【作者】 王鹏；

【导师】 高金吉；

【作者基本信息】 北京化工大学 ， 计算机应用技术， 2012， 硕士

【摘要】 当前旋转设备正朝着高功率、智能化、一体化方向发展,有效地对机械设备进行状态监测,在故障早期发展阶段及时发现并予以相应维护,是大型旋转机械设备运行安全性和可靠性的保证。而滚动轴承作为旋转机械设备组的重要组成部分,因此针对于滚动轴承的状态监测、故障诊断和趋势预测的研究就具有极其重要的现实意义。滚动轴承故障诊断融合了机械动力学、现代测试技术、现代信号处理、数据挖掘和人工智能等多学科知识,主要包括信号预处理、特征提取、模式识别和趋势预测四个阶段,其主要任务是提取能反映设备状态的故障特征量,判别故障类型,预测分析故障特征量的发展趋势,根据故障严重程度制定适当的维修计划。本文采用EMD和小波分析两种预处理方法按照频段或频率将信号分解去除噪声成分和其它干扰信息；并提出一种基于“能量-香农熵比’的小波基选取方法,该方法将小波分解过程中的频带能量泄露降到最低。特征提取方面,提出了RMS序列及基于RMS序列的相对熵,其计算简单,抑噪能力强,可有效处理故障早期或信噪比较低情况下的声发射信号；并介绍了近似熵及其快速算法,分析阐述了计算过程中各参数对熵值和计算时间的影响。本文利用改进粒子群优化的神经网络对滚动轴承故障进行模式识别。基于适应度值对粒子群算法优化,通过对标准速度更新公式中各参数和公式本身进行改进,并且结合其它智能算法来完善其搜索策略,突出了粒子在不同阶段的全局和局部搜索能力,有效规避了搜索过程中粒子陷入局部最优点的可能性。最后,采取基于遗传算法的回归预测模型对故障发展趋势进行预测,利用遗传算法优化不同阶次回归模型中的系数。通过轴承外圈故障在强负载不良润滑下的剩余寿命预测实验对此预测算法进行试验验证。更多还原

【Abstract】 Modern rotating equipment develops in a way of large-scale, complication, automation and high energy consumption. Effective condition monitoring of mechanical equipment, timely detection of failure in early stage and corresponding maintenance, are the foundation of secure and reliable operation of large-scale rotating mechanical equipments. Rolling element bearing is the significant component of rotating mechanical equipment, and condition monitoring, fault diagnosis and trend prediction of which is of important practical significance.Fault diagnosis of rolling bearing integrates mechanical dynamics, modern measuring and testing technology, modern signal processing, data mining and artificial intelligent, etc. It includes pre-processing, feature extraction, pattern recognition and trend prediction four parts, and the purpose of which is to extract fault characteristic parameters that could reflect operation status of equipments, identify fault type, predict development tendency of those parameters, and make appropriate maintenance plan according to fault severity. Preprocessing methods adopted in this paper is EMD and wavelet analysis, eliminating noise element through decomposition by frequence and frequence range.’Energy to Shannon Entropy’ is proposed to select wavelet base, reducing energy leakage of frequence range in wavelet decomposition. During feature extraction aspect, RMS index and relative entropy based on RMS index are put forward, which are of simple calculation and strong robustness against noise, could effectively process with AE signal under low SNR and early fault stage. Approximate entropy and its fast algorithm are introduced, and the effect of parameters in calculation process to entropy value and computing time is discussed.Neural network based on improved particle swarm optimization is proposed in this paper for pattern recognition of rolling bearing fault. PSO is impoved based on fitness, by modifying parameters and formula itself of standard velocity updating formula and consummating search strategy through combining with other intelligent algorithms. Improved PSO extrudes global search capability or local search capability in different stage, effectively avoding the possibility of falling into local minimum in search process. Finally, regression predition model based on genetic algorithm is introduced to predict fault trend, using genetic algorithm to modify coefficients in different degree regression model. And this method is tested through an residual life prediction experiment of bearing outer-race failure under strong load and badlubrication.更多还原