综掘工作面风幕阻尘效果影响因素研究

夏丁超, 吕品, 杜朋, 王金月

夏丁超,吕品,杜朋,等. 综掘工作面风幕阻尘效果影响因素研究[J]. 工矿自动化,2024,50(1):72-79. DOI: 10.13272/j.issn.1671-251x.2023060007
引用本文: 夏丁超,吕品,杜朋,等. 综掘工作面风幕阻尘效果影响因素研究[J]. 工矿自动化,2024,50(1):72-79. DOI: 10.13272/j.issn.1671-251x.2023060007
XIA Dingchao, LYU Pin, DU Peng, et al. Factors influencing the dust-blocking effect of air curtains during the fully mechanized excavation of working faces[J]. Journal of Mine Automation,2024,50(1):72-79. DOI: 10.13272/j.issn.1671-251x.2023060007
Citation: XIA Dingchao, LYU Pin, DU Peng, et al. Factors influencing the dust-blocking effect of air curtains during the fully mechanized excavation of working faces[J]. Journal of Mine Automation,2024,50(1):72-79. DOI: 10.13272/j.issn.1671-251x.2023060007

综掘工作面风幕阻尘效果影响因素研究

基金项目: 国家重点实验室开放基金项目(JYBSYS2019102)。
详细信息
    作者简介:

    夏丁超(1997—),男,安徽安庆人,硕士研究生,主要研究方向为矿井粉尘防治,E-mail:1546119121@qq.com

    通讯作者:

    吕品(1963—),男,安徽来安人,教授,博士,主要从事煤矿灾害事故控制理论和技术、安全评价理论和方法、火灾控制理论和技术等方面的教学与研究工作,E-mail:plv@aust.edu.cn

  • 中图分类号: TD714

Factors influencing the dust-blocking effect of air curtains during the fully mechanized excavation of working faces

