基于多源信息融合的数控机床进给系统机械故障诊断研究

发布时间：2018-05-09 18:58

本文选题：数控机床 + 故障诊断　；参考：《青岛理工大学》2016年博士论文

【摘要】：数控机床是指采用数字控制系统的自动化机床,可实现自动换刀以及复杂曲线、曲面的加工,具有加工精度高、加工质量稳定、生产效率高的特点,因而成为现代制造生产中的关键技术设备,其技术发展水平与拥有数目成为体现一个国家工业现代化水平的重要标志。从结构组成来看,数控机床是集机械、电子、液压等技术于一体的复杂系统。在使用过程中,任何一个部分出现故障,均会影响机床的正常运行,尤其机械部分出现故障时,长时间的停机检修,导致整个生产线停产,造成巨大的经济损失。目前数控机床机械部件仍然广泛采用定期维护与定期更换的维修制度,这种维修制度下维修过度与维修不足的矛盾突出,一方面造成人力与物质资源的极大浪费,另一方面无法避免数控机床突发性故障的发生。因此,开展数控机床状态监测与故障诊断研究,实现维护方式由定期更换到预防维护、预知维修的转换是非常必要的。本文以信息融合技术为基础,对数控机床状态监测与故障诊断的策略、信号处理与特征提取的方法、故障模式智能识别模型的建立以及全局综合决策融合方法等进行了深入地研究,设计了基于多层次信息融合的数控机床机械部件状态监测与故障诊断系统。论文从切削力分析入手,根据数控机床载荷多变、高频冲击的工况以及加工多样性的特点,得出其切削力传播路径上的机械零部件更易发生故障的结论,分析了数控机床上机械零部件与普通设备故障成因的不同,确定了论文的研究对象以及故障失效形式。研究了基于振动、温度、电机电流、伺服误差等多种参量的故障诊断机理,并按照参量信息来源的不同,构建了由外部传感器、内部信息、程序参数以及警报信息组成的数控机床多维感知状态监测体系,从机床本体、刀具磨损、加工过程、工件加工质量等多个角度、多个方面反映数控机床运行状态,实现数控机床的全方面监测,为后续诊断过程提供充足的信息。通过实验数据分析发现仅用传统的时域与频谱分析不能对复合故障进行有效地区分,为进一步挖掘隐藏在原始信号中的故障特征,本文提出了小波包与经验模态分解联合的信号处理方法,利用小波包对信号进行降噪,并将小波重构信号扩展为高频与低频两个窄带信号,再分别对两个窄带信号进行EMD处理的方法。这种小波包与经验模态分解联合的信号处理方法,利用小波包对信号进行降噪,大大提高EMD分解的精度与质量,而且通过重构节点的扩展,可以更加细致地分析故障信息。提取EMD分解后每个IMF的能量作为特征,与时域特征、频域特征组成多域混合特征集合。基于特征之间相关性分析的特征选择方法,以模糊聚类为主要手段进行特征降维,获取敏感特征子集。根据数控机床需要诊断的对象及其故障多的特点,提出了分级诊断的策略,将诊断划分为故障定位、故障类别与程度两个层次。主网络在对故障定位的同时,负责局部子网络模型结果的聚合;局部子网络诊断具体的故障类型与程度。通过任务分工与协作,达到了简化网络结构的目的。研究了数控机床故障诊断的BP神经网络模型、RBF神经网络模型及支持向量机(SVM)模型的构建依据和方法,以敏感特征作为模型输入,分别构建了基于BP、RBF与SVM的数控机床故障诊断主网络与局部诊断子网络的模型,对比研究了三种模型对不同故障类别的诊断能力。建立了基于模糊综合评判的全局诊断模型与基于加权D-S证据理论的全局诊断模型,进一步地提高了故障识别率。首先针对数控机床模糊综合评判建模的难点,提出了以多分类器的初步诊断结果为基础,将评价因素由高维特征转变为低维的初级诊断结果,降低了模型的复杂程度,成功构建了基于信息融合的数控机床单级模糊综合评判故障诊断模型,并提出了从正确性与诊断精度两个方面评价分类器分类能力的方法,构造了基于信息熵的评价函数以及分类器整体平均的权重分配方法,减少了人为主观因素的影响。构造了分类能力评价矩阵,有效地解决了分类器对不同故障类型识别率差异较大时的权重分配问题。针对证据理论合成规则在处理高冲突证据时,得出结论与事实相悖的问题,提出了基于加权的证据理论诊断模型,以分类器故障识别率作为权重对原始证据进行加权,有效地降低了证据冲突率,故障识别率得以提高。搭建了数控机床故障诊断实验系统,对本文所提出的模型与方法进行了实验验证。
[Abstract]:CNC machine tool is an automatic machine tool with digital control system, which can realize automatic knife exchange, complicated curve and surface processing. It has the characteristics of high machining precision, stable processing quality and high production efficiency. Therefore, it has become the key technology equipment in modern manufacturing and production, and its technical development level and the number of ownership become a country. The important symbol of the level of industrial modernization. From the structure composition, the CNC machine tool is a complex system which integrates mechanical, electronic, hydraulic and other technologies. In the process of use, any part of the machine can affect the normal operation of the machine tool, especially when the mechanical part occurs the obstacle, long time stop overhaul, causing the whole production line to stop. Production has caused huge economic losses. At present, the mechanical parts of CNC machine tools are still widely used for regular maintenance and regular replacement. The contradiction between excessive maintenance and insufficient maintenance is prominent under this maintenance system. On the one hand, it causes great waste of human and material resources, and on the other hand it can not avoid the occurrence of sudden failure of CNC machine tools. Therefore, the research of state monitoring and fault diagnosis of CNC machine tools is carried out, and the maintenance mode is changed from regular to preventive maintenance. The transformation of predictive maintenance is very necessary. Based on information fusion technology, the strategy of state monitoring and fault diagnosis of CNC machine tools, the method of signal processing and feature extraction, and intelligent identification of fault mode The establishment of the model and the fusion method of global comprehensive decision making are studied deeply. The system of state monitoring and fault diagnosis of the mechanical parts of CNC machine tools based on multilevel information fusion is designed. The paper starts with the analysis of the cutting force, and draws the characteristics of the variable load, the working condition of the high frequency punching and the diversity of the machining. The conclusion that the mechanical parts of the cutting force propagation path are more prone to failure is analyzed. The causes of the failure of the mechanical parts and the common equipment on the CNC machine tools are analyzed. The research objects and the failure modes of the paper are determined. The fault diagnosis mechanism based on the vibration, temperature, motor current and servo error is studied. According to the different