针对滚动轴承故障诊断中样本分布不均衡引起的模型泛化能力差、诊断精度低的问题,从两个方面展开研究:(1)故障样本增广,提出结合变分自编码器(VAE)和生成对抗网络(GAN)的VAE-GAN样本增广模型;(2)改进分类算法,提出基于焦点损失(FL)和卷...针对滚动轴承故障诊断中样本分布不均衡引起的模型泛化能力差、诊断精度低的问题,从两个方面展开研究:(1)故障样本增广,提出结合变分自编码器(VAE)和生成对抗网络(GAN)的VAE-GAN样本增广模型;(2)改进分类算法,提出基于焦点损失(FL)和卷积神经网络(CNN)的FLCNN(focal loss and convolutional neural network)样本分类模型。在此基础上,将VAE-GAN和FLCNN融合,构建VAE-GAN+FLCNN轴承故障诊断模型。首先,将样本量少的故障类输入VAE-GAN模型,通过交替训练编码网络、生成网络和判别网络,学习出真实故障样本的数据分布,从而实现故障样本的增广;然后用增广后的数据样本训练FLCNN分类模型,完成轴承故障识别。试验对比结果表明,所提方法能够有效提升样本不均衡条件下的轴承故障诊断效果,拥有更高的Recall值和F1-score值。展开更多
Nitrobenzene-containing industrial wastewater was degraded in the presence of ozonecoupled with H2O2 by high gravity technology. The effect of high gravity factor, H2O2 concentration, pH value, liquid flow-rate, and r...Nitrobenzene-containing industrial wastewater was degraded in the presence of ozonecoupled with H2O2 by high gravity technology. The effect of high gravity factor, H2O2 concentration, pH value, liquid flow-rate, and reaction time on the efficiency for removal of nitrobenzene was investigated. The experimental results show that the high gravity technology enhances the ozone utilization efficiency with O3 /H2O2 showing synergistic effect. The degradation efficiency in terms of the COD removal rate and nitrobenzene removal rate reached 45.8% and 50.4%, respectively, under the following reaction conditions, viz.: a high gravity factor of 66.54, a pH value of 9, a H2O2 /O3 molar ratio of 1:1, a liquid flow rate of 140L/h, an ozone concentration of 40mg/L, a H2O2 multiple dosing mode of 6mL/h, and a reaction time of 4h. Compared with the performance of conventional stirred aeration mixers, the high gravity technology could increase the COD and nitrobenzene removal rate related with the nitrobenzene-containing wastewater by 22.9% and 23.3%, respectively.展开更多
文摘针对滚动轴承故障诊断中样本分布不均衡引起的模型泛化能力差、诊断精度低的问题,从两个方面展开研究:(1)故障样本增广,提出结合变分自编码器(VAE)和生成对抗网络(GAN)的VAE-GAN样本增广模型;(2)改进分类算法,提出基于焦点损失(FL)和卷积神经网络(CNN)的FLCNN(focal loss and convolutional neural network)样本分类模型。在此基础上,将VAE-GAN和FLCNN融合,构建VAE-GAN+FLCNN轴承故障诊断模型。首先,将样本量少的故障类输入VAE-GAN模型,通过交替训练编码网络、生成网络和判别网络,学习出真实故障样本的数据分布,从而实现故障样本的增广;然后用增广后的数据样本训练FLCNN分类模型,完成轴承故障识别。试验对比结果表明,所提方法能够有效提升样本不均衡条件下的轴承故障诊断效果,拥有更高的Recall值和F1-score值。
文摘处理了Lageos-1,Lageos-2,Etalon-1和Etalon-2共4颗卫星在1993―2017年的全球卫星激光测距(satellite laser raging,SLR)观测数据,并对国际激光测距服务(International Laser Ranging Service,ILRS)各分析中心(analysis centers,ACs)的周解进行了技术内综合,分别通过几何法(参考SLRF2008和SLRF2014)和直接法(针对Lageos-1和Lageos-2)确定了SLR地心运动序列;利用傅里叶变换和最小二乘法分析了地心运动的长期变化、周年变化和半年变化,并与CSR(the center for space research)提供的动力学法结果进行了比较和分析。结果表明:地心运动的长期项比周期项小1个量级;与SLRF2008几何法相比,用SLRF2014几何法得到的地心运动长期项明显较小,特别是在Z方向;使用直接法分别针对Lageos-1和Lageos-2卫星激光测距数据解算的地心运动半年项有较好的一致性,但周年项相位相差较大,且在Z方向与几何法及CSR提供的动力学法结果差异也很大,这可能与该方法解算的地心运动与测站偏心改正强相关有关,也可能与单颗星解算地心运动几何结构不够好有关;使用几何法得到的地心运动与CSR动力学结果非常接近,证明利用上海天文台和ILRS ACs周解进行技术内综合来确定地心运动的方法是可行的。
基金financially supported by the National Natural Science Foundation of China(21206153)Science and Technology Development Program Fund of Taiyuan City(120164053)
文摘Nitrobenzene-containing industrial wastewater was degraded in the presence of ozonecoupled with H2O2 by high gravity technology. The effect of high gravity factor, H2O2 concentration, pH value, liquid flow-rate, and reaction time on the efficiency for removal of nitrobenzene was investigated. The experimental results show that the high gravity technology enhances the ozone utilization efficiency with O3 /H2O2 showing synergistic effect. The degradation efficiency in terms of the COD removal rate and nitrobenzene removal rate reached 45.8% and 50.4%, respectively, under the following reaction conditions, viz.: a high gravity factor of 66.54, a pH value of 9, a H2O2 /O3 molar ratio of 1:1, a liquid flow rate of 140L/h, an ozone concentration of 40mg/L, a H2O2 multiple dosing mode of 6mL/h, and a reaction time of 4h. Compared with the performance of conventional stirred aeration mixers, the high gravity technology could increase the COD and nitrobenzene removal rate related with the nitrobenzene-containing wastewater by 22.9% and 23.3%, respectively.