  • 摘要: 目前综掘工作面粉尘污染的研究多集中于单一因素对综掘工作面风幕阻尘效果的影响,而未充分考虑各因素间的交互作用,使得压风分流技术的工程应用效果欠佳。为明确附壁风筒径向出风距离、径向出风比及轴向出风距离对风幕阻尘效果的影响,以潘三矿810西翼机巷综掘工作面为研究对象,运用Fluent软件对径向出风距离为10~25 m、径向出风比为0.6~0.9及轴向出风距离为6~12 m条件下的风流分布和粉尘扩散情况进行数值模拟。结果表明:① 随着径向出风距离增大,径向涡流风幕在巷道内的转变更充分,综掘机司机前端的风流分布越均匀,更有利于形成风速方向均指向工作面的轴向阻尘风幕。当径向出风距离为10 m时,距工作面7 m断面内涡流特性明显,风速方向紊乱;当径向出风距离为25 m时,距工作面7 m断面内,风流分布趋于均匀,风速方向均指向工作面,形成了能够覆盖全断面的轴向阻尘风幕。② 随着径向出风比增大,整流风筒轴向风流风量减小,轴向风流风速和射流强度降低,轴向风流对综掘工作面前端气流的扰动减弱;径向出风比越大,越有利于形成风流方向指向工作面且能覆盖全断面的轴向阻尘流场,即轴向阻尘风幕。③ 径向涡流风幕的阻尘能力随径向出风比的增大先增强后减弱,轴向阻尘风幕的阻尘能力随径向出风比的增大而不断增强。④ 在采取压风分流风幕阻尘技术后,当压风总量为300 m3/min,吸风量为400 m3/min,附壁风筒径向出风距离为20 m,径向出风比为0.9,整流风筒轴向出风距离为8~10 m时,能很好地将粉尘聚集在吸尘口附近,达到高效控尘除尘的目的。在810西翼机巷综掘工作面进行现场测试,测点风速和粉尘质量浓度实测值与模拟值基本一致,高浓度粉尘被有效阻控于工作面前端,隔尘效果较为明显,验证了数值模拟的有效性。
    Abstract: Prevalent research on dust pollution during fully mechanized excavation has mainly focused on the impact of individual factors on the effectiveness of air curtains in fully mechanized excavation sites. However, scant research has been devoted to the interaction between factors, because of which pressure-induced air diversion technology has not been adequately applied to this context.To investigate the impact of the radial distance of the outlet of air, the ratio of this outlet, and the distance between the outlet and the wall-coated air duct on the effectiveness of dust blocking by air curtains, the authors of this study consider the excavation of the working face of the 810 west wing machine tunnel at the Pansan Mine . We used Fluent software to numerically simulate the distribution of wind flow and the diffusion of dust under a distance of the radial outlet of air of 10-25 m, a ratio of the outlet of 0.6-0.9, and an axial distance of the outlet of 6-12 m.The results showed that: ① As the distance of the radial outlet of air increased, the radial vortex air curtain transforms more fully in the tunnel . The wind flow at the front end of the excavation operator was more evenly distributed, and the wind speed was directed toward the working face such that this was more conducive to the formation of an axial dust-blocking air curtain.When the radial distance of the outlet of air was 10 m, vortical characteristics became apparent within a distance of 7 m from the working surface, and the direction of wind became disordered. When the radial distance of the air outlet was 25 m, the wind flow tended to be uniform within 7 m of the working surface, and its direction was evenly distributed toward the working surface. This led to the formation of an axial dust-blocking wind curtain that could cover the entire section.② As the ratio of the radial outlet of air increased, the volume of axial airflow of the rectifier air cylinder decreased to reduce the velocity of axial airflow and the intensity of the jet. This in turn reduced the disturbance caused by the axial airflow to that at the top of the mechanized working face that was being excavated. A higher ratio of the radial outlet of air was more conducive to the formation of an axial dust-blocking flow field, with the wind directed toward the working surface and covering the entire section. This led to an axial dust-blocking air curtain. ③ The dust-blocking ability of the radial vortical air curtain initially increased and then decreased as the ratio of the radial outlet of air increased. Its ability then continued to improve as the ratio was further increased. ④ We implemented the dust-control technology based on the air curtain with forced ventilation-induced diversion. When the pressure-induced volume of air was 300 m3/min and the volume of air suction was 400 m3/min, the distance between the radial outlet of air and the attached wall of the air duct was 20 m. The ratio of the radial outlet of air, and the distance between this outlet and the air duct of the rectifier was 8-10 m. The air curtain was able to collect dust near the port of the dust suction for efficient dust control and removal.We conducted an on-site test of the fully mechanized excavation working face of the 810 west wing machine tunnel. The empirically measured data of wind speed and dust mass concentration at measuring points and the results of numerical simulations were consistent with each other. Highly concentrated dust was blocked at the front end of the working face, and its isolation was noticeable. This confirms the effectiveness of the numerical simulations.
  • 刮板输送机是煤矿工作面唯一的运输设备,链条是其关键运行部件。当出现刮板输送机断链故障时,若不能及时发现,将导致链条堆积,严重影响煤矿生产安全和效率。因此,众多专家学者对刮板输送机断链监测技术进行了研究。初期大多针对刮板输送机断链故障原因和预防措施进行分析,建立链条的强度条件[1]和有限元仿真模型[2],探究卡、断链故障发生后刮板输送机动力学特性和链环之间接触力的变化规律[3],并对落煤冲击条件下的链条进行动力学分析[4],进而对断链后链条张力响应进行仿真研究[5]。目前刮板输送机断链监测方法主要包括压力监测法、差速监测法和视觉监测法[6]。吴孙阳等[7]设计了一种基于应力突变的刮板输送机断链监测系统,利用应变传感器测量与不同链条啮合的链轮轮齿受力面的应变,及时检测断链隐患;Zhang Xing等[8]提出了一种基于溜槽振动分析的刮板输送机链条故障检测策略,利用加速度传感器检测刮板输送机断链故障引起的溜槽振动信号,通过振幅识别断链故障;Hua Yilian等[9]通过安装在刮板两侧的超宽带节点实时反馈刮板是否出现倾斜情况,实现对刮板输送机断链故障的间接监测;高昌乐等[10]以刮板输送机链轮转速差为依据,当转速差超过预警值时,判断刮板输送机链条出现断裂现象;崔卫秀等[11]利用计算机诊断和AI视频识别技术,通过图像采集、分析和处理,对链条状态进行检测;Zou Huadong等[12]提出了一种基于划痕特征检测的视觉识别方法,通过监测识别链条划痕预测刮板输送机断链故障;Wang Zisheng等[13]采用Plackett−Burman试验确定刮板输送机链条裂纹深度、初始角度和拉伸载荷,为避免刮板输送机断链故障提供了参考。