sources of the parameter information, a multi-dimensional sensing state monitoring system of CNC machine tools, consisting of external sensors, internal information, program parameters and alarm information, has been constructed. From the machine tool body, tool wear, processing process, workpiece processing quality and many other angles, many sides reflect the running state of CNC machine tools and realize the whole CNC machine tools. It provides sufficient information for the follow-up diagnosis process. Through the analysis of experimental data, it is found that only the traditional time domain and spectrum analysis can not be effectively divided into the complex faults. In order to further excavate the fault features hidden in the original signal, this paper proposes a signal processing method combining the wavelet packet and the empirical mode decomposition. The wavelet packet is used to denoise the signal, and the wavelet reconstruction signal is extended to two narrow band signals of high frequency and low frequency, and then two narrow band signals are processed by EMD. The signal processing method combining the wavelet packet and the empirical mode decomposition is used to reduce the noise by the wavelet packet, which greatly improves the precision and quality of the EMD decomposition. In addition, the fault information can be analyzed more carefully by the expansion of the node. The energy of each IMF after EMD decomposition is extracted as the feature, and the multi domain mixed feature set is formed with the time domain features and frequency domain features. The feature selection method based on the correlation analysis between features is used as the main means to reduce the feature dimension and obtain the sensitivity. Feature subset. According to the characteristics that the CNC machine needs to diagnose and the characteristics of many faults, the strategy of grading diagnosis is put forward, and the diagnosis is divided into two levels of fault location, fault category and degree. The main network is responsible for the aggregation of the results of local subnetwork model while the fault location is located. Degree. Through task division and cooperation, the purpose of simplifying the network structure is achieved. The BP neural network model, the RBF neural network model and the support vector machine (SVM) model for CNC machine tool fault diagnosis are studied. The fault diagnosis of CNC machine tools based on BP, RBF and SVM is constructed by using the sensitive features as the model input. The model of the main network and the local diagnostic subnetwork is used to compare the diagnosis ability of the three models to different fault categories. The global diagnosis model based on fuzzy comprehensive evaluation and the global diagnosis model based on the weighted D-S evidence theory are established, and the fault recognition rate is further improved. First, the fuzzy comprehensive evaluation modeling of CNC machine tools is established. On the basis of the preliminary diagnosis results of multiple classifiers, the evaluation factors are transformed from high dimensional features to low dimension primary diagnosis results, and the complexity of the model is reduced. A single level fuzzy comprehensive evaluation model for numerical control machine tool based on information fusion is successfully constructed, and the correctness and diagnostic accuracy are two. To evaluate the classification ability of the classifier, the evaluation function based on information entropy and the weight allocation method of the overall average of the classifier are constructed, and the influence of the subjective factors is reduced. The classification ability evaluation matrix is constructed, which effectively solves the weight allocation problem when the classifier differs greatly from the recognition rate of different fault types. When the evidence theory synthesis rule is dealing with the high conflict evidence, the conclusion is contrary to the fact. The weighted evidence theory diagnosis model is put forward, which is weighted to the original evidence with the classifier fault recognition rate as the weight, thus effectively reducing the evidence conflict rate, so the recognition rate of the barrier is improved. The fault of the CNC machine tool is built up. The diagnostic experiment system is tested by the model and method proposed in this paper.

【学位授予单位】：青岛理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TG659

【参考文献】