基于自架构神经网络模型的磨削故障在线监测方法研究

Study on Self-configuration Method of Neural Network Model for Grinding Troubles Monitor On-line

提出利用神经网络非线性建模原理构建磨削故障在线监测模型,提取磨削过程声发射(AE)信号的FFT峰值、RMS峰值以及信号幅值偏差等特征值作为神经网络输入,磨削烧伤、颤振、砂轮钝化等磨削故障作为网络输出,神经网络的拓扑结构通过自架构方法建立。利用实验获得的典型数据训练和测试网络模型,结果表明,利用自架构方法获得的网络模型具有较高的磨削故障识别正确率,可用于磨削故障的在线监测。

A grinding trouble on-line monitoring mode was presented based on the nonlinear building mode principle of neural network. The input units were the peak of the FFT, the peak of RMS, and the standard deviation of AE signals. The outputs were the troubles of the grinding burning, grinding chatter, and grinding wheel dull. The structure of neural network was established by self-configuration method. The network mode was trained and tested by using the experiment data, and the results indicate that the neural network mode obtained by self-configuration method has high recognizing rate for grinding troubles, and can be used to monitor grinding troubles on-line.