基于样本扩充与IDANN的刀具状态识别方法

针对机床刀具磨损数据稀少与刀具磨损状态识别精度低的问题,提出了一种基于样本扩充与改进领域对抗网络(sample expansion and improved domain adversarial training of neural networks, SE-IDANN)的刀具状态识别方法。首先对机床刀具数据进行两次特征提取,并通过Smote算法进行样本扩充,解决机床刀具磨损数据量稀少的问题;其次在领域对抗网络(domain adversarial training of neural networks, DANN)模型特征提取器中加入残差块,进一步提取有效特征信息,解决刀具磨损特征微弱的难题;最后将Wasserstein距离作为目标域与源域的数据分布相似度标准引入DANN模型,实现对刀具磨损量的精确识别。通过对机床刀具数据的分析与仿真试验验证,证明该方法能够有效地识别刀具磨损量。

Tool status recognition method based on sample expansion and IDANN

To deal with the problems of scarce data of machine tool wear and low recognition accuracy of tool wear status, a tool status recognition method based on sample expansion and improved domain adversarial training of neural networks (SE-IDANN) was proposed. First, to solve the problem of scarce machine tool wear data, two feature extractions on the machine tool data were performed, and the sample was expanded through the Smooth algorithm. Secondly, a residual block was added to the domain adversarial training of neural networks (DANN) feature extractor to further extract effective feature information and solve the problem of weak tool wear characteristics. Finally, to realize the accurate identification of tool wear, the Wasserstein distance used as the data distribution similarity standard between the target domain and the source domain was introduced into the DANN model. Through the analysis and test verification of machine tool data, it is proved that this method can better identify tool wear.