CIESC Journal ›› 2020, Vol. 71 ›› Issue (3): 1264-1277.DOI: 10.11949/0438-1157.20190811

• Process system engineering • Previous Articles     Next Articles

Load identification method of ball mill based on improved EWT multi-scale entropy and KELM

Xiaoyan LUO(),Congcong DAI,Tiedong CHENG,Gaipin CAI,Xin LIU,Jishun LIU   

  1. School of Mechanical and Electrical Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, Jiangxi, China
  • Received:2019-07-12 Revised:2019-08-19 Online:2020-03-05 Published:2020-03-05
  • Contact: Xiaoyan LUO

基于改进EWT-多尺度熵和KELM的球磨机负荷识别方法

罗小燕(),戴聪聪,程铁栋,蔡改贫,刘鑫,刘吉顺   

  1. 江西理工大学机电工程学院,江西 赣州 341000
  • 通讯作者: 罗小燕
  • 基金资助:
    国家自然科学基金资助项目(51464017);江西省教育厅科技重点项目(GJJ150618)

Abstract:

Aiming at the problem that the load of the ball mill is difficult to accurately judge during the grinding process, a ball mill load identification method based on improved empirical wavelet transform (EWT)-multi-scale entropy and kernel extreme learning machine (KELM) is proposed. Firstly, according to the diversity and complexity of cylinder vibration signal, the EWT spectrum segmentation method is improved. By constructing the signal simulation model, the decomposition effect of EWT, EMD is compared, and the effectiveness of the method is proved. Secondly, the intrinsic modal function (IMF) is obtained by using the improved EWT algorithm to decompose the vibration signal of the cylinder under different load states, and then the effective IMF component is selected for reconstruction by correlation analysis. Finally, the multi-scale entropy of the reconstructed signal is extracted as the eigenvector to characterize the different load states of the mill, and the mean value of multi-scale entropy deviation is calculated. The results show that there are obvious differences in the mean value of multi-scale entropy and multi-scale entropy of the three load signals, and the relationship between the three load signals is underload > normal load > overload. The extracted multidimensional feature vector is normalized and used as the input of KELM and the load state of the mill is used as the output. The kernel target alignment (KTA) algorithm is used to optimize the kernel parameters and the optimal model of mill load state identification is established. The feasibility of the method is verified by grinding experiments. Compared with SVM, the overall recognition rate of EWT- is 3.4% higher, and for EMD-multi-scale entropy, EWT-multi-scale entropy is increased by 12.3% and 8.9%, respectively.

Key words: mill load, EWT, optimal, KELM, computer simulation, model-predictive control

摘要:

针对球磨机在磨矿过程中负荷靠经验难以准确判断的问题,提出了一种基于改进的经验小波变换(empirical wavelet transform, EWT)-多尺度熵和核极限学习机(KELM)的球磨机负荷识别方法。首先,针对筒体振动信号的多样性和复杂性特点,对EWT频谱分割方法进行改进,通过构建信号仿真模型,比较EWT、EMD的分解效果,证明该方法的有效性。再将不同负荷状态下的筒体振动信号用改进的EWT算法进行分解得到内禀模态函数(intrinsic mode function, IMF),接着,对分解后的IMF分量进行相关性分析得到敏感分量进行重构;最后,将重构信号的多尺度熵作为表征磨机不同负荷状态的特征向量,并计算多尺度熵偏均值。结果表明:三种负荷信号的多尺度熵及多尺度熵偏均值都存在明显的差异,关系表现为:欠负荷>正常负荷>过负荷。将提取的多维特征向量进行归一化处理并作为KELM的输入,磨机负荷状态作为输出,利用核排列(kernel target alignment, KTA)算法优化核参数,建立磨机负荷状态识别最优模型;通过磨矿实验验证了方法的可行性,相比SVM整体识别率提高了3.4%,且对于EMD-多尺度熵、EWT-多尺度熵分别提高了12.3%、8.9%。

关键词: 磨机负荷, 经验小波变换, 优化, KELM, 计算机模拟, 模型预测控制

CLC Number: