Study And Application On The Caving Automation In Fully Mechanized Top Coal Caving Face

After thirty years of development, coal caving mining technology has reached a high level in theoretical research and engineering application, and both mechanization and automation have been greatly improved with the development of equipment and technology of fully mechanized top coal caving mining face. In coal automation, although there are the basic automation production conditions, due to the lack of theoretical research and technology, the effect of automation production cannot be achieved. So the research and realization of caving automation and coal gangue recognition have important significance and value in the theory and practice. The main works in the caving automation control are showed as follows:1)Based on the existing fully mechanized top coal caving face automation system, the composition and the logical relations are analyzed about coal mining and caving automation in top coal caving face automation system, and the caving automation system platform is built. The method and experimental platform of coal gangue recognition based on vibration and acoustic signals is constructed. Through the introduction of proportional parameters of coal gangue and caving fault parameters B, the thesis solves the state parameters digitization in the caving process, such as:full coal falling (ζc),30% gangue mixture falling(ζ0.3),50% gangue mixture falling(ζ0.5), full gangue falling(ζg) and B, and it presents a method of stopping caving according to recognizing the coal gangue proportional and setting the critical point ζs, that is a precondition for solving coal process automation control.2)On account of analyzing the classic wavelet threshold denoising method, combined with the non-linear and non-stationary problem of vibration and acoustic signals in top coal caving process, and there still exists large difference between individual points and measured data after signal preprocessing, the thesis introduces the denoising factor associated with its wavelet coefficients, called the adaptive threshold wavelet denoising. The practical engineering analysis demonstrates that the algorithm can fully consider the energy distribution characteristic of the signal wavelet decomposition, and can achieve better denoising effect.3) Combined with Hilbert Huang Transformation (HHT) theory, we calculated the energy of the each IMF and the total energy, energy entropy and kurtosis characteristics.The vibration and acoustic signals are decomposed by the wavelet packet based on multi-resolution analysis, and the signals are restructured according to each subspace (spectrum).Then the energy and corresponding information entropy of each frequency band are calculated for the restructured signal. In addition, the fractal dimension, spectral centroid (SC), Mel-Frequency Cepstrum Coefficients (MFCC) and other important feature attributes are calculated, and then 16 feature attributes of coal gangue recognition are obtained in the thesis.4)The thesis proposes a new dimension reduction method for multi class feature based on F-Score, and realizes the feature reduction of coal gangue datasets during caving. Based on analyzing the domestic and foreign existing feature and attribute reduction algorithm, the multi class F-Score feature reduction algorithm (MF-Score) is put forward in the light of the defects that F-Score algorithm is mainly for two class feature reduction. In order to avoid the features information loss after feature reduction, the feature threshold with contributions is given, so that the threshold can be adaptively adjusted according to the classification accuracy requirement, and the new feature set is recombined. Simulation and experiment results verify that this method is very suitable for the feature reduction of multi class and multi features datasets.5)In order to solve the problem of five class classification about coal gangue dataset in caving process, the minimal enclosing sphere with adaptive width factor (σ-MEB) algorithm for multi class classification is proposed in the thesis. SVM usually uses the indirect ways in solving the multi classification problem, then the more training samples are needed, the learning efficiency is low, further more, the minimum enclosing ball often occurs over fitting or under fitting in solving multi classification problem. Aiming at these problems, the σ-MEB algorithm is presented. The six typical UCI datasets as well as vibration, acoustic datasets in caving verifies the reliability and adaptability of the method.6)The fusion structure and fusion method of multi information fusion technology are expatiated in the coal gangue recognition. Then, aiming at incompleteness and ambiguity of classification characteristics in caving process, the multi-model information fusion is put forward. Experiments show that the multi-model caving automation system fusing with acoustic and vibration singles of coal gangue has higher performance than the single mode state.7)The caving automation system based on Lab VIEW is developed and designed, then applied in the field. So the tail beam of hydraulic support, caving flashboard, rear conveyor with pulling are automatically controlled in caving process, according to coal gangue proportional.At last, the caving automation is completed in top coal caving face.