基于进化神经网络的电火花铣削加工电极损耗预测

TOOL WEAR PREDICTION IN ELECTRICAL DISCHARGE MILLING MACHINING BASED ON EVOLUTIONARY NEURAL NETWORK

针对电火花铣削加工的时变非线性特性,提出基于神经网络的电火花铣削加工电极损耗预测模型,利用该网络预测加工速度和工具的相对损耗,从而可在加工中实时计算出工具实际损耗量,为实现电极损耗的在线动态补偿打下基础。针对神经网络传统训练算法-BP算法的不足,提出了一种自适应调节变异率和变异量的进化算法来优化网络权值和网络结构,提高了网络的逼近精度和进化速度。

In according with the non-linear character in the process of EDMM, a tool wear prediction model is established based on artificial neural network. The tool relative wear and machining rate can be predicted by the network, which makes foundation for tool's dynamic compensation in EDMM. An improved evolutionary algorithm which could adaptively adjust mutation rate and magnitude of mutation is presented to optimize the neural network's weights and topology in case of local extremums obtained from traditional BP algorithm.