基于GA-BP神经网络的钣金折弯过程随动预测模型

Prediction Model of Bending Follow-up Based on GA-BP Neural Network

提出一种利用GA-BP神经网络预测钣金折弯过程中折弯机器人机械臂末端夹持点轨迹模型的方法。通过钣金折弯实验与ANSYS仿真对比,得到最佳的仿真条件;使用参数化建模,针对不同的影响因子如钣金材料的弹性模量、屈服强度、强化系数和钣金件及上下模几何形状等,进行批量化仿真获得模型训练数据;通过对BP神经网络训练建立折弯随动模型,并引入遗传算法对BP神经网络的初始权重和阈值进行优化,解决BP神经网络存在的过拟合和局部最优问题;通过测试集进行验证评估,结果表明该折弯随动模型相对误差在0.4%以内,可满足常规折弯工艺的应用需求。

A method for predicting the trajectory model of the end clamping point of bending robot manipulator in the process of sheet metal bending by using GA-BP neural network is proposed.Through the comparison between sheet metal bending experiment and ANSYS simulation,the best simulation conditions are obtained.For different influence factors such as elastic modulus,yield strength,strengthening coefficient of sheet metal materials,geometry of sheet metal parts and upper and lower dies,parametric modeling is applied to conduct batch simulation so as to gain the model training data.The bending follow-up model is established by training BP neural network,and the genetic algorithm is introduced to optimize the initial weight and threshold of BP neural network to solve the over fitting and local optimization of BP neural network.By test set,verification evaluation is carried out,whose result shows that the relative error of the bending follow-up model is less than 0.4%,which meets the application requirements of conventional bending process.