基于Triplet Network的小样本轴承、齿轮故障诊断方法

针对小样本条件下,深度神经网络在机械设备典型部件中(轴承、齿轮)故障诊断精度不高这一问题,提出了一种基于Triplet Network的机械设备典型部件(轴承、齿轮)故障诊断方法。首先,利用短时傅里叶变换,将原始时序信号转换为时频信号;然后,采用了基于Triplet Network搭建的模型,从时频信号中提取了同一故障和不同故障样本的特征,通过对比相同故障和不同故障样本特征的相似度,对模型参数进行了优化,达到了提取的同一故障样本特征相似度越来越高,不同故障样本特征相似度越来越低的效果;最后,通过比较未知样本与已知故障样本的特征相似度,实现了对机械设备典型部件(轴承、齿轮)的故障识别,并采用江南大学轴承故障数据集和康涅狄格大学齿轮故障数据集,对该故障诊断方法的有效性进行了实验验证。研究结果表明:基于Triplet Network的方法在每类训练样本只有5个的情况下,轴承故障识别率可以达到68%,齿轮故障识别率为96.8%,均优于传统的深度神经网络方法。

Extending application of asymmetric semi-circular bend specimen to investigate mixed mode Ⅰ/Ⅱ fracture behavior of granite

The asymmetric semi-circular bend(ASCB)specimen has been proposed to investigate the cracking behavior in different geo and construction materials and attracted the attention of researchers due to its advantages.However,there are few studies on the fracture toughness determination of rock materials.In this work,a series of fracture tests were performed with the ASCB specimens made of granite.The onset of fracture,crack initiation angle and crack propagating trajectory was analyzed in detail combined with several mixed mode fracture criteria.The influence of the crack length on the mode Ⅰ/Ⅱ fracture toughness was studied.A comparison between the fracture toughness ratios predicted by varying criteria and experimental results was conducted.The relationship between experimentally determined crack initiation angles and curves of the generalized maximum tangential stress(GMTS)criterion was obtained.The fracture process of the specimen was recorded with the high-speed camera.The shortcomings of the ASCB specimens for the fracture toughness determination of rock materials were discussed.The results may provide a reference for analysis of mixed mode I and II fracture behavior of brittle materials.