基于GA-WPT-ELM的6061铝合金表面粗糙度预测

为了提高工件表面粗糙度预测的准确性，针对振动信号特征识别和表面粗糙度预测建模时多个参数难以同步优化和人工经验调优误差较大的问题，提出基于遗传算法（GA）的信号特征识别和表面粗糙度预测的优化算法. 对采集的6061铝合金铣削振动信号进行小波包变换（WPT）和多个特征提取，利用GA优化WPT母小波和特征向量；将信号特征向量和表面粗糙度分别作为极限学习机（ELM）的输入和输出，对预测模型训练的同时，利用GA优化ELM隐含层的神经元个数；对训练好的预测模型进行测试. 实验结果表明，通过GA对振动信号识别和表面粗糙度预测的3类参数同步优化，获得了最佳的信号特征和较高的表面粗糙度预测精度，节省了建模分析计算成本.

Surface roughness prediction of 6061 aluminum alloy based on GA-WPT-ELM

The problem of multi-parameter simultaneous optimization and large error of empirical tuning should be solved in the vibration signal feature recognition and surface roughness prediction processes in order to improve the on-line prediction accuracy of workpiece surface roughness. The optimization method of signal feature recognition and surface roughness prediction modeling was proposed based on genetic algorithm (GA). GA was used to choose mother wavelet and feature quantities when signal features of milling 6061 aluminum alloy was recognized based on wavelet packet transform (WPT). The number of neurons in hidden layer was selected by GA when signal features and surface roughness were used to train extreme learning machine (ELM). The prediction modeling which has been trained was tested. The experimental results show that the three types of parameters for signal recognition and surface roughness prediction were simultaneously optimized by GA. GA-WPT-ELM not only obtains the best features of signal and the higher prediction accuracy of surface roughness, but also reduces analytical-computational cost in the modeling processes.