| Coal,as the primary energy source in China,can significantly enhance its value through washing and selection processes.Currently,wet separation methods such as vibrating screens and shallow troughs are predominantly used in coal separation,which consume excessive water and energy,involve complex processes,and pollute the environment.The proportion of coal selected for use in China differs by approximately15.0% compared to major coal-producing countries worldwide,mainly due to the lack of advanced separation methods,technologies,and equipment.Dual-energy X-ray transmission(DE-XRT)coal separation is based on the differential X-ray absorption responses of coal and gangue,which are detected by sensors to achieve identification.Combined with gas jet injection,this method achieves separation without water consumption,environmental pollution,or the need for human intervention,thus potentially replacing existing wet separation methods.However,the currently deployed separation devices suffer from drawbacks such as random particle size distribution,leading to low precision in segmentation and identification,as well as inefficient rejection of impurities.DE-XRT technology itself is affected by hardening,afterglow,fan-shaped effects,and thickness variations,which not only limit the range of identifiable particle sizes but also reduce recognition accuracy.The core algorithms fail to effectively establish control mechanisms for separation parameters tailored to different regions and qualities of coal and gangue,thus limiting the performance and application scope of the devices.In this study,the following research is conducted to address the deficiencies and technical challenges of existing DE-XRT intelligent separation devices:(1)Analysis of the principles of DE-XRT coal and gangue identification,revealing the absorption response laws of coal and gangue to X-rays and explaining the effects of hardening,afterglow,fan-shaped,and thickness variations.Preparation of materials with different qualities,thicknesses,and densities based on coal and gangue properties.Collection and establishment of a coal and gangue image dataset,followed by preprocessing using Gaussian filtering for noise reduction,non-uniformity correction using the two-point method,afterglow restoration using Lucy-Richardson method,and segmentation based on block maximum interclass variance thresholding.(2)Design of shape selection and recognition algorithms to guide segmentation in cases of overlapping and adhering particles.Prediction of state parameters such as overlapping rates and density distributions of dispersed queues.Detection of concave points based on the relationship between conjugate straight lines to separate overlapping coal and gangue.For complex multi-target adhesions,the design of a watershed segmentation algorithm based on multi-level grayscale threshold iteration and seed point marking is proposed.This achieves precise positioning of targets and lays the foundation for recognition.Research indicates that within a thickness range of5 mm to 150 mm,the accuracy of shape selection recognition exceeds 98.0%,and the prediction error of state parameters of dispersed queues is less than 4.1%.The accuracy of segmentation and positioning of overlapping and adhering targets is above 94.0%.(3)Considering the deficiencies in DE-XRT recognition technology such as thickness,afterglow,hardening,and fan-shaped effects,a machine learning model is introduced to automatically create classification models,proposing a multidimensional high-precision recognition method.Features are extracted from multiple images,layers,and angles,integrating grayscale,texture,geometric shape,and X-ray absorption response to mitigate the impact of defects.This not only achieves recognition of raw coal and pre-rejecting gangue but also establishes a mechanism for controlling separation parameters,meeting the needs of different regions for coal and gangue separation.Furthermore,the distance from multi-dimensional features in highdimensional space to the support vector machine classification hyperplane is used to characterize the density changes in DE-XRT imaging details.Research demonstrates that the recognition rate for raw coal pre-rejection within a thickness range of 5 mm to150 mm exceeds 99.0%,and the selection of separation parameters based on gangue density changes is relatively low,revealing the selection relationship between key separation parameters-density-ash content mapping.(4)Based on the regularity of dispersed queue distribution,an innovative design of a coal and gangue drum screening device is proposed to arrange particles from small to large diameters in a regularized manner,replacing the existing random arrangement.Improvements are made in the design of separation device structures and rejection schemes,effectively enhancing the accuracy of shape selection recognition and segmentation positioning,reducing gas consumption,and revealing the impact of overlapping and adhering target segmentation on recognition.Critical system components such as fabric,radiation protection,source temperature control,material conveying,rejection,and feeding systems are designed and tested.A DE-XRT separation device test platform is constructed to validate the effectiveness of segmentation-based coal and gangue identification and positioning methods.Comprehensive sorting test results show that under regularized arrangements,the accuracy of shape selection recognition improves by 5.6%,and the minimum improvement in segmentation accuracy is 16.9%,while the minimum improvement in centroid positioning accuracy is 12.0%.By utilizing nozzles with diameters of 10 mm and 6 mm to respectively eject large and small coal and gangue particles,a single response from the 6 mm nozzle can conserve 64.3% of the gas volume.The comprehensive separation test results show that the accuracy of coal and gangue identification based on segmentation is over 99.0%,and the quantity ratio of coal in gangue and gangue in coal after separation is less than 2.0%.This study provides theoretical references for the identification and positioning algorithms of existing DE-XRT intelligent sorting devices,and offers design solutions to enhance device performance.The proposed identification and positioning algorithms can be deployed rapidly as general algorithms and can be extended to the sorting of other mineral resources and recyclable solid waste.Figure [168] Table [36] Reference [161]... |
PDF Full Text Request