基于振动信号的采矿机截割负载分类

Mining machine cutting load classification based on vibration signal

针对人为判断采矿机截割负载类型的方式具有一定误差和滞后性的问题,提出了一种基于小波包分解和麻雀搜索算法优化BP神经网络(SSA−BPNN)的采矿机截割负载分类方法。该方法包括信号特征提取和模式分类2个部分:在信号特征提取部分,对采集的采矿机摇臂振动信号进行小波包分解,得到各子频带能量及信号总能量,经归一化处理后得到表征不同负载类型的特征向量,并利用主成分分析法对特征向量进行降维处理;在模式分类部分,通过SSA优化BPNN的初始权值和阈值,将特征向量作为SSA−BPNN的输入,从而实现负载分类识别。以MG500/1170−AWD1采矿机为对象,将磁吸式加速度传感器吸附于采矿机摇臂一轴靠近支架侧的壳体处,采集采矿机滚筒空载、截割铝土和岩石3种工况下的振动信号进行试验。试验结果表明:不同截割负载下振动信号在各子频带能量上表现出一定的差异性,表明经小波包分解后得到的能量特征可以作为区分不同负载类型的特征向量;与BPNN相比,SSA−BPNN收敛速度更快、识别准确率更高,负载分类识别准确率达95.3%。

There are some errors and lags in the way of judging the cutting load type of the mining machine manually.In order to solve the above problem,a classification method of mining machine cutting load based on wavelet packet decomposition and sparrow search algorithm optimized BP neural network(SSA-BPNN)is proposed.The method comprises two parts of signal feature extraction and mode classification.In the part of signal feature extraction,the collected vibration signal of the mining machine rocker arm is decomposed by wavelet packet to obtain the energy of each subband and the total energy of the signal.After normalization,feature vectors representing different load types are obtained.The principal component analysis is used to reduce the dimensions of the feature vector.In the mode classification part,SSA is used to optimize the initial weight and threshold of BPNN.The feature vector is used as the input of SSA-BPNN to realize the load classification and recognition.Taking the MG500/1170-AWD1 mining machine as an object,the magnetic acceleration sensor is attached to the shell of the rocker arm of the mining machine near the bracket side.The vibration signals of the mining machine drum under three working conditions of no-load,cutting bauxite and rock are collected and tested.The experimental results show that the vibration signals under different cutting loads have some differences in the energy of each sub-band.This result indicates that the energy features obtained by wavelet packet decomposition can be used as feature vectors to distinguish different load types.Compared with BPNN,SSA-BPNN has faster convergence speed and higher recognition accuracy,and the recognition accuracy of load classification is 95.3%.