基于机器学习的零件加工在线监测研究

Research on on-line monitoring of parts machining based on deep learning

针对数控机床执行零件加工时,因加工刀具磨损状态监测能力不足,刀具未能及时更换导致的零件加工质量偏低、合格品零件加工速率降低的问题,提出将极限学习机网络与自编码器相结合,构建ELM-AE模型,对零件加工的刀具磨损状态进行实时监测;为适应机床零件的实际加工情况,增加模型的实际监测效果,提出基于Coral距离构建刀具磨损状态特征迁移模型,对不同工况下的振动特征区域进行迁移,提高监测模型的实际应用性能;最后经过模型编码和振动信号采集设备选型,对模型进行性能验证实验。结果表明,设计的模型能够应对不同工况下的刀具磨损状态监测,在单一工况下,刀具磨损状态判断准确率最低为99.50%;用A工况进行训练用B工况进行测试,刀具磨损状态判断准确率最低为98.75%.综上所述,模型基本能够实现对零件加工过程中的刀具磨损状态的在线监测。

For when performing parts processing, CNC machine tools for machining tool wear condition monitoring ability is insufficient, the tool failed to replace the parts processing quality is low, product parts processing rate to reduce the problems, proposed to extreme learning machine network combined with the encoder, build the ELM-AE model, real-time monitoring the tool wear status of parts processing;In order to adapt to the actual machining situation of machine tool parts and increase the actual monitoring effect of the model, a feature migration model of tool wear state was proposed based on Coral distance to migrate the vibration characteristic regions under different working conditions and improve the practical application performance of the monitoring model. Finally, through model coding and selection of vibration signal acquisition equipment, the performance of the model is verified. The results show that the designed model can cope with the tool wear state monitoring under different working conditions, and the lowest accuracy of tool wear state judgment is 99.50% under a single working condition. The lowest judgment accuracy of tool wear state is 98.75% when A condition is used for training and B condition is used for testing. To sum up, the model can basically realize the on-line monitoring of tool wear state in the machining process of parts.