Microstructure of metastable austenitic manganese steel after reverse transformation treatment was investi gated using optical microscopy, X ray diffraction （XRD）, electrical resistivity and hardness testing. Austen...Microstructure of metastable austenitic manganese steel after reverse transformation treatment was investi gated using optical microscopy, X ray diffraction （XRD）, electrical resistivity and hardness testing. Austenite grain refinement was successfully achieved by a two-step heat treatment. First, martensite was produced by cooling the so- lution-treated samples to --196 ℃. Then, the deep cryogenic treated samples were heated to 850 ℃ upon slow or rapid heating. The mean size of original austenite grain was about 400 fire. But the mean size of equiaxed reversion austenite was refined to 50 μm. Microstructure evolution and electrical resistivity change showed that martensite plates underwent tempering action upon slow heating, and the residual austenite was decomposed, resulting in the formation of pearlite nodules at the austenite grains boundaries. The refinement mechanism upon slow heating is the diffusion-controlled nucleation and growth of austenite. However, the reverse transformation upon rapid heating was predominated by displacive manner. The residual austenite was not decomposed. The plate α-phase was carbon-super- saturated until the starting of reverse transformation. The reverse transformation was accompanied by surface effect, resulting in the formation of plate austenite with high density dislocations. The refinement mechanism upon rapid heating is the recrystallization of displacive reversed austenite.展开更多
High-resolution transmission electron microscope (HRTEM) was employed to investigate the deformation-induced α2→γ phase transformation phenomenon in a hot deformed Ti-45Al-10Nb alloy. Such a tronsformation can be n...High-resolution transmission electron microscope (HRTEM) was employed to investigate the deformation-induced α2→γ phase transformation phenomenon in a hot deformed Ti-45Al-10Nb alloy. Such a tronsformation can be nucleated either at α2/γ interfaces or at stacking faults on the basal planes of the α2 phase. The growth of deformation-induced γplate is accomplished by the motion of α/6<100> Shockley partials on alternate basal planes (0001)α2, and the α/6<100> Shockley partials move in coordination rather than sweep on (0001)α2 plane one by one. It appears that no atom transportation is involved in this stress-induced α2→γ transfromation.展开更多
针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为...针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为网格序列,从而获取建筑多边形化简前后的Token序列,构建出建筑多边形化简样本对数据;随后采用Transformer架构建立模型,基于样本数据利用模型的掩码自注意力机制学习点序列之间的依赖关系,最终逐点生成新的简化多边形,从而实现建筑多边形的化简。在训练过程中,模型使用结构化的样本数据,设计了忽略特定索引的交叉熵损失函数以提升化简质量。试验设计包括主试验与泛化验证两部分。主试验基于洛杉矶1∶2000建筑数据集,分别采用0.2、0.3和0.5 mm 3种网格尺寸对多边形进行编码,实现了目标比例尺为1∶5000与1∶10000的化简。试验结果表明,在0.3 mm的网格尺寸下模型性能最优,验证集上的化简结果与人工标注的一致率超过92.0%,且针对北京部分区域的建筑多边形数据的泛化试验验证了模型的迁移能力;与LSTM模型的对比分析显示,在参数规模相近的条件下,LSTM模型无法形成有效收敛,并生成可用结果。本文证实了Transformer在处理空间几何序列任务中的潜力,且能够有效复用已有化简样本,为智能建筑多边形化简提供了具有工程实用价值的途径。展开更多
基金Sponsored by National Natural Science Foundation of China(51071075)
文摘Microstructure of metastable austenitic manganese steel after reverse transformation treatment was investi gated using optical microscopy, X ray diffraction （XRD）, electrical resistivity and hardness testing. Austenite grain refinement was successfully achieved by a two-step heat treatment. First, martensite was produced by cooling the so- lution-treated samples to --196 ℃. Then, the deep cryogenic treated samples were heated to 850 ℃ upon slow or rapid heating. The mean size of original austenite grain was about 400 fire. But the mean size of equiaxed reversion austenite was refined to 50 μm. Microstructure evolution and electrical resistivity change showed that martensite plates underwent tempering action upon slow heating, and the residual austenite was decomposed, resulting in the formation of pearlite nodules at the austenite grains boundaries. The refinement mechanism upon slow heating is the diffusion-controlled nucleation and growth of austenite. However, the reverse transformation upon rapid heating was predominated by displacive manner. The residual austenite was not decomposed. The plate α-phase was carbon-super- saturated until the starting of reverse transformation. The reverse transformation was accompanied by surface effect, resulting in the formation of plate austenite with high density dislocations. The refinement mechanism upon rapid heating is the recrystallization of displacive reversed austenite.
文摘High-resolution transmission electron microscope (HRTEM) was employed to investigate the deformation-induced α2→γ phase transformation phenomenon in a hot deformed Ti-45Al-10Nb alloy. Such a tronsformation can be nucleated either at α2/γ interfaces or at stacking faults on the basal planes of the α2 phase. The growth of deformation-induced γplate is accomplished by the motion of α/6<100> Shockley partials on alternate basal planes (0001)α2, and the α/6<100> Shockley partials move in coordination rather than sweep on (0001)α2 plane one by one. It appears that no atom transportation is involved in this stress-induced α2→γ transfromation.
文摘针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为网格序列,从而获取建筑多边形化简前后的Token序列,构建出建筑多边形化简样本对数据;随后采用Transformer架构建立模型,基于样本数据利用模型的掩码自注意力机制学习点序列之间的依赖关系,最终逐点生成新的简化多边形,从而实现建筑多边形的化简。在训练过程中,模型使用结构化的样本数据,设计了忽略特定索引的交叉熵损失函数以提升化简质量。试验设计包括主试验与泛化验证两部分。主试验基于洛杉矶1∶2000建筑数据集,分别采用0.2、0.3和0.5 mm 3种网格尺寸对多边形进行编码,实现了目标比例尺为1∶5000与1∶10000的化简。试验结果表明,在0.3 mm的网格尺寸下模型性能最优,验证集上的化简结果与人工标注的一致率超过92.0%,且针对北京部分区域的建筑多边形数据的泛化试验验证了模型的迁移能力;与LSTM模型的对比分析显示,在参数规模相近的条件下,LSTM模型无法形成有效收敛,并生成可用结果。本文证实了Transformer在处理空间几何序列任务中的潜力,且能够有效复用已有化简样本,为智能建筑多边形化简提供了具有工程实用价值的途径。
