【作者】 郑建明；

【导师】 黄玉美； 李言；

【作者基本信息】 西安理工大学 ， 机械工程， 2004， 博士

【摘要】 制造装备自动化与制造模式集成化的发展，使得刀具成为制造系统薄弱的一环，因此，研究和开发加工过程刀具状态监控技术对实现加工过程自动化、提升制造业水平具有重要意义。论文以钻削过程为研究对象，建立了以钻削力作为监测信号的钻头磨损监测实验系统，对钻削力信号的特征提取以及基于HMM的多特征融合钻头磨损监测技术进行了系统的理论与实验研究。 分别从时域和频域的角度出发，研究了钻削过程中钻削力信号及其特征值与钻头磨损之间的关系，发现时域统计特征与钻头磨损之间相关性较差，而在频域，钻削力信号功率谱密度与钻头磨损之间则表现出较强的相关性，频域统计特征呈现出与钻头磨损规律基本一致的变化趋势。 讨论了小波变换的多分辨特性以及Mallat算法的基本原理，采用Daubechies5正交小波对钻削力时域和功率谱信号进行了多分辨分析，研究了小波分解信号及其统计特征的变化规律，在钻削力信号低频小波分解系数与功率谱重构包络中获得了与钻头磨损相关性较强的特征参数，为实现钻头磨损状态的监测奠定了基础。 成功地把分形理论引入到钻头状态监测领域，研究了离散时序信号盒维数计算过程中的网格建立及时间序列重构中确定延迟时间和嵌入维数的方法，给出了离散时序信号三种分维数的具体实现算法，研究了钻削过程钻削力时域信号和各频段西安理工大学博士学位论文小波重构信号分维数的变化规律，结果表明:随着钻头磨损量的增加，钻削力信号的盒维数、信息维数和关联维数均呈现出明显的下降趋势，利用这一特征可有效实现钻头磨损状态的监测。 针对各钻头磨损状态下特征模式高度重叠的特点，提出采用隐马尔可夫模型HMM解决钻头磨损监测的思想，对1肠IM在钻头磨损监测中的原理及实现方法进行了系统的研究，完成了基于前后向算法和Baurn一Welch算法进行模型训练和识别的具体实施步骤，建立了基于S服网络的矢量量化器，解决了钻削力特征矢量的融合与编码问题。 提出了两种基于1侧颐的钻头磨损监测方法，利用时域、频域、小波及分形方法所提取的有效特征进行了多特征融合钻头磨损监测，实验结果表明:利用未知观察序列在单一1翎服下的概率输出值，可反映不同钻头磨损状态下观察序列的统计相似性，跟踪钻头磨损的发展趋势;而利用多模型方法成功实现了三种典型钻头磨损状态的识别，为加工过程刀具状态开辟了新的途径。关键词:钻头磨损监测;小波变换;分形;特征提取;隐马尔可夫模型(HMM)。 论文的研究得到了原机械工业部发展基金项目《深孔加工过程多传感融合监控技术研究))(项目编号:97JFOo13)的资助。更多还原

【Abstract】 The development of manufacturing equipment automation and manufacturing mode integration makes tools become a weak link in manufacturing system. Accordingly the research and development of the tool condition monitoring technology in machining process is of great significance in realizing automation of machining process and upgrading the manufacturing level. With the drilling process as the research objective, the drill wear monitoring experimental system with the drilling force as the monitoring signal is established in this thesis to carry out the systematically theoretical and experimental researches on the feature extraction of drilling force signal as well as the technology of the multi-feature fusion drill wear monitoring based on HMM.Starting from the angles of the time and frequency domain respectively, the relationships between the drilling force signal as well as its feature value and drill wear in drilling process are investigated in this thesis, and it has been found that there is a rather poor correlation between the statistic feature and drill wears in the time domain while in the frequency domain, the power spectrums of drilling force signal showed good correlation with drill wear, and the statistic features appear to have the changing trend basically agreeable with the drill wear laws.The multi-resolution performances of wavelet transform as well as the basis principle of Mallat algorithm are discussed. The Daubechies5 orthogonal wavelet is adopted to carry out the multi-resolution analysis to the drilling force signals in the time and frequency domain. The changing laws of the wavelet decomposed signals and its statistic features are also studied so that the statistical features that are correspond to drill wears are obtained from the low frequency wavelet decomposed coefficient and the reconstructed power spectrum envelope of the drilling force signals, and this lay a solid foundation for the realization of drill wear condition monitoring.The fractal theory has been successfully introduced into the field of drill conditionmonitoring. Also the grid establishment in the process of calculation the box dimension of the discrete time series signals as well as the method for determining the delay time and implanted dimension in time series reconstruction are studied. The algorithms for calculating the three kinds of fractal dimension of the discrete time series signals are given in this thesis. The changing laws of fractal dimension of the signals in the time domain and the wavelet reconstruction signals in each frequency band are studied in drilling process. The results show that with the increase of drill wears, the box dimension and the information dimension as well as the correlation dimension of drilling force signals appear obviously to have the downtrends. Accordingly, this characteristic can be effectively used to realize the monitoring of drill wears.Aiming at the characteristics of the highest overlap of the feature modes in each drill wear condition, the hidden Markov model is put forward to resolve the drill wear monitoring. The systematic studies on the principle and realization method of HMM used in drill wear monitoring are carried out. The implementing procedures for model training and identification based on forward and backward algorithm and Baum-Welch algorithm are completed. The vector quantization system based on SOM network has been formed to resolve the problem of the fusion and coding of the drilling force feature vectors.Two kinds of methods for drill wear monitoring based on HMM are put forward. The effective features obtained from the time and frequency domain, wavelet transform and fractal are used to carry out drill wear monitoring of multi-feature fusion. The experimental results indicate that probabilities of unknown observation series can reflect the statistic similarities of observation series in different drill wear status and track the development trend of drill wear. Moreover using the multi-model method has successfully realize the recognition of three typical drill wea更多还原