基于增量学习的数控机床故障诊断系统

针对数控机床中主轴轴承和刀具同时出现故障或机床主轴转速改变时的故障诊断问题,提出了基于增量学习的深度卷积诊断模型。首先,将常用转速下的主轴轴承和刀具振动数据集,输入结合了批量归一化算法的一维卷积神经网络,实现单一转速下故障诊断;然后,人工判断跨转速诊断时的未知故障类型,对其打标签后重新输入网络,通过增量学习实现知识迁移并使模型学习新数据特征;最后模型在跨转速故障诊断领域的准确率为76.49%~86.09%,且与Fine Tuning和Joint Training两种经典跨领域算法相比,基于增量学习的深度卷积诊断模型提高了准确率,缩短了训练用时。

Fault Diagnosis System of Numerical Control Machine Based on Incremental Learning

In order to solve the problem of fault diagnosis when the spindle bearing and tool in numerical control machine malfunction at the same time or working conditions of numerical control machine are changed, a fault diagnosis model named deep convolutional neural network based on incremental learning was proposed. First, the vibration data sets of spindle bearings and tools at common speeds were input into a one-dimensional convolutional neural network, which combined a batch normalization algorithm. Secondly, manually judged the unknown fault type during cross-speed diagnosis, tagged it and re-entered the network. Incremental learning was used to retain old knowledge and learn the characteristics of new data to improve model performance. The fault diagnosis accuracy rate of the model at different speeds is between 76.49% and 86.09%. Compared with the two classic cross-domain algorithms of fine tuning and joint training, deep convolutional neural network based on incremental learning improves accuracy and shortens training time.