电火花线切割火花参数分析研究

发布时间：2019-02-16 04:54

【摘要】：电火花线切割加工原理是利用移动的细金属丝线(铜丝或钼丝)作电极,对工件与金属丝线进行脉冲电火花放电、切割成型。在电火花线切割加工中,各项加工参数相互关联、制约,甚至互相矛盾,耦合性极强,难以用数学公式描述其加工模型。电火花线切割主要加工阶段是脉冲电源接通时电极丝与工件之间产生电火花,但接通阶段往往出现断流现象,本文重点研究加工时火花面积量的变化规律,对所采集数据进行分析、训练及预测,预测系统可作为加工控制策略制定的专家系统。数据聚类分析,能够研究数据的相似程度以发现隐含的、未知的、有潜在应用价值的信息或模式,是作为数据处理的基础或分析的起点。机器学习中,传统数据按有无确定性可划分为监督学习和无监督学习；聚类分析是一个无监督的学习过程。半监督聚类是通过引入成对约束解决无监督聚类容易陷入局部最优值问题。本文采用矩阵成对约束优化K均值聚类算法,利用部分标记数据和属性矩阵提高聚类中心搜索能力。测试数据选用UCI机器学习数据集与现有聚类算法进行比较分析,结果表明本算法在低维度数据中聚合能力更优。聚类中心搜索目的是优化径向基(RBF)神经网络中心向量的选取,RBF神经网络对于复杂的函数关系或模型难以确定具有较高的逼近效果,理论上能以任意精度逼近连续函数。径向基神经网络学习缺陷在于中心向量选取困难,传统选取方式是以输出结果为导向,采用梯度下降法调整学习,输出模型具有较高的拟合精度,但预测精度不高。本文采用电火花线切割火花面积与加工影响参数建立神经网络二步预测学习模型,提出了半监督聚类优化RBF神经网络预测算法,利用数据半监督聚类分析系统参数影响程度,聚类结果优化径向基神经网络中心向量和宽度选择,对火花面积和加工效率进行预测学习。实验结果表明算法无论是拟合能力还是预测能力均优于传统神经网络,预测相对误差较小,整体预判火花量成功率较高。
[Abstract]:The principle of WEDM is to use moving thin wire (copper wire or molybdenum wire) as electrode to cut the workpiece and metal wire by pulse spark discharge. In WEDM, the machining parameters are interrelated, restricted, even contradictory, and the coupling is very strong, so it is difficult to describe the machining model by mathematical formula. The main stage of WEDM is the generation of EDM between electrode wire and workpiece when the pulse power supply is on, but the phenomenon of breakage often occurs in the turn-on stage. This paper focuses on the variation law of spark area during machining. The data collected are analyzed, trained and predicted. The prediction system can be used as an expert system for manufacturing control strategy. Data clustering analysis, which can study the similarity of data to find hidden, unknown and potentially valuable information or patterns, is the basis of data processing or the starting point of analysis. In machine learning, traditional data can be divided into supervised learning and unsupervised learning according to the determinacy, and clustering analysis is an unsupervised learning process. Semi-supervised clustering is to solve the problem that unsupervised clustering is prone to fall into local optimal value by introducing pairwise constraints. In this paper, the K-means clustering algorithm with matrix pairwise constraints is used to improve the ability of clustering center searching by using partial marking data and attribute matrix. The test data is compared with the existing clustering algorithms using UCI machine learning data set. The results show that the algorithm has better aggregation ability in low dimensional data. The purpose of clustering center search is to optimize the selection of center vector of radial basis function (RBF) neural network. RBF neural network has high approximation effect for complex function relation or model, and can approach continuous function with arbitrary precision theoretically. The learning defect of radial basis function neural network is that it is difficult to select the center vector. The traditional selection method is based on the output result and the gradient descent method is used to adjust the learning. The output model has high fitting accuracy but the prediction accuracy is not high. In this paper, the neural network two-step prediction learning model is established by using EDM spark area and machining parameters, and a semi-supervised clustering optimization RBF neural network prediction algorithm is proposed, and the influence degree of system parameters is analyzed by using data semi-supervised clustering. The selection of center vector and width of radial basis function neural network is optimized by clustering results, and the prediction of spark area and machining efficiency is carried out. The experimental results show that the algorithm is superior to the traditional neural network in both fitting ability and prediction ability, the relative error of prediction is small, and the success rate of predicting sparks is higher as a whole.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TG484

【相似文献】