带式输送机的托辊状态监测及材料失效相关性的研究

RESEARCH ON CONDITION MONITORING OF BELT CONVEYOR ROLLERS AND CORRELATION OF MATERIAL FAILURE

王高山

11930216

硕士

材料工程

吴景深

2021-05-23

2021-06-15

南方科技大学

深圳

带式输送机主要负责矿石等散装物料输送，因其具有成本低、效率高、传输距离远等优点，在矿石开采行业中被广泛采用。带式输送机是运输系统的中枢环节，若其发生故障，不仅会造成经济损失，甚至可能导致人员伤亡等严重事故。托辊作为矿用带式输送机的主要承载部件，其数量较多，托辊失效是导致输送机系统发生安全事故的最主要因素。目前，托辊的故障检测主要依赖人工巡检，存在效率低、准确性差、人力成本高等不足。针对当前现状，本文通过采集托辊在运行过程中的声学信号，结合信号处理、特征提取、故障诊断等方法，对托辊健康监测展开相关研究。本文深入分析托辊故障机理及故障演化规律，确定以声学信号作为托辊的状态监测参数。同时搭建了小型带式输送机托辊的实验平台，采集四种不同健康状态的托辊分别在不同转速下旋转时的声学信号，开展实验研究。通过结合时域统计特征参数、梅尔倒谱系数、小波包能量谱、经验模态分解-奇异值分解（EMD-SVD）四种不同的信号处理技术，本文从托辊音频信号中提取出30个信号特征，形成特征向量。并且，对每一种方法提取出的托辊特征参数进行相关分析，验证这些特征提取技术对于判断托辊健康状态的有效性。实验中对经过信号处理技术提取到的30维特征向量进一步采用主成分分析（PCA）进行降维处理，提取95%主成分（共9维）构成新特征矩阵，作为支持向量机（SVM）算法的输入。基于SVM对托辊进行故障诊断，结果表明，将数据样本分成正常与故障两类情况，预测正确率达到99.4%；将数据样本分成正常托辊、轴承磨损、轴承与轴间无润滑油、筒皮磨破四类情况，预测正确率达到94.3%。实验结果有效证明了本实验提出的特征提取及分类算法对于区分正常托辊与故障托辊，及判别故障托辊的不同失效模式具有很好效果。本文研究成果可望为带式输送机托辊的健康监测提供一定的理论依据。论文最后对全文进行了总结，并对课题未来的研究方向进行了展望。

Belt conveyors are widely used in the ore mining industry because of the advantages of low cost, high efficiency, and long transmission distance. The belt conveyor is the core part of the transportation system. The failure of the belt conveyor will not only cause economic losses, but may even lead to serious accidents and casualties. The rollers are the main load-bearing component of mining belt conveyors. The number of rollers is large, and roller failure is the most important factor leading to service failures of the conveyor system. At present, the failure detection of rollers mainly relies on manual inspection, which has the weaknesses such as low efficiency, poor accuracy, and high labor costs. Considering current situation, this paper is focus on roller fault detection based on the acoustic signals of the rollers during operation.This paper analyzes the failure mechanism and failure evolution of rollers, and the acoustic signal is chosen as the key condition monitoring parameter of the rollers. What’s more, an experimental platform of a small-size belt conveyor is designed and established. The experimental research is carried out to record the audio signals of rollers with different conditions at different speeds.During the experiment, a feature map with 30 features is constructed based on four different signal processing technologies, including time-domain statistical feature parameters, Mel-cepstrum coefficients, wavelet packet energy spectrum, and empirical mode decomposition-singular value decomposition (EMD-SVD). Moreover, correlation analysis is carried out to verify the effectiveness of these 30 features.Principal component analysis (PCA) is applied for dimensionality reduction of the 30 features, where the 95% dominate proportion, the largest 9 principal components, are selected to form a new feature map, which is used as the input of SVM models. The result of SVM models shows that a high prediction accuracy of 99.4% can be achieved when the data samples are divided into two categories, the normal rollers and rollers with defects. Moreover, the prediction accuracy still keeps at 94.3% when the data samples are divided into four categories, including normal rollers, rollers with worn bearings, rollers with no lubricating oil between the bearing and shaft, and rollers with worn-out rollers. The high prediction accuracy indicates that the feature extraction and classification model used in this paper have a good performance on fault detection of rollers.At last, a conclusion is given and future prospects are pointed out. The research outcomes of this paper are expected to provide a certain theoretical basis for the health monitoring of belt conveyor rollers.

带式输送机 托辊 信号处理 特征提取 故障诊断

belt conveyor roller signal processing feature extraction fault diagnosis

中文

独立培养

南方科技大学

王高山. 带式输送机的托辊状态监测及材料失效相关性的研究[D]. 深圳. 南方科技大学,2021.