含砂雾封层性能试验研究

通过对含砂雾封层封水性能、抗滑性能及耐磨性能进行试验研究,结果表明:总体上不同含砂雾封层及涂刷量对封水性能、抗滑性能及耐磨性能影响存在差异,3~#含砂雾封层封水性能最好,2~#含砂雾封层抗滑性能最好,1~#含砂雾封层耐磨性能最好;建议在实体工程中如果要较大规模使用含砂雾封层,应尽量保证材料供应稳定唯一,洒布量应通过试验段性能验证,综合分析慎重确定洒布量。     

High-speed friction and wear behaviors of bulk Ti3SiC2

High-speed friction and wear behaviors of bulk Ti3 SiC2 sliding drily against low carbon steel were investigated. Tests were carried out using a block-on-disk type tester with normal pressures ranging from 0.1 to 0.8 MPa and several sliding speeds from 20 to 60 m/s. The results show that, in the case of sliding speeds of 20 40 m/s, the friction coefficient exhibits a decreasing tendency with increasing the normal pressure after an increment in the smaller pressure range, and the worn quantity of Ti3SiC2 exhibits a nearly linear increase with increasing the normal pressure. However, when the sliding speed is up to 60 m/s, the friction coefficient exhibits a monotonous increase and the worn quantity exhibits a quadric increase with increasing the normal pressure. These speed-dependent and pressure-dependent behaviors are attributed to the antifriction effects of a frictionally generated oxide film covering the friction surface of Ti3 SiC2, and a balance between the generating rate and the removing (wearing) rate of the film.     

基于生成对抗网络的PPG-ECG信号转换方法

心电（Electrocardiogram, ECG）信号的长期检测与评估对心血管疾病的诊断和预防至关重要。心电信号的检测通常需要在患者身上安装电极,易使受试者产生不适感,适用范围有限。相对而言,使用光电容积描记法（Photoplethysmography, PPG）检测得到的脉搏波（Pulse wave）信号不仅包含丰富的心血管生理和病理信息,而且易于测量。考虑到PPG与ECG信号间存在固有的映射关系,本文基于生成对抗网络（Generative adversarial network, GAN）提出了一种将PPG转换为ECG信号的模型。该模型生成器由Unet模型组成,并且在特征图融合方面参考了Unet++的结构,而其判别器由卷积神经网络组成。在训练过程中,采用梯度惩罚方式增加了生成模型的稳定性。基于公用数据集进行了实验,通过对比53名受试者样本的处理结果,新模型所生成ECG信号的均方根误差（Root mean square error, RMSE）、Pearson相关系数（ρ）和Fréchet距离（Fréchet distance, FD）三个指标分别提升了3.4%、5.5%和0.4%,证明新模型具有更好的PPG-ECG转换效果。     

汉语声调识别中的基音后处理方法

汉语是一种带有声调的语言,声调信息主要体现在韵母的基音轨迹中,但是由于提取的基音不够稳健,所以必须要对基音进行后处理.通过归纳以帧为单位和以韵母为单位的两类基音后处理方式,并在第一种方式中提出基于韵母平均值进行归一化算法,在第二种方式中提出了帧叠靠前和帧叠靠后的韵母四等分长算法,经实验结果(以标准HTK为平台)得出后者更优的结论.考虑到前后声调的影响,采用声调三音子模型进行声调识别测试,可以比单音子模型识别效果提高10%左右.