基于云理论的刀具磨损状态监测与磨损量预测理论研究

发布时间：2018-05-24 23:29

本文选题：刀具磨损 + 声发射信号　；参考：《东北电力大学》2017年硕士论文

【摘要】：随着装备制造业的发展,刀具磨损状态监测技术已成为制约现代自动化机床的一项关键技术,该技术目前尚未得到有效解决。实时地监测刀具状态,可提高零件加工质量和机床的加工效率,减少机床事故的发生,最大限度地减少人对机床的操作,实现机床的智能化和无人化,保证系统在最优参数下运行。因此,刀具磨损状态监测技术的研究是非常迫切且重要的。本文针对不同切削条件下刀具磨损状态监测和磨损量预测研究课题,通过正交试验法安排切削试验,在采集的声发射信号的基础上,应用现代信号处理方法小波包分析和最优熵理论相结合实现信号的滤波处理,提出了基于云模型理论和最小二乘支持向量机的刀具磨损状态识别方法,最后应用不确定性云推理方法实现磨损量的不确定性预测。主要研究内容由以下几个部分构成:以往的刀具磨损监测信号滤波采用时域分析(经验模态分解)、频域分析(功率谱分析)等传统的信号预处理方法。由于所采用的声发射信号的非平稳和非线性特点,本文将适合处理非平稳信号处理的小波包分析方法引入到信号预处理中,实现信号的滤波。首先通过频谱分析得到不同磨损阶段声发射信号的频带分布范围,作为小波包分解层次的定性参考;其次应用信息熵理论中的Shannon熵表征噪声的大小,确定小波包分解最佳树;最后通过最佳树统计分析确定小波包分解的最优分枝,并通过阈值处理后进行信号重构,信噪比可达35dB以上。提出了基于云理论的不确定性声发射信号特征提取方法。首先通过改进的逆向云算法提取不同磨损量声发射信号的特征参数,期望、熵和超熵;其次,定量分析刀具在不同切削条件下三种云特征参数随磨损量增大所呈现的变化趋势和规律;最后,通过散点图验证三种特征参数表征刀具磨损声发射信号的有效性。提出了将云特征参数与最小二乘支持向量机相结合的刀具磨损状态识别方法。针对神经网络学习算法收敛速度慢、易陷入局部极值以及对特征要求较高等问题,提出基于云理论与最小二乘支持向量机结合刀具磨损状态识别方法。实例分析表明,在优化选取支持向量机参数的条件下,云-支持向量机结合的方法比传统神经网络识别方法的识别率更高。将不确定性云推理模型应用到刀具磨损量预测领域。首先,通过条件云发生器挖掘不同磨损阶段磨损趋势与该阶段云特征参数数据之间的关系;其次,在此基础上构建云预测规则;最后,建立了多条件单规则不确定性磨损量预测方法。实例分析结果显示,云推理磨损量预测结果符合刀具磨损规律;对非确定模型进行预测,云推理比模糊推理更接近实际情况。此外,该方法能可推广到不同工况条件下的磨损量预测,具有较强的实用性。
[Abstract]:With the development of equipment manufacturing industry, tool wear monitoring technology has become a key technology restricting modern automatic machine tools, which has not been effectively solved. Monitoring the cutting tool condition in real time can improve the machining quality and efficiency of the machine tool, reduce the accident of the machine tool, reduce the operation of the machine tool to the maximum extent, and realize the intellectualization and inhumanity of the machine tool. Make sure the system runs under the optimal parameters. Therefore, the research of tool wear monitoring technology is very urgent and important. In this paper, aiming at the research of tool wear condition monitoring and wear quantity prediction under different cutting conditions, cutting test is arranged by orthogonal test method, and on the basis of the collected acoustic emission signal, A new method of tool wear recognition based on cloud model theory and least square support vector machine (LS-SVM) is proposed, which combines wavelet packet analysis and optimal entropy theory. Finally, uncertainty cloud reasoning method is used to predict the uncertainty of wear quantity. The main research contents are as follows: the traditional signal preprocessing methods such as empirical mode decomposition (EMD) and frequency domain analysis (power spectrum analysis) are used in the previous tool wear monitoring signal filtering. Because of the nonstationary and nonlinear characteristics of the acoustic emission signal, the wavelet packet analysis method, which is suitable for the processing of the non-stationary signal, is introduced to the signal preprocessing to realize the signal filtering. First, the frequency band distribution range of acoustic emission signals at different wear stages is obtained by spectrum analysis, which can be used as the qualitative reference of wavelet packet decomposition level. Secondly, the Shannon entropy of information entropy theory is used to characterize the size of noise, and the best tree of wavelet packet decomposition is determined. Finally, the optimal branch of wavelet packet decomposition is determined by the statistical analysis of the best tree, and the signal to noise ratio (SNR) is higher than 35dB after threshold processing. A method for feature extraction of uncertain acoustic emission signals based on cloud theory is proposed. Firstly, the characteristic parameters, expectation, entropy and superentropy of acoustic emission signals with different wear quantities are extracted by the improved reverse cloud algorithm. Quantitative analysis of the change trend and law of the three cloud characteristic parameters with the increase of wear quantity under different cutting conditions. Finally, the validity of the three characteristic parameters to characterize the acoustic emission signal of tool wear is verified by scatter plot. A tool wear recognition method combining cloud feature parameters with least squares support vector machine (LS-SVM) is proposed. Aiming at the problems of slow convergence, easy to fall into local extremum and high demand for features of neural network learning algorithm, a tool wear recognition method based on cloud theory and least squares support vector machine (LS-SVM) is proposed. The analysis of examples shows that the cloud-support vector machine method has a higher recognition rate than the traditional neural network method under the condition of optimizing the selection of support vector machine parameters. The uncertain cloud reasoning model is applied to the field of tool wear prediction. Firstly, the relationship between wear trend of different wear stages and cloud characteristic parameter data is mined by conditional cloud generator. Secondly, cloud prediction rules are constructed on this basis. Finally, A prediction method for uncertain wear volume with multiple conditions and single rules is established. The result of case analysis shows that the prediction results of cloud reasoning wear quantity accord with the law of tool wear, and the cloud reasoning is closer to the actual situation than fuzzy reasoning to predict the uncertain model. In addition, the method can be extended to predict the wear quantity under different working conditions, and it has strong practicability.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG71

【参考文献】