数控机床主轴组件故障的知识发现研究
发布时间：2018-08-30 15:23
【摘要】：数控机床故障诊断及维护是机床调试和使用过程中的重要组成部分,是目前制约数控机床发挥正常作用的主要因素之一。现有数控机床故障自诊断系统能够诊断常见电器系统故障及简单的与系统相连部件的故障,但故障出现率较高且引起机床加工质量下降的机械类故障的自诊断基本上还是盲点,而主轴组件的故障在此类故障中占了相当比重,这也一直是国内外数控机床故障诊断领域的难题。 本文针对性提出从软计算理论的全新视角解决该问题,对在获取故障知识的数据准备阶段和知识发现阶段的几个关键问题展开了较为深入的研究和探索。 在知识获取的数据准备阶段,进行了两个方面的研究工作。 首先选取数控机床主轴系统的两大组件即滚动轴承和齿轮作为研究对象,通过对比分析它们与一般机械振动的机理后得到结论,即滚动轴承故障主要表现为表面磨损和剥落,而主轴齿轮最主要的故障来源于运动中产生的齿面均匀磨损和局部剥落故障。对前者,在进行知识获取过程提取特征时可以用基频及其整数或分数倍频处幅值为特征参数；对后者,可以依据振动信号啮合频率及其两侧产生的边频带的组合频谱诊断故障。针对主轴齿轮故障数据获取时的测点布置优化问题,采用有限元建模分析和谐响应分析,确定出主轴箱振动测点的理论最佳位置。搭建了以上两种组件故障模拟实验系统,为后续研究工作获取原始数据做了准备。 其次,从数据采集和处理角度,特别提出使用一个三阶低通巴特沃斯滤波器和一个三阶高通巴特沃斯滤波器建立的带通滤波器进行滤波,并对该滤波过程进行了数学分析；为了实现将传感器获取数据的融合,针对单一传感器数据融合的时问性问题,提出了结合算术均值与递推估计的数据融合方法,获得了比算术平均值更可靠的测量结果,而针对多传感器数据融合的空间性问题,提出一种多传感器数据的加权融合算法,不同的传感器按照相应的权数,在总均方误差最小这一最优条件下,根据各个传感器所得到的测量值以自适应的方式寻找其对应的权数,使融合后的数据结果达到最优,并提出采用信息熵来评价数据融合的效果。 在故障数据的知识发现过程阶段,分别对两种组件的故障采取不同软计算方法获取了故障知识规则,实现了故障诊断。 针对滚动轴承故障实验所获取数据,分别运用基于等间距聚类与属性重要度约简算法和基于k-均值聚类与区分矩阵约简算法,均实现表面磨损和剥落故障及正常状态三种模式的知识及规则的获取。 针对数控机床主轴齿轮的典型故障诊断,构建了一种具有三层网络结构模型的BP神经网络,经过实验数据样本的训练和仿真,实例结果验证了该方法可以实现对齿轮齿面均匀磨损故障、齿面局部剥落故障以及正常状态的识别。
[Abstract]:The fault diagnosis and maintenance of NC machine tool is an important part in the process of debugging and using, and it is one of the main factors restricting the normal function of NC machine tool at present. The existing CNC machine tool fault self-diagnosis system can diagnose the common electrical system faults and simple faults connected with the system, but the self-diagnosis of mechanical faults, which have a high occurrence rate and cause the machine tool machining quality to decline, is basically a blind spot. The malfunction of spindle assembly occupies a considerable proportion in this kind of fault, which has always been a difficult problem in the field of fault diagnosis of CNC machine tools at home and abroad. In this paper, a new perspective of soft computing theory is proposed to solve this problem, and some key problems in the stage of data preparation and knowledge discovery of fault knowledge acquisition are studied and explored deeply. In the data preparation stage of knowledge acquisition, two aspects of research work are carried out. Firstly, two main components of the spindle system of CNC machine tools, that is, rolling bearings and gears, are selected as the research objects. By comparing them with the mechanism of general mechanical vibration, the conclusion is drawn that the fault of rolling bearings is mainly manifested by surface wear and spalling. The main fault of spindle gear is caused by uniform wear and local spalling. For the former, the fundamental frequency, its integer or fractional frequency amplitude can be used as the feature parameter in the process of knowledge acquisition, and the fault can be diagnosed according to the meshing frequency of the vibration signal and the combined frequency spectrum of the edge band generated by both sides of the vibration signal. Aiming at the optimization of measuring point arrangement when the fault data of spindle gear is acquired, the theoretical optimum position of vibration measuring point of spindle box is determined by using finite element modeling and harmonious response analysis. The above two component fault simulation experiment systems are built to prepare for the subsequent research work to obtain the original data. Secondly, from the point of view of data acquisition and processing, a band-pass filter based on a third-order low-Tombatworth filter and a third-order high-Tunbartworth filter is proposed to filter, and the process of the filter is analyzed mathematically. In order to achieve the fusion of sensor data acquisition, a data fusion method combining arithmetic mean and recursive estimation is proposed to solve the temporal problem of single sensor data fusion. The measurement results are more reliable than arithmetic average. Aiming at the spatial problem of multi-sensor data fusion, a weighted fusion algorithm for multi-sensor data is proposed. According to the corresponding weights, different sensors are optimized under the optimal condition of minimum total mean square error. According to the measured values obtained from each sensor, the corresponding weights are found in an adaptive manner, so that the results of the fused data are optimized, and the information entropy is proposed to evaluate the effect of the data fusion. In the process of knowledge discovery of fault data, different soft computing methods are used to obtain fault knowledge rules and fault diagnosis is realized. In view of the data obtained from rolling bearing fault experiment, the algorithm based on equidistant clustering and attribute importance reduction and the algorithm based on k-means clustering and discernibility matrix reduction are used, respectively. The knowledge and rules of surface wear and peeling fault and normal state are obtained. Aiming at the typical fault diagnosis of spindle gear of NC machine tool, a BP neural network with three-layer network structure model is constructed, which is trained and simulated by experimental data sample. The results show that the method can recognize the uniform wear fault, the local spalling fault and the normal state of gear tooth surface.
