrystal can reach that family's two-beam condition, a question posed in 2019 and open since. At fixed tilt $t$, stage rotation sweeps the incident beam around a cone of half-angle $t$ in each grain's crystal frame, and a family with Bragg angle $θ_B$ is reachable exactly when one of its plane normals lies within $(t + θ_B)$ of the specimen surface. On a measured 163-grain austenitic stainless steel map at 20 kV, {111} reachability rises from 38.7% at 7 deg tilt to 77.3% at 15 deg; {220} and {311} already reach 79.8% and 91.4% at 7 deg. Ranking survivors by predicted darkness alone is wrong: same-family candidates are equivalent with respect to their own band at exact Bragg incidence, and the darkest tend to lie closest to rival-band Bragg conditions. GRADAR therefore ranks darkest-clean, gating every candidate on a minimum clearance from all rivals, and returns a dial-ready stage move per selected grain. Because the beam moves only ~0.12 deg per degree of stage rotation at 7 deg tilt, an ordinary rotation stage sets the selected condition to better than 0.1 deg. The predicted sweep is checked against the AstroECP dataset's published silicon precession series: simulated and measured traces agree at r = 0.84, and the measured intensity is dark at all 16 predicted two-beam edges. Per-grain polycrystal validation is deferred to a companion study.
Comments17 pages, 8 figures
