Research On Ball Mill Load Prediction Based On Vibration Feature Extraction

Ball is the key of grinding equipment after the material to be broken.The efficiency of grinding will be directly affected by the changes of internal ball load.Theinterior ofcylinder so like a black box that the load can’t bedescribed explicitly and controlled as well as can’t determine the actual performance of the mill.Thus,the effectiveprediction of realizing the mill internal loads and the running millat optimal working condition are fundamental task,which is to improve the efficiency of ore-dressing and reduce cost of production.The passage,which takes theΦ330mm × 330 mmexperimental ball as research subjects,through using the method of theoreticalanalysis,numerical simulation,experimental study,completes the work as follows.Vibration feature extraction and prediction methods of ball load are researched in-depth,considering the impact of product quality and energy consumption,to achieve predict accurately of ball load status parameters.The main results are as follows.First,through analyzing motion distribution of ball mill load,the vibration measurement position is fixed according to the principle of force balance.The vibration signal analysis period is determined by numerical simulation according to the theory of media motion stratification.For the effect of product quality and energy consumption,the ball load test analysis is conducted according to the filling ratio and material-ball.From the two angles of energy consumption and yields of-200 mesh particle,evaluation of “yield ratio” is established to achieve grinding efficiency of ball.Second,based on the fact that vibration signalexists the component of interfering signal,the wavelet thresholding algorithmis used to pretreat the vibration signal in the load state of ball.Analysing the effect of different wavelet functions,decomposition scale,threshold,threshold function,based on the SNR and RMSE,the optimal solution of wavelet thresholding will be received.Third,time-domain signal statistical feature are achieved by analysing the correlation between the vibration signal and ball mill load.Analysis of statistical principle between load parameters and the frequency component of ball load,the frequency domain statistical model is established to achieve frequency features.According to the vibration signal characteristics of ball load state,feature vector of peak to peak,mean,standard deviation,skewness,kurtosis and spectrum band energy of AR model are constructed.A multi-feature parameters fusion model is established based on D-S evidence theory to obtain the contribution rate for different characteristic parameters of ball mill load.Finally,the support vector machine parameters C and g are optimized by using grid search and cross validation algorithm(K-fold)and genetic algorithm(GA)respectively.The ball load forecasting model of time-domain feature and frequency domain features are established by the best performance matching parameters.The simulation results show that the parameters of SVM are optimized quickly and accurately to achieve the prediction of ball load by the K-SVM and GA-SVM algorithm.Through the above research shows,the prediction of ball load parameters of considering product quality and energy consumption has a positive role in guiding to improve grinding efficiency.