【学位授予单位】：兰州理工大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TG659;TH165.3
[Abstract]:The fault diagnosis and maintenance of NC machine tool is an important part in the process of debugging and using, and it is one of the main factors restricting the normal function of NC machine tool at present. The existing CNC machine tool fault self-diagnosis system can diagnose the common electrical system faults and simple faults connected with the system, but the self-diagnosis of mechanical faults, which have a high occurrence rate and cause the machine tool machining quality to decline, is basically a blind spot. The malfunction of spindle assembly occupies a considerable proportion in this kind of fault, which has always been a difficult problem in the field of fault diagnosis of CNC machine tools at home and abroad. In this paper, a new perspective of soft computing theory is proposed to solve this problem, and some key problems in the stage of data preparation and knowledge discovery of fault knowledge acquisition are studied and explored deeply. In the data preparation stage of knowledge acquisition, two aspects of research work are carried out. Firstly, two main components of the spindle system of CNC machine tools, that is, rolling bearings and gears, are selected as the research objects. By comparing them with the mechanism of general mechanical vibration, the conclusion is drawn that the fault of rolling bearings is mainly manifested by surface wear and spalling. The main fault of spindle gear is caused by uniform wear and local spalling. For the former, the fundamental frequency, its integer or fractional frequency amplitude can be used as the feature parameter in the process of knowledge acquisition, and the fault can be diagnosed according to the meshing frequency of the vibration signal and the combined frequency spectrum of the edge band generated by both sides of the vibration signal. Aiming at the optimization of measuring point arrangement when the fault data of spindle gear is acquired, the theoretical optimum position of vibration measuring point of spindle box is determined by using finite element modeling and harmonious response analysis. The above two component fault simulation experiment systems are built to prepare for the subsequent research work to obtain the original data. Secondly, from the point of view of data acquisition and processing, a band-pass filter based on a third-order low-Tombatworth filter and a third-order high-Tunbartworth filter is proposed to filter, and the process of the filter is analyzed mathematically. In order to achieve the fusion of sensor data acquisition, a data fusion method combining arithmetic mean and recursive estimation is proposed to solve the temporal problem of single sensor data fusion. The measurement results are more reliable than arithmetic average. Aiming at the spatial problem of multi-sensor data fusion, a weighted fusion algorithm for multi-sensor data is proposed. According to the corresponding weights, different sensors are optimized under the optimal condition of minimum total mean square error. According to the measured values obtained from each sensor, the corresponding weights are found in an adaptive manner, so that the results of the fused data are optimized, and the information entropy is proposed to evaluate the effect of the data fusion. In the process of knowledge discovery of fault data, different soft computing methods are used to obtain fault knowledge rules and fault diagnosis is realized. In view of the data obtained from rolling bearing fault experiment, the algorithm based on equidistant clustering and attribute importance reduction and the algorithm based on k-means clustering and discernibility matrix reduction are used, respectively. The knowledge and rules of surface wear and peeling fault and normal state are obtained. Aiming at the typical fault diagnosis of spindle gear of NC machine tool, a BP neural network with three-layer network structure model is constructed, which is trained and simulated by experimental data sample. The results show that the method can recognize the uniform wear fault, the local spalling fault and the normal state of gear tooth surface.
【学位授予单位】：兰州理工大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TG659;TH165.3
【参考文献】
相关期刊论文 前10条
1 李云松;任艳君;;智能诊断技术发展综述[J];四川兵工学报;2010年04期
2 安治永;李应红;苏长兵;;航空电子设备故障诊断技术研究综述[J];电光与控制;2006年03期
3 董立新,肖登明,王俏华,吕干云,刘奕路;模糊粗糙集数据挖掘方法在电力变压器故障诊断中的应用研究——基于油中溶解气体的分析诊断[J];电力系统及其自动化学报;2004年05期
4 李建兰;黄树红;张燕平;;基于Rough Set理论的典型振动故障诊断[J];动力工程;2008年01期
5 董辛e，
本文编号：2213487
本文链接：https://www.wllwen.com/kejilunwen/jixiegongcheng/2213487.html
