陶瓷材料电加工表面粗糙度的预测

Surface Roughness Prediction of Electrical Discharge Machining Ceramics Based on Evolutionary Neural Network

针对电加工工艺参数与性能指标的函数映射关系大多具有非线性的特征,提出了将BP神经网络引入电加工领域中。考虑到BP算法的不足,提出用遗传算法来优化BP神经网络的连接权值,设计了基于进化神经网络的学习算法,建立了陶瓷材料电加工表面粗糙度随工艺参数变化的预测模型。试验结果表明,该算法可以避免BP神经网络易陷入局部极小值等问题,预测精度高,相对误差在4%之内,进而验证了该模型的可靠性。

Owing to that the function mapping relationship between the technological parameters and performance index of wire electrical discharge machining （WEDM） has a non-linear characteristic, the artificial neural networks were incorporated into WEDM calculations. To compensate the disadvantage of the conventional back propagation algorithm （CBPA）, an improved learning algorithm, which trained a BP neural network by the genetic algorithm, was developed. A predictive model for surface roughness of ceramic by WEDM was developed based on the evolutionary neural networks （ENN）. The results show that the ENN can effectively overcome the problems of easily falling into local minimum point and of weak global search capability. The errors between the prediction values and the practical measured ones are less than 4%.