    煤矿井下工作面环境复杂恶劣,常规在刮板输送机中部槽或刮板上安装传感器检测断链的方法常出现传感器损坏现象,因此,基于视频监控的断链监测方法得到越来越多的应用。早期的煤矿井下视频监控技术主要是将井下各监控点的图像传输至地面中心监控室显示,并对设备参数异常、设备工作状态等情况进行人工识别。随着AI技术的快速发展,视频AI识别逐渐被引入煤矿井下安全监控领域。基于视频AI识别技术的刮板输送机断链监测技术[14-18]以AI算法为核心,依据视频AI摄像仪对井下刮板输送机断链状态进行实时监控。此类算法适应性强,能够解决复杂的非线性问题,但对数据样本集的要求较高,在线学习能力差,导致断链监测准确率和精确率相对较低。此外,现有技术通过采集样本数据进行离线算法训练,导致算法在陌生环境中适应性差、检测精度和鲁棒性不足。因此,提出一种基于在线贯序极限学习机(Online Sequential Extreme Learning Machine,OSELM)网络的刮板输送机断链智能监测技术。OSELM网络由极限学习机(Extreme Learning Machine,ELM)改进而来,采用增量式在线学习算法,通过分析样本数据集的自相关性和输入与输出关系,自动随机确定网络隐含层节点数量,不需要复杂的超参数调优和迭代优化过程,具备训练速度快、精度高、泛化能力强等优点。此外,OSELM网络能够更好地适应流式样本数据的训练,进而对网络输出权重进行更新迭代,使网络始终保持在最优状态,在处理不断变化的样本数据时具有显著优势。

    ELM是一类单隐含层的FNN(Feedforward Neuron Network,前馈神经网络)[19-20]。ELM网络拓扑如图1所示,其包含n个输入层节点、L个隐含层节点和m个输出层节点。

    图  1  ELM网络拓扑
    Figure  1.  Topology of ELM network

    ELM网络与传统人工智能网络不同,它随机获取神经网络输入层权值和隐含层偏置,利用最小二乘法准则,通过计算穆尔−彭罗斯广义逆矩阵得出网络输出权值,具有训练速度快、学习误差小等优势,且泛化性能极高[21-23]

    ELM网络在学习训练过程中随机获取样本{(xiti)},其中xi为样本输入,ti为样本输出,i=1,2$ ,\cdots, $n。通过ELM网络隐含层S型激活函数g(·)计算隐含层输出函数:

    $$ {h}_{k}\left({x}_{i}\right)=g\left({{\boldsymbol{w}}}_{k},{{\boldsymbol{b}}}_{k},{x}_{i}\right)=g\left({{\boldsymbol{w}}}_{k} {x}_{i}+{{\boldsymbol{b}}}_{k}\right) $$ (1)

    式中:wk为第k个隐含层节点的输入权重矩阵,k=1,2$,\cdots , $Lbk为第k个隐含层节点偏置矩阵。

    ELM网络模型可表示为

    $$ {t}_{i}={\displaystyle \sum _{k=1}^{L}{{\boldsymbol{\beta}} }_{k}g\left({{\boldsymbol{w}}}_{k} {{{x}}}_{i}+{{\boldsymbol{b}}}_{k}\right)} $$ (2)

    式中βk为ELM网络第k个隐含层节点的输出权重矩阵。

    将训练ELM网络转换为求解一个期望输出线性矩阵系统问题,表示为

    $$ {\boldsymbol{T}} = {\boldsymbol{H\beta}} $$ (3)

    式中:T为ELM网络输出矩阵;H为隐含层输出矩阵;β为ELM网络输出权重矩阵。

    为了最小化所有样本的整体预测误差,ELM网络通过最小二乘法计算网络输出权重矩阵β

    $$ {\boldsymbol{\beta}} {\text{ = }}{{\boldsymbol{H}}^\dagger }{\boldsymbol{T}} = {\left( {{{\boldsymbol{H}}^{\text{T}}}{\boldsymbol{H}}} \right)^{ - 1}}{{\boldsymbol{H}}^{\text{T}}}{\boldsymbol{T}} $$ (4)

