基于神经网络的金属板材折弯回弹预测与研究

发布时间：2017-12-27 07:15

本文关键词：基于神经网络的金属板材折弯回弹预测与研究　出处：《上海应用技术学院》2015年硕士论文　论文类型：学位论文

【摘要】：在弯曲成形过程中必然存在着回弹问题,回弹的存在极大的影响着零件的加工精度和生产效率,因此对回弹进行预测和研究有着重要的意义。工程实际中,一般依靠操作者的经验和多次加工、反复试验来减小或消除回弹对精度产生的影响。20世纪末以来,回弹的预测与研究逐渐成为板料成形领域的热点和重点问题。将人工智能技术和方法应用于金属板材折弯回弹的预测和研究当中是板料成形领域的一个热点问题。人工神经网络是一种模拟人脑脑神经处理信息的方式而建立起来的一种数学运算模型,它模拟人脑对信息的传递和人脑对环境的学习能力,通过多次的迭代反复的修改神经元之间的连接权值和阈值,实现输入神经元到输出神经元的复杂的非线性映射关系。由于网络具有很强的自适应、自学习能力和泛化能力,它常被用来替代许多计算过程复杂、耗时特别长的传统算法；基于上述情况,本研究采用神经网络对折弯回弹建立模型,分别用不同的激活函数和训练函数训练网络,对折弯回弹半径进行预测；然后分别采用遗传算法(GA)和粒子群算法(PSO)优化BP神经网络的权值和阈值后对折弯回弹进行预测,对三种预测结果进行分析比较,得出对于折弯回弹的预测,用粒子群算法优化后的BP神经网络在预测精度和误差方面更优；最后通过实验论证证明本文所建立的模型的准确性和可靠性。
[Abstract]:In the process of bending, springback is inevitable. The existence of springback greatly affects the machining accuracy and production efficiency of parts. Therefore, prediction and research of springback is of great significance. In engineering practice, the influence of resilience on precision is usually reduced or eliminated by the operator's experience and repeated processing and repeated tests. Since the end of the twentieth Century, the prediction and research of springback has gradually become a hot and key issue in the field of sheet metal forming. It is a hot issue in the field of sheet metal forming that artificial intelligence technology and methods are applied to the prediction and study of sheet metal bending and springback. Artificial neural network is a simulation of human brain neural information processing method is a mathematical model established, it simulates the human brain to the information transmission and the learning ability of the human brain, through repeated modifications between iterative neuron weights and threshold value, the realization of the input neurons to the output neurons of the complex nonlinear mapping relationship. Because the network is adaptive, strong self-learning ability and generalization ability, it is often used to replace many complicated calculation process, especially the long time-consuming traditional algorithm; based on the above, the neural network model is established in this study by bending springback, with different activation function and the training function training network, to predict the bending springback then using radius; genetic algorithm (GA) and particle swarm optimization (PSO) algorithm BP neural network weights and threshold of bending springback prediction, the prediction results of three kinds of analysis and comparison, the prediction for bending springback, with BP neural network and particle swarm optimization in prediction accuracy and error better; experiments demonstrate that the accuracy and reliability of the model.
【学位授予单位】：上海应用技术学院
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TG306;TP183

【相似文献】