【作者】 马辉；

【导师】 夏士雄；

【作者基本信息】 中国矿业大学 ， 计算机系统结构， 2020， 硕士

【摘要】 矿井提升机是广泛应用于煤矿产业的关键设备,其稳定性和安全性是煤矿生产和井下工作人员生命的重要保障。滚动轴承是提升机设备中的核心部件,其运行状态直接影响到提升机的效率及安全。如何对轴承部件的故障进行高效精准的诊断,进而实现对设备的智能维护,是当前大型煤矿生产中亟待解决的问题。首先,针对传统诊断方法中人工提取特征的不稳定性,以及数据稀缺的情况,本文提出了一种基于深度神经网络的双层次故障诊断算法,对无噪声的轴承振动数据进行精准的故障检测。算法通过滑动窗口重叠采样技术扩充数据集,使用自编码技术提取数据中的特征。将所提取的特征作为双层次神经网络的输入,采用反向传播算法训练双层次故障诊断分类器。最后通过集成学习投票决定故障的具体分类。本文利用凯斯西储大学的轴承公开数据集,模拟理想状态无噪声环境下的轴承故障诊断,从而验证所提算法的有效性,并将所提算法与经典的支持向量机、反向传播算法进行了对比实验,实验结果表明所提算法的故障诊断平均准确率约为99.95%,明显高于对比算法。其次,针对工业数据实际噪声较大以及机械运行负载变化的问题,本文提出了一种基于域适应批标准化的双层次神经网络故障诊断算法。所提算法将自适应批标准化方法与无噪双层次故障诊断算法融合,并且增加Dropout层,提高网络的泛化能力。在训练阶段中,以反向传播算法训练神经网络模型,记录所有批标准化层的缩放变量和平移变量。在诊断阶段,若测试集与训练集域分布不同,则计算测试集的均值与方差并用其替换掉已训练模型中各批标准化层的均值与方差,再进行缩放和平移,从而达到域适应的效果。将所提方法在不同噪声,不同负载以及噪声变负载混合三种情况下对其故障诊断性能分别进行了验证。实验结果表明所提方法具有较高的抗噪性、变负载能力以及故障诊断准确率。综上所述,本文对提升机轴承无噪故障诊断以及有噪且变负载的故障诊断问题进行了研究,分别提出了基于深度神经网络的双层次故障诊断算法以及基于域适应批标准化的双层次神经网络故障诊断算法。实验结果表明,所提算法可以出色完成复杂工作环境状态下的提升机轴承故障诊断任务,对煤矿的故障诊断及智能生产具有一定指导作用。该论文有图21幅,表10个,参考文献111篇。更多还原

【Abstract】 Mine hoist is a key equipment widely used in the coal mine industry.Its stability and good safety conditions are important guarantees for the coal mine production and life of underground workers.Rolling bearings are the core components of the hoisting equipment,and their operating status directly affects the efficiency and safety of the hoisting equipment.How to carry out efficient and accurate diagnosis of bearing components and then realize intelligent maintenance of equipment are several urgent problems to be solved in the current large-scale coal mine production.First,aiming at the instability of artificially extracted features and data scarcity in traditional diagnostic methods,this paper proposes a double-layer fault diagnosis algorithm based on deep neural network.This algorithm can perform accurate fault detection on noise-free bearing vibration data.Firstly,the original data set is expanded by the sliding window overlap sampling technique,and then,self-autoencoder technology is used to extract the features of the data.Subsequently,using the extracted features as the input of the double-layer neural network,the back-propagation algorithm is used to train the fault diagnosis classifier.Finally,the specific classification of the fault is decided by voting based on the ensemble learning.In this paper,the bearing public data set of Case Western Reserve University is used to simulate bearing fault diagnosis in an ideal and noise-free environment to verify the effectiveness of the proposed algorithm.The proposed algorithm is compared with the classic support vector machine and back propagation neural network algorithm.Experimental results show that the average accuracy of the proposed algorithm for fault diagnosis is about 99.95%,which is significantly higher than comparison algorithms.Second,aiming at the problems of large actual noise of industrial data and variable load of mechanical operation,this paper proposes a double-layer neural network fault diagnosis algorithm based on adaptive batch normalization.The proposed algorithm integrates the adaptive batch normalization method with the noise-free fault diagnosis algorithm,and the dropout layer is added to improve the generalization ability of the network.In the training phase,the neural network model is trained with back-propagation algorithm.Then,the scaling variables and translation variables of all batch normalization layers are recorded.In the diagnosis phase,if the test set and the training set have different domain distributions,the mean and variance of the test set are calculated to replace the training mean and variance of each batch normalization layer in the trained model.We perform the normalization to the result,and then scaled and translated to achieve the effect of domain adaptation.The proposed method is validated for its fault diagnosis performance under different noise,variable load,and mixed noise-to-load conditions.Experimental results show that the proposed method has high noise immunity,variable load capacity and accuracy of fault diagnosis.In summary,this paper studies the noise-free fault diagnosis of hoist bearings and the fault diagnosis problems of noisy and variable load,and respectively proposes a double-layer fault diagnosis algorithm based on deep neural network and a double-layer neural network fault diagnosis algorithm based on adaptive batch normalization.The experimental results show that the proposed algorithm can excellently perform the fault diagnosis task of hoist bearing under complex working environment conditions,which has certain guidance for fault diagnosis and intelligent production of coal mine.There are 21 pictures,10 tables and 111 references in this paper.更多还原