    式中${{\boldsymbol{H}}^\dagger }$为ELM网络输出矩阵H的穆尔−彭罗斯广义逆矩阵。

    传统ELM网络不能实时处理动态模型,对此,提出能产生大量序列化数据的在线增量生长式ELM网络,即OSELM网络。其将单隐含层神经网络输出权重的学习训练过程分为2个阶段:① 初始化阶段,通过海量刮板输送机断链故障状态的样本训练得到网络输出权重矩阵β。② 序列化阶段,利用序列化在线样本数据集不断完善β

    将获取的刮板输送机断链离线样本进行高精度可靠筛选,得到高质量刮板输送机断链离线样本。将离线样本输入OSELM网络进行初始化训练,同时将学习到的知识和经验存储到隐含层节点。OSELM网络根据离线样本特征随机获取隐含层节点的输入权重矩阵wk及偏置矩阵bk,结合广义逆矩阵计算方法,对批量离线样本数据进行训练,计算出初始化的网络输出权重矩阵β0

    $$ {{\boldsymbol{\beta}} ^0} = {\left( {{{\boldsymbol{H}}_0}^{\text{T}}{{\boldsymbol{H}}_0}} \right)^{ - 1}}{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} $$ (5)

    式中:H0为初始化的隐含层输出矩阵;T0为初始化的网络输出矩阵。

    β0作为OSELM网络序列化迭代阶段的初始权重,进一步完善网络。

    定义隐含层输出矩阵为

    $$ {\boldsymbol{H}} = \left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{H}}_0}} \\ {{{\boldsymbol{H}}_1}} \end{array}} \right] $$ (6)

    式中H1为第1次迭代后的隐含层输出矩阵,为已知量。

    定义ELM网络输出矩阵为

    $$ {\boldsymbol{T}} = \left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{T}}_0}} \\ {{{\boldsymbol{T}}_1}} \end{array}} \right] $$ (7)

    式中T1为第1次迭代后的网络输出矩阵,为已知量。

    因此,ELM网络输出权重矩阵更新为

    $$ \begin{split} {\boldsymbol{\beta}} =& {\left( {{{\boldsymbol{H}}^{\text{T}}}{\boldsymbol{H}}} \right)^{ - 1}}{{\boldsymbol{H}}^{\text{T}}}{\boldsymbol{T}}= \\ &{\left\{ {{{\left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{H}}_0}} \\ {{{\boldsymbol{H}}_1}} \end{array}} \right]}^{\text{T}}}\left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{H}}_0}} \\ {{{\boldsymbol{H}}_1}} \end{array}} \right]} \right\}^{ - 1}}{\left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{H}}_0}} \\ {{{\boldsymbol{H}}_1}} \end{array}} \right]^{\text{T}}}\left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{T}}_0}} \\ {{{\boldsymbol{T}}_1}} \end{array}} \right]= \\ & {\left( {{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} \right)^{ - 1}}\left( {{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{T}}_1}} \right) \end{split} $$ (8)

    P0=(${{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0} $)−1P1=(${{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0} $+${{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1} $)−1,可得到P1P0的迭代计算公式:

    $$ {{\boldsymbol{P}}_1} = {\left( {{{\boldsymbol{P}}_0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} \right)^{ - 1}} $$ (9)

    根据Sherman−Morrison矩阵求逆公式[24],可将式(9)简化为

    $$ {{\boldsymbol{P}}_1} = {{\boldsymbol{P}}_0} - \frac{{{{\boldsymbol{P}}_0}{{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}{{\boldsymbol{P}}_0}}}{{1 + {{\boldsymbol{H}}_1}{{\boldsymbol{P}}_0}{{\boldsymbol{H}}_1^{\text{T}}}}} $$ (10)

    将式(9)两边同时求逆,得

    $$ {{\boldsymbol{P}}_1^{ - 1} }= {{\boldsymbol{P}}_0^{ - 1}} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1} $$ (11)

    通过式(11)可推导出${{\boldsymbol{P}}_0^{ - 1}} $计算公式:

    $$ {{\boldsymbol{P}}_0^{ - 1}} = {{\boldsymbol{P}}_1^{ - 1}} - {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1} $$ (12)

