基于双路并行卷积信息融合的刀具磨损识别

Tool wear identification based on dual-channel convolutional information fusion

针对机械加工现场环境复杂多变,使用单一信号进行刀具磨损识别难以获取全面的刀具磨损特征信息的问题,提出一种同时利用声音信号和工件表面图像信息结合深度学习网络识别刀具磨损状态的方法。首先采集铣削加工过程中声音信号和工件表面图像数据,然后使用双路并行卷积神经网络对声音信号和工件表面图像进行特征提取及融合,最后进行刀具磨损识别。结果表明,和单一信号识别结果相比,采用信息融合方法能获取更全面的刀具磨损特征信息,有利于增强刀具磨损识别效果,且刀具磨损识别准确率和F1-score均在95%以上,能有效识别刀具磨损状况。

During the machining process,it is difficult to obtain comprehensive information about tool wear characteristics using a single signal due to the complex and varied machining environment.A method is proposed that utilizes a combination of sound signals and surface image signals of workpieces through deep learning networks to recognize tool wear states.Initially,sound signals and surface image data from milling processes are synchronously collected,followed by the establishment of a dual-channel parallel convolutional neural network for feature extraction and fusion of one-dimensional sound signals and two-dimensional workpiece surface images.Finally,tool wear recognition is conducted.The results indicate that the use of information fusion methods can obtain more comprehensive tool wear characteristic information compared to single information recognition results,which is advantageous in improving the tool wear recognition effect.Moreover,the tool wear recognition accuracy and F1-score are above 95%,allowing for effective recognition of the tool wear condition.