基于IMB-CNN的薄壁件超声铣削颤振辨识方法

针对薄壁件超声铣削加工时产生的颤振严重影响工件质量,加剧刀具磨损的问题,搭建了颤振图像监测系统,利用卷积神经网络(CNN)进行颤振图像辨识,综合运用趋磁细菌算法(MB)、爬山算法(HC)和禁忌算法(TS)的优点,改进MB算法进行超参数优化,提出了一种基于改进趋磁细菌卷积神经网络(IMB-CNN)的薄壁件超声铣削颤振辨识方法。首先,通过MB算法进行全局搜索,再以最优解为初始点,通过HC算法进行邻域搜索,避免了MB算法在最优解附近的振荡;同时,通过禁忌列表跳过已搜索的节点,减小计算规模,加快计算效率;最后,将获得的最优超参数用于CNN,实现颤振图像的精确辨识。与其他方法相比,该方法实现了97.69%的识别率,判断时间为363ms,能有效地进行颤振监测,且整体性能较优。

Chatter Identification Method for Ultrasonic Milling of Thin Walled Parts Based on IMB-CNN

Chatter in ultrasonic milling of thin-walled parts seriously affects the quality of workpiece and aggravates tool wear, so a chatter image monitoring system was built. Convolutional neural network (CNN) was used to identify chatter images and the advantages of magnetotactic bacteria algorithm (MB), hill climbing algorithm (HC) and tabu search algorithm (TS) were taken synthetically to improve MB algorithm for optimizing the parameters. Therefore, a chatter identification method based on the improved magnetic bacteria convolution neural network (IMB-CNN) for ultrasonic milling of thin-walled parts was proposed. Firstly, the global search was carried out by MB algorithm, and then neighborhood search was carried out by HC algorithm with the optimal solution as the initial point, so as to avoid the oscillation of MB algorithm near the optimal solution. At the same time, the tabu list was used to skip the searched nodes to reduce the calculation scale and speed up the calculation efficiency. Finally, the optimal hyperparameters were applied to the CNN to realize the accurate identification of flutter images. Compared with other methods, this method achieves 97.69% recognition rate, and the judgment time is 363ms. The chatter is identified effectively, and the overall performance is better than other algorithms.