【作者】 张伟；

【导师】 李军霞；

【作者基本信息】 太原理工大学 ， 机械工程， 2024， 博士

【摘要】 煤炭是我国能源的“压舱石”,每年有40多亿吨煤炭通过多台串联带式输送机组成的主煤流系统完成输送。托辊是带式输送机用量最大、更换频率最高的零部件,其运行状态决定了整个运输系统的效率和使用寿命。由于托辊长时间在高速、重载、潮湿、多尘的恶劣环境下运行,轴承作为托辊内部的关键部件,极易发生旋转卡死和旋转卡顿等故障,引起托辊温度升高,进而引发火灾甚至瓦斯爆炸等恶性事故。因此,必须对托辊轴承开展故障诊断方法研究,便于及时发现托辊轴承存在的故障,确保主煤流系统安全高效运行。论文以矿用带式输送机托辊轴承为研究对象,以振动信号为切入点,结合信号分解、特征提取、特征选择与模式识别等相关理论和技术,采用理论分析、仿真建模和实验研究等手段,开展矿用带式输送机托辊轴承故障诊断方法研究。主要内容包括:(1)针对现有动力学模型无法准确描述滚动体经过故障区域诱发冲击效应的问题,开展了基于时变激励的托辊轴承故障动力学建模研究。基于Hertz接触理论研究了滚动体与滚道之间的接触特性,根据能量守恒定律和动量定理计算了滚动体碰撞故障区域产生的冲击力,建立了时变位移激励和时变接触力激励相耦合的托辊轴承六自由度动力学模型。探究了托辊轴承分别发生内圈故障和外圈故障时的振动响应规律,阐明了滚动体经过故障区域的冲击特性,并通过理论分析和模拟实验验证了所建动力学模型的有效性。(2)针对强噪声干扰下难以提取故障特征导致诊断准确率低的问题,开展了基于自适应变分模态分解和混合过滤法的托辊轴承信号降噪与特征选择方法研究。提出了一个新的故障衡量指标——综合冲击系数,用于识别故障特征,并将其作为优化算法的适应度函数。对采集到的振动信号进行自适应变分模态分解,从中挑选综合冲击系数最大的分量作为输入信号从而达到降噪目的,减少了噪声干扰对故障诊断结果的影响。对降噪后的信号进行多域特征提取从而得到高维故障特征集,采用混合过滤法从中挑选最优特征子集,并通过麻雀搜索算法优化的最小二乘支持向量机建立了故障特征与故障类型之间的映射关系。将煤矿带式输送机工作现场噪声信号分别加载至仿真数据集、凯斯西储大学轴承数据集和托辊轴承数据集中,并进行测试来验证所提方法的有效性。(3)针对基于机器学习的故障诊断方法存在特征提取过分依赖人工经验的问题,开展了基于多尺度自适应局部三值卷积神经网络的托辊轴承故障诊断方法研究。通过计算邻域像素点与中心像素点的标准差,提出了一种能够根据局部特征选取合适阈值的自适应局部三值模式,并设计了相应的卷积层。将多尺度卷积神经网络、自适应局部三值卷积层和极限学习机相结合,提出了多尺度自适应局部三值卷积神经网络模型。该模型能够从通过连续小波变换得到的二维时频图像中自适应提取特征,并确定故障形态。采用含有真实噪声的仿真数据集、凯斯西储大学轴承数据集和托辊轴承数据集进行测试,以运行时间和测试精度为评价指标,评估了所提模型的性能,在模型假设条件下,其故障诊断准确率分别为100%、99.65%和99.53%。(4)针对复杂运行环境下托辊轴承寿命难以预测的问题,开展了基于深度残差时间卷积网络的托辊轴承寿命预测方法研究。在卷积注意力模块基础上,通过对深层特征赋予不同权重来增加对重要特征信息的关注;结合时间卷积网络和软阈值函数,提出了深度残差时间卷积网络模型,能够预测相同工况下不同轴承的寿命。进一步引入无监督领域自适应策略,在缺少数据标签的情况下,依然能对不同工况下的轴承寿命进行预测,降低了模型对数据标签的依赖。采用XJTU-SY数据集、PHM 2012数据集和托辊轴承数据集进行测试,以平均绝对误差、均方根误差和自主设计的得分函数为评价指标,评估了所提模型的准确性,其平均得分分别提高了3.7%、3.9%和5.0%。更多还原

