A Study On Coal And Gangue Identification Method Based On Vibration Feature Recognition

The mining method of fully-mechanized top-coal caving is one of the main methods of mining extra-thick coal seams in China.Accurate identification of coal and gangue during the top-coal caving process is one of the key technologies to realize the mining automation of fully-mechanized caving.Realization of coal and gangue identification and automatic coal-caving technologies can increase the recovery rate and reduce the mixed rate of gangue,the washing cost,and the impact on people in harsh underground working environment.Different physical properties of coal and gangue lead to differences in vibration information generated by impacting hydraulic supports.Based on the vibration signal analysis,this property is the basic principle of coal and gangue identification.Pattern recognition technology based on vibration analysis has been widely used in fault diagnosis and modal parameters analysis and other fields.Based on the vibration feature identification,the present study takes the coal-gangue identification method as research direction.With the support of national key research and development projects,the present experiment is conducted in a fullymechanized caving face.Main work and achievements are presented as follows:(1)On-site investigations of 8222 working face in Tashan Coal Mine were conducted and the experimental plan for the fully-mechanized caving site based on the actual situation was formulated.Moreover,on the basis of the built vibration test system,vibration data collection tests were carried out.During the top-coal caving process,the tail beam of the hydraulic support was impacted and collapsed by top-coal and gangue.Vibration signals generated by the tail beam were collected.Then,the beginning and end time under different working conditions was recorded.It provided data conditions for the research of coal and gangue identification method based on vibration feature identification in a fully mechanized caving face.(2)The least square method was used to detrend the collected original coal and gangue vibration data,which eliminated the influence of the DC component mixed with the signal and the long period trend term on the signal,and improved the accuracy of the vibration signal.The vibration data of coal and gangue were denoised by EEMD decomposition and reconstruction.Results show that after performing EEMD decomposition on the coal gangue signal,the first 8 IMF components are highly correlated with the original signal.Therefore,the first 8 IMF components can be selected as effective IMF components for further feature extraction.Besides,the feature information can be extracted to represent the two working conditions of coal caving and gangue caving.(3)On the basis of performing EEMD processing on vibration samples,IMF energy,IMF energy moment,IMF kurtosis,and IMF singular value were extracted from the first 8 IMFs as feature vectors.A preliminary analysis of the effectiveness of each feature was carried out.A 960x43-dimension original coal and gangue vibration feature data set was constructed.The principal component analysis method was used to reduce each feature vector’s dimensionality.After doing this,the dimensionality of the feature data set is 960x24.A part of feature data after dimensionality reduction are analyzed visually.Results show that feature data after dimensionality reduction have good separability for the two working conditions,which further verifies the effectiveness of the extracted features.(4)Based on EEMD feature extraction,PCA feature dimensionality reduction,and SVM recognition,the EEMD-PCA-SVM coal and gangue recognition algorithm was proposed.The coal and gangue feature data set,which was reduced dimensionality,conducted modeling training and then model parameters were optimized.The effectiveness of this recognition algorithm is verified by using 100 groups of vibration data under two working conditions of coal caving and gangue caving.Results show that the average recognition accuracy of the EEMD-PCA-SVM algorithm for vibration signals under the two working conditions is 89.5%,indicating a significantly high recognition accuracy rate.Based on EEMD feature extraction and random forest recognition,the EEMD-RF coal and gangue recognition algorithm was proposed.The original vibration feature data set was used for model training and related parameters were optimized.The effectiveness of the recognition algorithm is also verified by using100 groups of vibration data of coal caving and gangue caving.Results show that the average recognition accuracy of the EEMDRF algorithm for vibration signals under the two working conditions is up to 92%.In terms of the coal-gangue identification problem based on vibration feature identification,the prediction performance of EEMD-RF algorithm is slightly better than that of EEMD-PCA-SVM algorithm.In summary,the two recognition algorithms proposed in the thesis have high accuracy in identifying coal and gangue samples.Both of them can better complete the identification of the two conditions of coal caving and gangue caving.The thesis contains 55 diagrams,20 tables and 122 references.