Research On Hybrid Ensemble Modeling Method Of Mill Load Based On Multi-Scale Spectrum

The grinding process is to crush the broken ore through the ball mill into a qualified pulp,such that certain concentrate is obtained.This process is a key link between the production of iron ore and non-ferrous metal minerals.The mill load refers to the ore load(including broken ore,etc.),ball load and water load inside the mill.The mill load is directly related to quality,efficiency,energy consumption,material consumption and the safe operation of the grinding process.Overloading can cause the spitting ore,coarsening the mill output,mill blocked,and belly being full.All of these broke the production process.Conversely,the low load can cause the "running with only ball".This leads to energy consumption and steel consumption,even destroying milling devices.Accurate detection ofthe mill load is one of the key factors in the optimum operation of the grinding process.Because the mineral and water in the grinding process are continuously changed,corrosion and wear of the steel ball are not known.The closed and continuous operation of the mill and the grinding mechanism are difficult problems.These need online measurement of mill load.The mill load cannot be measured directly,while the mechanical vibration signal and the acoustic signal can be measured.In real application,the operation experts identify the"sharp" and "boring" acoustic blurring characteristics via their ears.They use their experience to reason the mill's charging into "high","medium" and "low" blur states.Recently,some researchers use the vibration signal,acoustic signal and the load parameters of the mill(ratio material to ball,pulp density and load volume ratio),to estimate the load of the mill.Their methods are based on the models of the grinding.In this thesis,milling load is estimated by using the high frequency and multi-frequencycharacteristics of the vibration/acoustic spectrum.Based on multiple scale vibration/acoustic spectrum characteristics,the efficacy of the proposed mill load sensing model is verified by the experimental data of a wet ball mill.The main contributions of this thesis can be summarized as follows:(1)Based on the analysis of the dynamic characteristics of the mill load,the modelingstructure of milling load sensors and the function of each component are presented.The model consists of multi-scale vibration/acoustic spectrum feature selection and extraction,mill load parameters(MBVR,PD and CVR within the ball mill)model,and the mill load hybrid ensemble model uses mill load parameters.(2)Since it is difficult to extract and select the characteristic of the vibration signal,a method of extraction and selection of vibrations spectrum characteristics of multi-scales is proposed.EMD and FFT technology are used to obtain the spectrum of vibrations of multi-scales.KPCA and MI approach are applied to extract and select the multi-scale vibration spectrum function.We use these methods to model mill load parameters as in[28].The experimental study was carried out with the experimental mill.The experimental results show that the method of extraction and selection of multi-scale frequency spectrum characteristic can improve the accuracy of mill load parameters compared with the single-scale frequency spectrum method[28].(3)There are fuzzy type characteristics in the multi-frequency spectrum of vibration/acoustic signal,mill load parameters,redundancy and complementarities of the multi-frequency spectrum.Also,it is difficult to simulate the operation of expert "listen"inference to identify the load parameters of the mill.In this thesis,by using KPLS,MI,online clustering,Madani fuzzy model,derivation and linkage(BB),weighted adaptation fusion(AWF)and modeling of selective sets,we proposed a fuzzy logic based model,which selects and extracts adaptively the characteristics of the vibration/acoustic spectrum at multiple scales.The experiments are carried out in the experimental wet mill.The experimental results show that the proposed model can simulate the expert's reasoning mechanism and has better modeling accuracy.(4)Based on the proposed mill load parameters fuzzy inference model,multi-scale vibration/acoustic spectrum characteristics,an integrating model combining the latent selective mapping is proposed.This method consists of the load model and the load compensation model by using the random weight neural network.The experiments are carried out in the experimental mill.The experimental results show that the MSE,MSPE,RMSR and RMSRE of the mill load estimation are within the required accuracy ranges.