| Unmanned working face and intelligent mining are the urgent needs and effective ways to achieve safe,efficient and green targets in coal mines all over the world.The key basic problems that restrict intelligent mining in underground coal mine are intelligent and precise perception of mining environments and autonomous and cooperative control of mining equipments.Coal-rock recognition is one of the key technologies to realize unmanned working face and intelligent mining,and is also the core technology to achieve intelligent and precise perception of mining environments.However,coal-rock recognition has always been a major problem that has not been solved in research and application of unmanned and intelligent mining of coal mine.The difference in material compositions between ground objects makes their reflectance spectra characteristics different.This basic property is the basic principle of hyperspectral remote sensing detection and recognition of ground objects.At the same time,reflectance spectroscopy technologies have been applied in the field of remote sensing of coal and rock mines.Inspired by remote sensing detection of ground objects,this paper aims to study characteristics and differences of visible and near-infrared reflectance spectra of coals and coal measures rocks of roof and floor,and explore methods for recognizing coal and rock based on reflectance spectroscopy technology.The main works and results are as follows:(1)Seventy-five types of coal and coal measures rock in typical mining areas of China were collected.Two types of block samples and nine types of powder samples were made for each type of coal and rock.Visible and near-infrared reflectance spectrum of each sample prepared was measured at close range.A spherical coordinate device for measuring coal-rock reflectance spectra was designed.Geometric near-infrared reflectance spectra of each sample prepared were measured.Compositions of each type of coal and rock were analyzed.Based on spectral reflectance data of 11 samples prepared,material compositions etc.of the 75 coal-rock types,the reflectance spectra library of coals and coal measures rocks was established,and based on the MATLAB GUI module,the query software system of the reflectance spectra library of coals and coal measures rocks was designed,which establishes the foundation for research of reflectance spectra characterization of coal and rock,recognition methods,light source-probe sensor device,etc.(2)Characteristics of visible and near-infrared reflectance spectra of typical coal types and their mechanisms in the aspect of material composition were analyzed.It was found that with decreasing coal rank,positive inclination,reflectance and number of absorption valleys of the near-infrared reflectance curve all increase.Two types of parametrization using spectral slope and sum of depth of absorption valleys were carried out to process reflectance spectra of coals.The relationships between the two parameters and coal rank and related material composition were obtained.When the coal rank is higher or lower,the spectral slope increases linearly or exponentially with decreasing coal rank,respectively.The moisture content,the content of main mineral elements and the volatile matter yield of coal based on air drying are are strongly,moderately and weakly linearly related to their sum of depth of absorption valleys,respectively.Characteristics of visible and near-infrared reflectance spectra of typical coal measures rock types and their mechanisms in the aspect of material composition were analyzed.The results showed that the more obvious absorption valleys of coal measures rocks mainly distribute near four wavelength points of 1400 nm,1900 nm,2200 nm and 2350 nm.The differences between reflectance spectra curves of typical coal and rock in each coal mine of four representative coal mines were analyzed,and it was found that the band with the most universal differential absorption valley between coal and rock mainly distribute near two wavelength points of 2200 nm and 2350 nm.The effect of carbon content on reflectance spectrum of carbonaceous shale was studied through a simulation experiment.The results showed that when the content of carbonaceous matter is more than about 35%,the reflectance spectrum waveform of carbonaceous shale changes from convex to concave and the main absorption valleys become not obvious.(3)No.1 anthracite,meager coal,gas coal,No.2 lignite,carbonaceous shale,siltstone and argillaceous limestone were selected as representative coals and rocks.Under different azimuth and reflection angles,minimum included angle between incident and reflection light,different detection distances,different particle sizes and surface roughness,near-infrared reflectance spectrum curves in the spectral library and change rules of their characteristic parameters of these seven coals and rocks were analyzed.The results showed that with decreasing coal rank,the back reflectance of coal is enhanced,and the bidirectional reflection is more and more obvious.All the three rocks have more obvious back reflection characteristics,and the ranges of the forward and backward reflection angles with the maximum reflectance were selected as the best spectral detection angle of coal and rock.With increasing reflection angle under minimum included angle between incident and reflection light and increasing detection distance,reflectance spectrum curve,spectral slope and depth of more obvious absorption valleys of coal and rock all decrease.With decreasing particle size,the reflectance spectrum curves of coal and rock show a rising rule,and the reflectance of rough surface is less than that of smooth surface.Near-infrared reflectance spectrum curves and change rules of their characteristic parameters of these seven representative coals and rocks under different surface water contents were studied through an experiment and it was found that different types of coal and rock show different characteristics.The fitting equations between reflectance at characteristic wavelength points and detection conditions of main influencing factors were established and the coefficients of determination were all greater than 0.8.The equations can be used to predict and modify spectral reflectance in wavelength ranges near the characteristic wavelength points under different detection conditions.(4)Preprocessed spectral curves of reflectance spectra through five methods of these seven representative coals and rocks were analyzed and continuum removal was selected as the preferred preprocessing algorithm for coal and rock reflectance spectra.For coal and rock with obvious differences in absorption valleys,four basic recognition algorithms based on waveform matching with the spectral library were studied.These four algorithms are RBF neural network classification using spectra in full-band,RBF neural network classification using characteristic parameters of absorption valley in differential band,included angle cosine matching of spectral vectors in differential band and classification using spectral slope index threshold in differential band.For bituminous coal and carbonaceous shale with similar spectral waveforms,based on extraction and classification of principal components of continuum removal spectra in eight characteristic bands related to major material compositions of coal and rock,two recognition algorithms-PCA-SVM and GRB-KPCA-SVM were established.The prediction accuracies of test coal and rock samples were all over 90% by the algorithms established based on reflectance spectra matching and features extraction and the real-time performances were good.(5)Taking the typical in-situ block coal and rock samples of Xinglongzhuang and Malan coal mines as examples,near-infrared reflectance spectra detection experiments of different back reflection angles and long distance were carried out respectively,and two types of in-situ coal-rock recognition models based on geometric reflectance spectra were established.For three coal-rock combinations with obviously different spectral waveforms in Xinglongzhuang coal mine including gas coal-mudstone,gas coal-siltstone and gas coal-argillaceous limestone,the common band range with the greatest different spectral waveforms of the four coal and rocks at different back reflection angles of 0°--75° were obtained,and the range was 2150-2400 nm.Three unsupervised and self-adaptive in-situ coal-rock recognition models based on reflectance spectra at different detection angles in the common band range were established.The CFCM clustering with cosine distance as clustering distance was chosen as the preferred model.The optimal weighted indexes for recognition of the three coal-rock combinations were 2.0,2.0 and 1.1,respectively.The recognition rates were 93%,98% and 100%,respectively.The total recognition time of each combination was less than 0.1 s.For bituminous coal and carbonaceous shale with similar spectral waveforms in Malan coal mine,the correlations between long-range reflectance spectra and ash yields of in-situ coal and rock samples were studied.The maximum correlation coefficient was 0.777,which appearred at the wavelength point of 1698 nm and was obtained by the preprocessing method of continuous removal.Based on the long-range continuum removal spectra at 11 wavelength points near the wavelength point of 1698 nm,two in-situ coal-rock prediction and recognition models-SVR for ash yield and SVC for coal-rock type were established.The remote recognition rates of in-situ coals and rocks from the same coal mine were both over 90% by the two models,and the prediction time for each sample was less than 0.1 s.There are 147 figures,18 tables and 195 references in this dissertation. |
PDF Full Text Request