电加工与模具
電加工與模具
전가공여모구
ELECTROMACHINING & MOULD
2014年
1期
36-40
,共5页
杨光美%张云鹏%李铠月%闫妍
楊光美%張雲鵬%李鎧月%閆妍
양광미%장운붕%리개월%염연
超声振动磨削放电加工%BP神经网络%支持向量机%模型预测
超聲振動磨削放電加工%BP神經網絡%支持嚮量機%模型預測
초성진동마삭방전가공%BP신경망락%지지향량궤%모형예측
ultrasonic vibration grinding assisted EDM%BP neural network%support vector machines (SVM)%model prediction
超声振动磨削放电加工过程复杂,难以用精确的理论公式进行描述,通常在试验基础上,借助于机器学习理论做出分析。针对实际加工中试验样本数量有限、预测量数值变化波动大的情况,采用BP神经网络和支持向量机两种方法分别建立超声振动磨削放电加工SiCp/Al指标预测模型,并利用两个模型预测零件表面粗糙度和加工速度等工艺指标。预测结果表明,零件表面粗糙度的数值变化范围较小,两种模型预测值与试验值均具有较好的一致性,预测精度较高;加工速度的数值变化较大,支持向量机模型的预测精度优于BP模型。因此,支持向量机模型更适合于解决小样本及指标变化范围大的预测问题。
超聲振動磨削放電加工過程複雜,難以用精確的理論公式進行描述,通常在試驗基礎上,藉助于機器學習理論做齣分析。針對實際加工中試驗樣本數量有限、預測量數值變化波動大的情況,採用BP神經網絡和支持嚮量機兩種方法分彆建立超聲振動磨削放電加工SiCp/Al指標預測模型,併利用兩箇模型預測零件錶麵粗糙度和加工速度等工藝指標。預測結果錶明,零件錶麵粗糙度的數值變化範圍較小,兩種模型預測值與試驗值均具有較好的一緻性,預測精度較高;加工速度的數值變化較大,支持嚮量機模型的預測精度優于BP模型。因此,支持嚮量機模型更適閤于解決小樣本及指標變化範圍大的預測問題。
초성진동마삭방전가공과정복잡,난이용정학적이론공식진행묘술,통상재시험기출상,차조우궤기학습이론주출분석。침대실제가공중시험양본수량유한、예측량수치변화파동대적정황,채용BP신경망락화지지향량궤량충방법분별건립초성진동마삭방전가공SiCp/Al지표예측모형,병이용량개모형예측령건표면조조도화가공속도등공예지표。예측결과표명,령건표면조조도적수치변화범위교소,량충모형예측치여시험치균구유교호적일치성,예측정도교고;가공속도적수치변화교대,지지향량궤모형적예측정도우우BP모형。인차,지지향량궤모형경괄합우해결소양본급지표변화범위대적예측문제。
The ultrasonic vibration grinding assisted EDM is so complex that it is difficult to describe the machining process with precise theoretical formula. Usually,based on the experiments,the process is analyzed with the machine learning theory. In the actual situation of few research samples but numerically fluctuated prediction,this paper has built a processing predictive model of ultrasonic vibration grinding assisted EDM SiCp/Al based on both BP neutral network and support vector machines partly,and has used each model to predict process indices including surface roughness of workpiece and machining. The result shows that the predictive values are consistent with the test results based on the two models as the surface roughness is just in a small change range,and the predictive precision is higher. The predictive precision of SVM is preferable to the result of BP as machining rate change is bigger. Therefore,the predictive model based on SVM is more suitable to solve the problems with smaller samples and larger change range of indices.