Dynamic crystallization was introduced to improve the magnetic properties of NdFeB nanocrystalline permanent magnets by optimizing microstructure.The microstructure was studied by X-ray diffraction(XRD)and transmissio...Dynamic crystallization was introduced to improve the magnetic properties of NdFeB nanocrystalline permanent magnets by optimizing microstructure.The microstructure was studied by X-ray diffraction(XRD)and transmission electron microscopy(TEM).It has been determined that,compared with the conventional heat treatment,dynamic crystallization can shorten the crystallization time.Moreover,dynamic crystallization can refine grains,enhance the exchange-coupled interaction among grains,and promote the magnetic properties.As a result,the optimal magnetic properties of Nd_(10.5)(FeCoZr)_(83.4)B_(6.1)(B_(r)=0.685 T,H_(ci)=732 kA·m^(-1),H_(cb)=429 kA·m^(-1),(BH)_(m)=75 kJ·m^(-3))are obtained after dynamic crystallization heat treatment at 700℃for 10 min.展开更多
为了在变电站的低计算能力设备上部署火灾检测算法,通过多种方式的结合改进YOLOv3的网络结构,实现准确而快速的火灾检测。鉴于火灾图像数据集不足以训练深度神经网络,通过多种手段收集火灾图像,自建火灾图像数据集,并基于线上数据增强方...为了在变电站的低计算能力设备上部署火灾检测算法,通过多种方式的结合改进YOLOv3的网络结构,实现准确而快速的火灾检测。鉴于火灾图像数据集不足以训练深度神经网络,通过多种手段收集火灾图像,自建火灾图像数据集,并基于线上数据增强方法,进一步扩充数据集;鉴于原YOLOv3网络参数众多,引入MobileNetv3-Large主干网络替换原DarkNet53主干网络来降低网络复杂度,并通过在预测网络部分引入Inverted-bneck-shortcut结构实现多尺度特征图的融合预测;进一步通过引入锚框聚类优化、随机带泄漏修正线性单元(randomized leaky rectified linear unit,RLReLU)激活函数改进网络,提升算法的检测精度。实验结果表明,所提改进YOLOv3火灾检测模型的大小近似为原YOLOv3模型的1/3,推断速度提高了近12%,并且算法的平均识别精度提高了近10%,说明所提改进YOLOv3变电站火灾检测算法能较为快速和准确地识别并定位图像中的火焰。展开更多
基金This work was financially supported by New Century Excellent Person Support Program of China(No.NCET-04-0873)Science Found for Distinguished Young Scholars of Sichuan Province(No.03ZQ026-006)Major Science Plan of Sichuan Province(No.03GG009-006).
文摘Dynamic crystallization was introduced to improve the magnetic properties of NdFeB nanocrystalline permanent magnets by optimizing microstructure.The microstructure was studied by X-ray diffraction(XRD)and transmission electron microscopy(TEM).It has been determined that,compared with the conventional heat treatment,dynamic crystallization can shorten the crystallization time.Moreover,dynamic crystallization can refine grains,enhance the exchange-coupled interaction among grains,and promote the magnetic properties.As a result,the optimal magnetic properties of Nd_(10.5)(FeCoZr)_(83.4)B_(6.1)(B_(r)=0.685 T,H_(ci)=732 kA·m^(-1),H_(cb)=429 kA·m^(-1),(BH)_(m)=75 kJ·m^(-3))are obtained after dynamic crystallization heat treatment at 700℃for 10 min.
文摘为了在变电站的低计算能力设备上部署火灾检测算法,通过多种方式的结合改进YOLOv3的网络结构,实现准确而快速的火灾检测。鉴于火灾图像数据集不足以训练深度神经网络,通过多种手段收集火灾图像,自建火灾图像数据集,并基于线上数据增强方法,进一步扩充数据集;鉴于原YOLOv3网络参数众多,引入MobileNetv3-Large主干网络替换原DarkNet53主干网络来降低网络复杂度,并通过在预测网络部分引入Inverted-bneck-shortcut结构实现多尺度特征图的融合预测;进一步通过引入锚框聚类优化、随机带泄漏修正线性单元(randomized leaky rectified linear unit,RLReLU)激活函数改进网络,提升算法的检测精度。实验结果表明,所提改进YOLOv3火灾检测模型的大小近似为原YOLOv3模型的1/3,推断速度提高了近12%,并且算法的平均识别精度提高了近10%,说明所提改进YOLOv3变电站火灾检测算法能较为快速和准确地识别并定位图像中的火焰。