【Abstract】 Coal serves as the"ballast stone"of energy in China.Every year,over 4 billion tons of coal are transported through major coal flow systems,which are composed of numerous belt conveyors.Idlers are components with the largest consumption and the highest replacement frequency in belt conveyors,and their running state determines the efficiency and service life of the entire transportation system.Because idlers operate in a harsh environment with high speed,heavy load,humidity,and dust for a long time,bearings,as their key components,are prone to failures such as intermittent or complete stoppage of rotation,causing the temperature of the idlers to rise,leading to serious accidents such as fires and even gas explosions.Therefore,it is necessary to research fault diagnosis methods for idler bearings to timely detect faults in idler bearings and ensure the safe and efficient operation of major coal flow systems.In this paper,idler bearings of mine belt conveyors are taken as the research object,and the vibration signals are used as the focal point.By integrating relevant theories and technologies,including signal decomposition,feature extraction,feature selection,and pattern recognition,fault diagnosis methods for idler bearings of mine belt conveyors were studied.The main contents include:(1)Aiming at the problem that existing dynamic models cannot accurately describe the impact effect induced by a rolling body passing through a fault area,the fault dynamic modelling of idler bearings based on time-varying excitation was studied.Based on Hertz contact theory,the contact characteristics between the roller and the raceway were analyzed.According to the law of conservation of energy and momentum,the impact force resulting from the collision between the rolling body and the fault area was calculated.A six-degree-of-freedom dynamic model for idler bearings,coupled with time-varying displacement excitation and time-varying contact force excitation,was established.The vibration response law of idler bearings was explored when an inner ring fault or an outer ring fault occurred separately.The impact characteristics of the rolling body passing through the fault area were expounded,and the effectiveness of the dynamic model was verified through theoretical analysis and simulation experiments.(2)Aiming at the problem that it is challenging to extract fault features under strong noise interference,resulting in low diagnosis accuracy,a fault diagnosis method for idler bearings was proposed based on adaptive variational mode decomposition and a hybrid filtering method.A new fault measurement index,named the comprehensive impact coefficient,was proposed to identify fault features and serve as the fitness function for an optimization algorithm.The collected vibration signals were decomposed using adaptive variational mode decomposition,and the component with the highest comprehensive impact coefficient was selected as the input signal to mitigate the effects of noise interference on fault diagnosis results.Multi-domain feature extraction was carried out on the denoised signal to obtain a high-dimensional fault feature set.A hybrid filtering method was then employed to select the optimal feature subset,and the mapping relationship between fault features and types was established using a least squares support vector machine optimized by the sparrow search algorithm.Noise signals from the working site of mining belt conveyors were loaded into a simulation dataset,the Case Western Reserve University bearing dataset,and an idler bearing dataset,respectively,and fault diagnosis tests were carried out to verify the effectiveness of the proposed method.(3)Aiming at the problem that fault diagnosis methods based on machine learning heavily rely on manual experience,a fault diagnosis method for idler bearings based on a multi-scale adaptive local ternary convolutional neural network was proposed.By calculating the standard deviation between neighboring pixels and the central pixel,an adaptive local ternary model was proposed.This model can select the appropriate threshold based on local characteristics,and a corresponding convolution layer was designed accordingly.A multi-scale adaptive local ternary convolutional neural network model was proposed by combining multi-scale convolutional neural networks,adaptive local ternary convolutional layers,and extreme learning machines.This model can adaptively extract features and determine fault types from two-dimensional time-frequency images obtained through a continuous wavelet transform.The performance of the proposed model was evaluated using a simulation dataset with real noise,the Case Western Reserve University bearing dataset with real noise,and an idler bearing dataset with real noise.The evaluation indicators were running time and testing accuracy.Under the model’s assumptions,the fault diagnosis accuracies were 100%,99.65%,and 99.53%,respectively.(4)Aiming at the problem that it is difficult to predict the life of idler bearings in complex operation environments,a life prediction method for idler bearings based on a deep residual time convolution network was proposed.Based on the convolution block attention module,deep features were given different weights to increase attention to important feature information.By combining time convolutional networks and soft threshold functions,a deep residual time convolutional network model was proposed,capable of predicting bearing life under similar working conditions.Furthermore,an unsupervised domain adaptive strategy was introduced,enabling the prediction of bearing life under varying working conditions without data labels,thereby reducing the model’s reliance on labelled data.The XJTU-SY dataset,the PHM 2012 dataset,and an idler bearing dataset were used for experiments.The model’s prediction performance was evaluated with the average absolute error,root mean square error,and self-designed score function as evaluation indicators,and the average scores were increased by 3.7%,3.9%,and 5.0%,respectively.更多还原