织构化热处理对钢轨钢耐磨性和寿命的影响

目的减小钢轨钢表面因离散硬化产生的硬化区与基材边界处的硬度突变,从而消除磨损过程中硬化区与基材边界处因材质不连续而产生的早期碎裂现象,进一步提高钢轨钢的耐磨性和寿命。方法提出了一种基于层流等离子体束离散硬化的表面织构化热处理方法。与之前的离散圆点状硬化不同,通过调节等离子体束在试样表面的运动速度,改变试样表面的加热时间,在硬化区与基材之间人为地附加了一个过渡区。为了揭示织构化热处理对钢轨钢滚滑性能的影响,在摩擦磨损试验机上进行了磨损试验。磨损后的试样经切割、打磨、抛光和腐蚀,在光学显微镜和扫描电子显微镜下进行了观察,以分析试样的磨损过程和织构化热处理提高试样耐磨性的机理。结果由于附加过渡区的存在,相变区的一边被拉长,形状变为泪滴状。硬化区到基材的硬度变化率由3500 HV/mm降低至200 HV/mm,在圆点状硬化中可观察到的硬度突变消失。同时,在磨损实验中,泪滴状硬化试样的磨损量较圆点状硬化试样降低了27%,边界碎裂消失。结论具有附加过渡区的试样比圆点状硬化的试样具有更好的耐磨性和抗塑性变形能力,该方法对提高钢轨钢的使用寿命具有一定的应用前景。     

基于自训练SGAN的调压器半监督故障诊断模型

针对实际工况下调压器标记样本稀缺导致故障诊断效果不佳的问题,基于自训练算法与半监督生成对抗网络(semi-supervisedgenerativeadversarialnetwork,SGAN)设计了故障诊断模型。首先,对燃气调压器一维压力信号进行预处理,得到灰度图像样本。之后,基于深度卷积生成对抗网络,设计SGAN进行特征提取,判别器采用具有共享权值的堆叠鉴别器模型。然后,设计自训练算法,使用训练好的初始分类器预测无标签样本的类别标签。最后,采用重复标记方式将满足要求的样本扩充到有标签样本集重新训练,保存最终的分类器。实验结果表明,在少量调压器标签样本的情况下,所提模型依旧具有良好的性能。     

Design and characteristics of a triple-cathode cascade plasma torch for spheroidization of metallic powders

Arc plasma torch is an effective tool for spheroidization of metallic powders. However, as most conventional plasma torches were not specifically designed for plasma spheroidization, they may exhibit the disadvantages of the radial injection of powders, large fluctuations in the arc voltage, large gas flow rate, and disequilibrium between multiple plasma jets during the spheroidization process. Therefore, this paper presents a triple-cathode cascade plasma torch (TCCPT) for plasma spheroidization. Its structural design, including three cathodes, a common anode, and three sets of inter-electrodes, are detailed to ensure that powders can be inserted into the plasma jet by axial injection, the arc voltage fluctuations are easily maintained at a low level, and the plasma torches can work at a relatively small gas flow rate. Experimental results showed that the proposed TCCPT exhibits the following characteristics: (1) a relatively small arc voltage fluctuation within 5.3%; (2) a relatively high arc voltage of 75 V and low gas flow rate range of 10–30 SLM; (3) easy to be maintained at the equilibrium state with the equilibrium index of the three plasma jets within 3.5 V. Furthermore, plasma spheroidization experiments of SUS304 stainless steel powers were carried out using the proposed TCCPT. Results verified that the proposed TCCPT is applicable and effective for the spheroidization of metallic powders with wide size distribution.     

基于自注意力机制辅助分类生成对抗网络的轴承故障诊断

针对传统判别式轴承故障诊断算法在复杂工况下依赖人工特征提取、诊断效果不佳的问题,提出将生成式模型辅助分类生成对抗网络(auxiliary classifier generative adversarial network,ACGAN)用于轴承故障诊断研究。首先,通过快速傅里叶变换将轴承振动信号转为2维频域特征灰度图,设计卷积网络作为模型主体结构,添加批量归一化和LeakyReLU激活函数缓解梯度消失问题;其次,引入自注意力机制(self-attention mechanism,SA),将数据中相距较远的特征相互关联建立新的SA-ACGAN模型,实现多分类场景下对原始数据分布特征的有效学习;最后,将模型应用于电机轴承进行对比验证,结果表明所提方法故障诊断准确率高达99.7％,并具有良好的鲁棒性和泛化性。