机械工程学报
機械工程學報
궤계공정학보
CHINESE JOURNAL OF MECHANICAL ENGINEERING
2015年
1期
203-212
,共10页
薄壁件%装夹变形%神经网络%遗传算法
薄壁件%裝夾變形%神經網絡%遺傳算法
박벽건%장협변형%신경망락%유전산법
thin-walled workpiece%fixturing deformation%neural network%genetic algorithm
在多重装夹元件装夹过程中,由于装夹顺序、夹紧力、定位元件位置等装夹布局参数的不同,薄壁件的装夹变形程度也不一样。单个装夹布局参数引起的工件装夹变形规律能够通过有限元方法获得。但是,若同时考虑多个装夹布局参数的影响,仅仅利用有限元方法难以揭示装夹布局参数与装夹变形之间的关系。为此,针对薄壁件的装夹布局方案建立三维有限元模型,以便利用有限元法获取神经网络的训练样本。借助神经网络的非线性映射能力,通过有限的训练样本构建装夹变形的预测模型。以减小工件的最大装夹变形为目标,并根据每一代装夹布局中工件的最大装夹变形定义个体的适应度,建立装夹布局方案的优化模型及其遗传算法求解技术。试验结果表明,网络预测值与相应的有限元仿真值、试验数据之间的相对误差均不超过3%。提出的基于神经网络与遗传算法的装夹变形“分析-预测-控制”方法,不仅能够提高装夹变形的计算效率,而且为薄壁件装夹布局方案的合理设计提供基础理论。
在多重裝夾元件裝夾過程中,由于裝夾順序、夾緊力、定位元件位置等裝夾佈跼參數的不同,薄壁件的裝夾變形程度也不一樣。單箇裝夾佈跼參數引起的工件裝夾變形規律能夠通過有限元方法穫得。但是,若同時攷慮多箇裝夾佈跼參數的影響,僅僅利用有限元方法難以揭示裝夾佈跼參數與裝夾變形之間的關繫。為此,針對薄壁件的裝夾佈跼方案建立三維有限元模型,以便利用有限元法穫取神經網絡的訓練樣本。藉助神經網絡的非線性映射能力,通過有限的訓練樣本構建裝夾變形的預測模型。以減小工件的最大裝夾變形為目標,併根據每一代裝夾佈跼中工件的最大裝夾變形定義箇體的適應度,建立裝夾佈跼方案的優化模型及其遺傳算法求解技術。試驗結果錶明,網絡預測值與相應的有限元倣真值、試驗數據之間的相對誤差均不超過3%。提齣的基于神經網絡與遺傳算法的裝夾變形“分析-預測-控製”方法,不僅能夠提高裝夾變形的計算效率,而且為薄壁件裝夾佈跼方案的閤理設計提供基礎理論。
재다중장협원건장협과정중,유우장협순서、협긴력、정위원건위치등장협포국삼수적불동,박벽건적장협변형정도야불일양。단개장협포국삼수인기적공건장협변형규률능구통과유한원방법획득。단시,약동시고필다개장협포국삼수적영향,부부이용유한원방법난이게시장협포국삼수여장협변형지간적관계。위차,침대박벽건적장협포국방안건립삼유유한원모형,이편이용유한원법획취신경망락적훈련양본。차조신경망락적비선성영사능력,통과유한적훈련양본구건장협변형적예측모형。이감소공건적최대장협변형위목표,병근거매일대장협포국중공건적최대장협변형정의개체적괄응도,건립장협포국방안적우화모형급기유전산법구해기술。시험결과표명,망락예측치여상응적유한원방진치、시험수거지간적상대오차균불초과3%。제출적기우신경망락여유전산법적장협변형“분석-예측-공제”방법,불부능구제고장협변형적계산효솔,이차위박벽건장협포국방안적합리설계제공기출이론。
In the fixturing processing with the multiple clamps, the different fixturing parameters, including the fixturing sequence, the magnitude of clamping force, the locator position, and so forth, can cause the different fixturing deformation of the thin-walled workpiece. Deformation law of workpiece which is caused by a single fixturing parameter can be obtained by finite element method. However, if multiple fixturing parameters are synchronously considered, finite element method is difficult in revealing the relationship between the fixturing parameters and fixturing deformation of workpiece. Therefore, the finite element model of the multi-fixturing layout is above all established for the thin-walled workpiece. Fixturing deformations can be analyzed to be the training samples of neural network. And then, with the nonlinear mapping of neural network, the prediction model of fixturing deformation is suggested according to the training samples. Finally, the optimal model of multi-fixturing layout with the objective of minimizing the minimum fixturing deformation is presented. According to the maximum fixturing deformation of each generation, the fitness of the individual is defined to develop the genetic algorithm so that the optimal model can be solved to obtain the fixturing sequence and the locator position. The prediction model is able to predict experimental results within a 3% error margin as well as predict simulated results within a 3% error margin. The presented “analysis-prediction-control” method of fixturing deformation can not only improve the calculation efficiency of fixturing deformation, but also provide a basic theory of multi-fixturing layout design for the thin-walled workpiece.