【摘要】 提出了基于频域特征提取与多传感器信息融合的磨机负荷(ML)软测量新方法。针对磨矿过程主要依靠人工经验定性判断ML状态,难以定量检测ML参数的现状,通过融合磨机筒体振动、振声及驱动电机电流信号,建立了以料球比、矿浆浓度、充填率为输出的ML软测量模型。该方法首先采用快速傅里叶变换(FFT)将时域振动及振声信号转换为频谱变量,再对频谱变量通过主元分析(PCA)进行谱特征提取,然后采用径向基函数(RBF)变换生成的激活矩阵实现谱特征的非线性映射,最后采用偏最小二乘(PLS)算法建立以谱特征、激活矩阵、电流信号为输入的回归模型,从而有效克服了多传感器信息之间及RBF变换引起的多重共线性等问题。实验表明,该方法能够较准确地检测ML参数,融合多传感器的软测量方法具有更好的预测效果。更多还原

【Abstract】 A novel approach based on frequency domain feature extraction and multi-sensor information fusion for soft sensing of mill load (ML) is proposed. Aiming at the problem that the state of mill load is mainly qualitatively estimated by the experience of operators and the ML parameters can not be quantitatively monitored in grinding process, a soft sensor model of the ML parameters (material to ball volume ratio, pulp density and charge volume ratio) is built based on fusing the mill shell vibration, acoustic signal and driver motor electric signal. In this approach, fast Fourier transform (FFT) is first used to transform the vibration and acoustic signals in time domain into frequency spectral variables, then the spectral features of the frequency spectral variables are extracted using principal component analysis (PCA); the mapping of the spectral features in nonlinear space is realize with the active matrix generated from radical basis function (RBF) transformation. At last, partial least squares (PLS) algorithm is used to establish a regressive model whose inputs are spectral features, active matrix and driver motor electric signal in time domain. The multi-collinearity caused by multi-sensor information and RBF transformation is overcome effectively. Experiment result shows that the proposed soft sensing approach can predict the ML parameters accurately, and the method of multi-sensor fusion has better performance than single sensor method.更多还原