基于差异照度的综放开采煤矸混合度精准识别研究

Intelligent control of caving mining is the key technology to realize intellectualization in longwall top-coal caving mining, and automatic recognition of coal gangue mixture degree is the main theoretical and technical bottleneck restricting intelligent caving. At present, when the mechanical properties of coal and gangue are not different, the vibration signal recognition method which has been studied more often is liable to produce larger recognition errors. Based on the basic principle of image recognition and deep learning theory, the approch of accurate recognition of coal gangue mixture degree by differential illumination method is proposed in this project. On the one hand, the response differences of gray, texture and fractal characteristics of coal gangue image under different illumination are studied, and based on this, a deep learning algorithm for recognition of coal gangue mixture degree is established. On the other hand, the reliability test, evaluation and feedback model of coal gangue mixture degree recognition is established by numerical simulation software. Combined with the laboratory test results, the reliability of the proposed algorithm is evaluated comprehensively and optimized to achieve the purpose of precision recognition of coal gangue mixture degree. This project aims to provide theoretical support for the precision recognition of coal and gangue in longwall top-coal caving mining, lay the foundation for the development of automatic caving equipment, and promote the breakthrough of the bottleneck of intelligent longwall top-coal caving mining technology in China.

智能放煤是综放开采实现智能化的关键核心技术，而煤矸混合度的自动识别是制约智能放煤的主要理论和技术瓶颈。目前研究较多的震动信号识别法在煤矸力学性质相差不大时，易产生较大的识别误差。本项目结合图像识别原理和深度学习理论，提出采用差异照度法实现煤矸混合度精准识别的思路。一方面，研究煤矸图像的灰度、纹理以及分形等特征参量在不同照度下的响应差异，并基于此建立煤矸混合度识别深度学习算法；另一方面，通过数值仿真软件建立煤矸混合度识别可靠性检验、评价与反馈模型，结合室内试验结果，对所提出算法的可靠性进行综合评价并针对性优化，达到煤矸混合度精准识别的目的。本项目旨在为综放开采煤矸精准识别提供理论支撑，为自动化放煤成套设备的研发奠定基础，从而推动我国智能化综放开采技术瓶颈的突破。

智能放煤是综放开采实现智能化的关键核心技术，而煤矸混合度的自动识别是制约智能放煤的主要理论和技术瓶颈。低照度、小空间、高粉尘、煤矸叠压、声振信号干扰、夹矸误识别等问题严重制约智能放煤技术开发，本项目提出差异照度煤矸识别方法与技术，将照度概念引入到煤矸图像识别中，揭示了不同照度下煤矸图像特征的变化规律；研究发现可见光图像识别智能放煤技术适用于灰度差异较大的煤矸种类，针对灰度差异较小的难辨别煤矸种类，提出了基于红外图像的液体介入煤矸识别新技术；发现了液体介入后煤、矸石均会出现明显的温降现象，且煤矸间温差会显著增大的规律；揭示了煤、矸石表面微观结构的差异导致煤样表面产生的液滴聚集现象，是液体介入引发煤矸温差增大的内在原因。通过不同液体介入参数条件下的煤矸识别实验，发现液体介入时间及水温对煤矸温差影响显著，提出了采用低于环境温度的液体介入来提高煤矸识别准确率的思路；分析了二次介入对煤矸识别准确率的影响，发现一次介入的识别准确率均值可达92.17%，二次介入识别准确率均值约82.11%，二次介入形成的煤矸石温差小于一次介入。揭示了液体介入后煤矸红外图像中最高温随红外仪发射率的增大而降低、最低温随发射率的增大而升高的规律，发现了将红外发射率设置在较低范围时，煤矸间最大温差达到4.5℃~4.8℃，最大温差效率比可达9，可有效提高难辨别煤矸图像识别效率，构建了基于低发射率的液体介入煤-矸温差调控方法，开发了液体介入煤矸混合度自动识别算法及系列软件，实现了运动状态下基于煤矸红外图像的混合度自动计算。项目研究成果对提高基于图像的煤矸识别准确率有重要意义，为推动我国智能化综放开采技术瓶颈的突破奠定了理论基础。本项目在研究期间，在国内外学术期刊上发表学术论文14篇(SCI收录11篇，El收录3篇)，授权国家发明专利5件，登记软件著作权3件，获得科技奖励2项，协助培养研究生3人(其中博士1人，硕士2人)。