    P0=(${{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0} $)−1代入式(5),得

    $$ {{\boldsymbol{\beta}} ^0} = {\left( {{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0}} \right)^{ - 1}}{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} = {P_0}{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} $$ (13)
    $$ {{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} = {{\boldsymbol{P}}_0^{ - 1}}{{\boldsymbol{\beta}} ^0} = \left( {{{\boldsymbol{P}}_1^{ - 1}} - {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} \right){{\boldsymbol{\beta}} ^0} $$ (14)

    P1=(${{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0}} $+${{{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} $)−1和式(14)同时代入式(8),得到第1次迭代后的网络输出权重矩阵:

    $$\begin{split} {{\boldsymbol{\beta}} ^1} = &{\left( {{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{H}}_0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} \right)^{ - 1}}\left( {{{\boldsymbol{H}}_0^{\text{T}}}{{\boldsymbol{T}}_0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{T}}_1}} \right) = \\ &{{\boldsymbol{P}}_1}\left[ {\left( {{{\boldsymbol{P}}_1^{ - 1}} - {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{H}}_1}} \right){{\boldsymbol{\beta}} ^0} + {{\boldsymbol{H}}_1^{\text{T}}}{{\boldsymbol{T}}_1}} \right] = \\ &{{\boldsymbol{\beta}} ^0} + {{\boldsymbol{P}}_1}{{\boldsymbol{H}}_1^{\text{T}}}\left( {{{\boldsymbol{T}}_1} - {{\boldsymbol{H}}_1}{{\boldsymbol{\beta}} ^0}} \right) \end{split} $$ (15)

    定义OSELM网络第k+1次迭代的网络输出权重矩阵为βk+1,中间参数矩阵为Pk+1。根据OSELM网络在线迭代关系,可由第k次迭代参数计算出第k+1次迭代参数,在线学习递推公式为

    $$ {{\boldsymbol{P}}_{k + 1}} = {{\boldsymbol{P}}_k} - \frac{{{{\boldsymbol{P}}_k}{{\boldsymbol{H}}_{k + 1}^{\text{T}}}{{\boldsymbol{H}}_{k + 1}}{{\boldsymbol{P}}_k}}}{{1 + {{\boldsymbol{H}}_{k + 1}}{{\boldsymbol{P}}_k}{{\boldsymbol{H}}_{k + 1}^{\text{T}}}}}\qquad $$ (16)
    $$ {{\boldsymbol{\beta}} ^{k + 1}} = {{\boldsymbol{\beta}} ^k} + {{\boldsymbol{P}}_{k + 1}}{{\boldsymbol{H}}_{k + 1}^{\text{T}}}\left( {{{\boldsymbol{T}}_{k + 1}} - {{\boldsymbol{H}}_{k + 1}}{{\boldsymbol{\beta}} ^k}} \right) $$ (17)

    OSELM网络初始化阶段训练流程如下。

    1) 获取样本数据集D={(xi, ti)},在其中选取n0n0L)个高可靠度采样数据并组成集合D0={(xj, tj)},j=1, 2$, \cdots ,n_0 $,将其输入OSELM网络。

    2) OSELM网络随机获取隐含层节点的输入权重矩阵wk和偏置矩阵bk,并计算初始化的隐含层输出矩阵H0

    3) 计算初始化的网络输出权重矩阵β0

    序列化阶段训练流程如下。

    1) 通过在线学习,训练数据样本(xi+1, ti+1)。

    2) 计算在线学习数据样本的隐含层输出矩阵Hk+1

    3) 计算OSELM网络的输出权重矩阵βk+1

    与多数在线生长型人工神经网络模型相比,OSELM网络具有参数少、训练速度快和在线学习泛化性能强等优势。

    设计OSELM网络算法包含5个有限元模型,定义为

    $$ \text{OSELM}=\langle g(\cdot ),{\boldsymbol{H}},{\boldsymbol{P}},{\boldsymbol{T}},{\boldsymbol{\beta}} \rangle $$ (18)

    式(18)中,隐含层激活函数g·)由系统模型和外界环境决定;中间参数PPk×Hk+1Pk+1,其第k+1次迭代的计算结果Pk+1总是由第k次的中间参数Pk和OSELM网络第k+1次隐含层输出矩阵Hk+1共同决定;网络输出权重矩阵ββk×Pk+1×Hk+1×Tk+1βk+1,其第k+1次迭代的计算结果βk+1总是由OSELM网络第k次的输出权重矩阵βkk+1次中间参数Pk+1和第k+1次网络输出矩阵Hk+1三者共同决定。

