Research On Techniques Of Ball-mill Vibration Signal Feature Extraction And Mill Load Parameters Online Detection

As one of the most critical equipments in the process of grinding,ball mill's load is directly related to many techn-economic indexes of concentration plant such as production efficiency,safety,energy consumption and so on.Butit is difficult to describe explicitly the rules and achieve the real-time control of mill loadwith several time-varying and nonlinear uncertainties effected during the ball mill operation as well as its rotary sealing working characteristic,which bring about low degree of automation of grinding process and large waste of resources.In order to improve the operation stability and economy of the ball mill,it is the key to realize the online detection of mill load to provide guidance of feeding and make the ball mill operate under the best condition.In this thesis,the vibration signal of ball mill shell is taken as the main researchobject and the methods of signal analysis and modeling are taken as the basic means,the methods of ball mill vibration signal feature extraction and mill load detection are given a thoroughly research on,the main work as follows:Firstly,the existing mill load detection methods are compared,the current research situation of vibration signal feature extraction and modeling methods are illuminated.Through the research on mechanical structure,operational mechanism of wet ball mill and explanation of grinding process,the feasibility of taking the shell vibration signal as mill load representational signal is guaranteed.On the basis of the industrial process and expert experiences,the mill load is divided into four categories(empty load,under-load,ideal load and over-load),and the major parameters relevant to mill load are obtained.Seven parameters(the ore feed,water feed,makeup water feed,pulp level in the pump sump,overflow flux,percentage of-200 mesh and motor power)are chosen to be the input arguments of multi-source information fusion model.Aiming at the problem of the unavailable mill load,amethod to identify mil loadbased on an improved K-means clustering method is proposed.After dimensionality reduction using kernel principal component analysis(KPCA),an improved K-means clustering method is used to get the final cluster centers,which are adopted to identity mill load.The improved K-means clustering method is improved by integrating density and partitioning method with agenetic algorithm to set initial cluster centers.Through experiments and comparative analysis,the results show that the improved K-means clustering method proposed are more exact and stable than traditional K-means clustering method.The results of applied on measured data show that the method can be applied to identify mill load and its outcome has a higher reliability.In view of the characteristics of ball mill's vibration signal,a feature extraction method based on harmonic wavelet packet and improved power spectrum is proposed.The vibration signal is decomposed into components in different frequency bands by harmonic wavelet packet decomposition,and the autocorrelation function of each component signals is computed by the improved autocorrelation.The improved autocorrelation method overcomes the shortcoming of autocorrelation function's not making the most of all sample data by shifting and adding operation.The energy centrobaric method is performed to power spectrum correction and improved by constructing the second-order Hanning self-convolution window function combined with frequency deviation method to obtain the preciser frequency of the power spectrum maximums in every layer.The frequency is used as features vector to reflect the mill load.The results of the emulation experiment and applied on measured data indicate that the method can extract more accurate features and have a better differentiation degree and a higher correct recognition rate compared with other methods.Meanwhile,a fuzzy least square support vector machine(LS-SVM)algorithm is selected to realize mill load detection model by comprehensively compared with SVM algorithm and LS-SVM algorithm.The experiment results show that it is available to detect mill load by using the method proposed and t he correct recognition rate reaches 87.5%.At last,by developing a laboratory ball mill,the test experiments platform is built.The methods of vibration signal feature extraction and mill load detection are applied and the virtualization test experiments platform of mill load parameters detection is construted.The feasibility and effectivity of the vibration feature extraction and mill load parameters detection methods proposed is verified by the tests of small ball mill system and measured data.