基于声发射的刀具磨损状态识别与预测

发布时间：2018-06-25 01:03

本文选题：刀具磨损 + 时域分析　；参考：《电子科技大学》2017年硕士论文

【摘要】：如今,从国际角度来看,制造业地位日益凸显,以智能制造为代表的科技变革,正在将全球制造业推倒重建,形成新的“工业互联网”世界,并成为国际竞争战略高地。在制造业都在向智能制造方向发展的同时,数控加工技术的智能化水平也得到迅速提高。在机械加工中,大部分的零件都是由切削加工生产得到的,刀具的使用是最直接最频繁的。实时准确获知刀具的磨损状态对提高加工产品精度和表面质量、实现个性化制造,提高机床智能化水平、提高系统误差补偿技术具有实际意义。本文的工作内容如下:1、研究了声发射信号的特点,确定了以声发射信号为监测信号的在线监测方案。搭建了刀具磨损试验系统,通过试验研究了切削过程中刀具磨损形式,确定了刀具磨损的磨钝标准,将刀具磨损划分为前期、中期、后期磨损三个阶段。基于正交试验研究了声发射信号随刀具磨损、主轴转速、进给量和背吃刀量四个因素的变化规律。2、用不同的信号处理方法提取刀具磨损的特征值。基于时域分析方法,提取了均值、均方根、方差和方根幅值;基于小波包变换的分析方法,提取频段的能量比作为刀具磨损的特征值;基于经验模态分析方法,提取特征模态函数的均方根作为刀具磨损的特征值。对提取的时域特征值和时频特征值进行选择和优化。3、将优化后的特征值输入到LS-SVM算法中进行学习,建立刀具磨损状态识别模型。LS-SVM算法中的惩罚因子c和核参数g对识别的准确率有很大的影响,将粒子群算法应用到LS-SVM算法中,在全局化与收敛速度方面具有较大优势,能够实现参数c和g的快速寻优。建立基于PSO-LS-SVM的刀具磨损状态识别模型和磨损量预测模型,建立未优化的LS-SVM和BP神经网络刀具磨损模型,用测试样本检测三个模型的预测效果,结果表明PSO-LS-SVM的准确率最高。最后,离线检测刀具磨损对已加工表面的残余应力和粗糙度,将离线检测的结果用于间接评估刀具磨损,同时验证了刀具磨损的预测结果。
[Abstract]:Now, from the international point of view, the status of manufacturing industry is increasingly prominent. The technological changes represented by intelligent manufacturing are pushing the global manufacturing industry down and rebuilding, forming a new "industrial Internet" world and becoming the strategic high ground of international competition. With the development of manufacturing industry towards intelligent manufacturing, the intelligent level of NC machining technology has been improved rapidly. In machining, most parts are produced by cutting, and the use of cutting tools is the most direct and frequent. It is of practical significance to know the wear state of cutting tools in real time and accurately for improving the precision and surface quality of machining products, realizing individualized manufacturing, improving the intelligent level of machine tools and improving the compensation technology of system errors. The work of this paper is as follows: 1. The characteristics of acoustic emission signal are studied and the on-line monitoring scheme with acoustic emission signal as monitoring signal is determined. The tool wear test system was set up, and the tool wear form in the cutting process was studied, and the grinding bluntness standard was determined. The tool wear was divided into three stages: early, middle and late wear stages. Based on orthogonal test, the variation of acoustic emission signal with tool wear, spindle speed, feed rate and feed rate was studied. Different signal processing methods were used to extract the characteristic value of tool wear. Based on time domain analysis method, the mean value, root mean square, variance and square root amplitude are extracted. Based on wavelet packet transform, the energy ratio of frequency band is extracted as the characteristic value of tool wear. The root mean square (RMS) of the eigenmode function is extracted as the eigenvalue of tool wear. The extracted time-domain eigenvalues and time-frequency eigenvalues are selected and optimized. The optimized eigenvalues are input into the LS-SVM algorithm for learning. The penalty factor c and kernel parameter g of tool wear recognition model. LS-SVM algorithm have great influence on the accuracy of recognition. The particle swarm optimization algorithm is applied to LS-SVM algorithm, which has a great advantage in globalizing and convergence speed. The parameters c and g can be optimized quickly. Based on PSO-LS-SVM, the tool wear recognition model and wear quantity prediction model are established, and the unoptimized LS-SVM and BP neural network tool wear models are established. The results show that the accuracy of PSO-LS-SVM is the highest. Finally, the residual stress and roughness of tool wear on machined surface are detected offline. The results of off-line testing are used to evaluate the tool wear indirectly, and the prediction results of tool wear are verified at the same time.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG71

【参考文献】