基于BP网络的机床主轴故障诊断研究

Research of Spindle Fault Diagnosis Based on BP Neural Network

针对机床主轴在运转过程中由于高速、热变形和应力集中等原因而发生平移、旋转、压缩和伸展等,提出了一种基于BP神经网络和误差标定拟合对机床主轴轴心轨迹误差预测的方法。该方法首先通过实验测量出机床主轴轴心轨迹的偏心数据形成样本,运用BP神经网络对样本进行训练,然后根据样本训练结果预测机床主轴轴心偏转的将来值,最后通过三维张量空间分布函数分析将来值与理论值拟合情况得出机床运转状态。实验结果显示,当迭代次数epoch=2,训练误差为Validation=0.0052442时,训练后的拟合曲线拟合效果较好,此时BP训练状态最佳,训练后的主轴偏转结果能够反映和预测机床运转状态。本方法对于生产过程中的机床定期维修保养具有重要的指导意义。

In the state of high speed,thermal deformation and stress concentration,machine tool spindle may produce translation,rotation,compression,and stretch,thus we proposed an error forecast method based Back-Propagation neural network and error calibration match.Firstly,we did the experiment to measure eccentric data of the axis of spindle of the machine spindle,and made it pattern,then used BP nerve network to train pattern.Secondly we could analysis the result of training to predict future value of deflection.Finally,we used the three-dimensional tensor space distribution function to analysis the fitting effect between future values and theoretical values,we could predict the state of machine tool spindle.The experimental result shows When the number of iterations epoch=2 and training error Validation=0.0052442,fitting effect is preferable,training condition is also the best,thus the results after training can predict the state of machine tool.The method can provide great help for periodic maintenance of machine tool.