基于时序数据的列车牵引系统故障预测方法

牵引系统作为列车动能转换的关键模块,如果发生故障会给整车正常运行带来重大安全隐患,所以对其进行故障预测具有重要意义。然而,传统预测方法存在高度依赖人工经验判断、不能包含大量故障特征、预测精度不足等问题。为此,文中提出一种基于时序数据的故障预测方法。利用XGBoost算法对列车牵引变流器系统的故障特征进行计算和筛选,确定与变流器故障相关性较强的关键特征;采用贝叶斯优化的LSTM模型自适应地学习多源变量数据特征,利用时间窗对特征变量数据进行截取,实现对不同类型故障的预测。实验结果表明,所提方法在预测变流器场景下的6种故障时准确率可达到91%以上。

基于MSIF-CNN的地铁车辆制动系统故障诊断方法

研究地铁车辆制动系统的故障诊断对保障交通安全、提高运营效率具有重要意义。针对当前的制动系统故障诊断研究存在过度依赖于专家的知识经验、数据融合效率不高以及现有模型训练参数过多的问题,提出了一种基于多传感器信息融合和改进卷积神经网络的“端到端”制动系统故障诊断方法。该方法不需要专家知识对数据进行特征提取,而是利用一维卷积神经网络(1D-CNN)来处理多传感器信息融合问题,并引入一维全局平均池化层(1D-GAP)改进神经网络结构,以减少模型训练参数。最终利用极端梯度提升模型(XGBoost)作为分类判别器,以提高故障诊断的准确性。实验结果表明,所提方法的准确率、精确率、召回率和F1值分别为95.86%、96.59%、92.68%和93.15%,同时,在地铁车辆制动系统故障诊断方面展现了更优的性能。

BEHAVIOUR OF BORON SEGREGATION AT GRAIN BOUNDARIES IN BCC Fe-3% Si ALLOY

The behaviour of baron segregation at grain boundaries in Fe-3% Si has been studied by means of particle tracking autoradiography(PTA).The results indicate that the tendency of equilibrium segregation of B at grain boundaries in Fe-3% Si is higher than that in austenitic alloys.No observable nonequilibrium segregation of B at grain boundaries was revealed in Fe-3% Si alloy during cooling and isothermal holding.The binding energy between boron at- oms and grain boundaries is 55.7±1.7 kJ/mol.Based on the complex mechanism of B segre- gation of quenching-induce-nonequilibrium segregation,different behaviours of B segrega- tion in γ-Fe and bcc Fe-3% Si alloy were discussed.

Effects of inclusions on microstructure and properties of heat-affected-zone for low-carbon steels

The characteristics of inclusions in two types of low-carbon steels by different deoxidization methods have been investigated by using the welding thermal simulation, the optical microscopy and scanning electron microscopy. In addition, the effects of inclusions on microstructure and properties of heat-affected-zone were studied. The nucleation and growth of intragranular acicular ferrite was observed in situ by the laser scanning confocal microscopy. The distribution of Mn element near the inclu- sion was also analyzed by the auger electron spectroscopy. The results showed that the inclusions in A1 killed steel are mainly aluminum oxides, manganese sulfide and titanium nitrides, and that the inclusions in Ti killed steel are mainly titanium oxide, manganese sulfide complex inclusion and single manganese sulfide. The auger electron spectroscopy showed that there is an Mn-depleted zone near the interface of TiOffMnS complex inclusion in the size of 1-3 gm. It could be the effective nucleus of intragranular acicular ferrite which could divide the prior austenite grains, inhibit the growth of low-temperature microstruc- ture, and refine the final microstructure, so as to improve the toughness of heat-affected-zone significantly.