    OSELM网络框架如图2所示。将采集的刮板输送机断链离线样本和AI摄像仪实时在线感知图像输入OSELM网络,输出为AI摄像仪的决策信息。

    图  2  OSELM网络框架
    Figure  2.  OSELM network framework

    采用离线样本对构建的OSELM网络进行训练,进而构建刮板输送机断链识别模型。试验硬件配置为13th Gen Intel(R) Core(TM) i9−13900K CPU @ 3.00 GHz处理器、12 GiB的NVIDIA RTX A2000GPU、Windows10操作系统,编程语言采用Python 3.10.1,开发环境为PyCharm。

    样本数据集来自兖矿能源集团股份有限公司金鸡滩煤矿、中国华能集团有限公司高头窑煤矿、淮南矿业(集团)有限责任公司丁集煤矿等大型综采(放)工作面,由隔爆兼本安型高清摄像仪采集。为提升刮板输送机断链识别效果,使用LabelImg对样本数据集进行标注,得到6 952张有效图像,按照7∶1∶2的比例划分为训练集、测试集和验证集,用于网络训练和性能评估。部分样本如图3所示。

    图  3  样本数据集(部分)
    Figure  3.  Sample dataset (partial)

    OSELM网络经离线样本训练和测试,其对刮板输送机断链状态识别的平均精度均值(Mean Average Precision,mAP)、准确率和精确率均达到90%以上,平均检测速度为183.5帧/s。

    基于OSELM的刮板输送机断链智能监测模型如图4所示。

    图  4  刮板输送机断链智能监测模型
    Figure  4.  Intelligent chain-broken monitoring model for scraper conveyor

    依托兖矿能源集团股份有限公司石拉乌素煤矿工作面配置的海康威视KBA18(D)型AI摄像仪进行井下工业性试验。AI摄像仪安装于刮板输送机机尾架,如图5所示。考虑煤矿井下环境复杂、恶劣,AI摄像仪时常会附着大量煤尘,影响摄像仪视觉清晰度,对AI摄像仪设置定时雨刷功能,可根据工作面不同的采煤工艺及工况自定义雨刷工作时间间隔,在特殊情况下可手动操作雨刷工作。同时,为避免补光灯光照强度对拍摄质量和监测准确度的影响,为AI摄像仪配置自动变焦、曝光度自动调节和强光抑制功能。

    图  5  刮板输送机断链智能监测AI摄像仪布置
    Figure  5.  AI camera deployment for broken chain monitoring system on scraper conveyor

    将经过离线样本训练的OSELM网络与AI摄像仪融合,进行序列化在线学习训练。在线学习过程中,AI摄像仪先从外界环境中随机获取刮板输送机链条当前状态集合,之后产生相应的判断。随着AI摄像仪采集的在线样本不断增多,OSELM网络可以获取到最佳的“状态−决策”集合,从而进一步完成自主认知发育学习。

    AI摄像仪实时采集的刮板输送机链条图像通过井下工业性千兆光纤环网上传至刮板输送机集中控制系统平台。在该平台主界面(图6)可显示刮板输送机断链监测的故障和通信信息,且具有断链监测的可视化界面,对断链识别结果进行全方位显示,如图7所示,蓝色框为采样识别区域,绿色和橙色锚框分别为左右两侧链环的识别状态和mAP。

    图  6  刮板输送机集中控制系统主界面
    Figure  6.  Main interface of centralized control system for scraper conveyor
    图  7  断链监测可视化界面
    Figure  7.  Visualization interface of broken chain monitoring

    采用文献[18-21]中的网络模型(分别为深度神经网络融合网络、RT−DETR、YOLOv5、YOLOv8)、ELM和OSELM网络进行可视化识别分析,结果如图8所示。可看出OSELM网络对刮板输送机断链和正常链环的识别准确度均高于对比模型。

    图  8  不同网络模型的断链识别可视化结果
    Figure  8.  Visualization results of broken chain identification using different networks

    从mAP50、准确率、精确率、检测速度4个指标方面,将OSELM网络与文献[18-22]所提网络模型、ELM和OSELM网络进行对比分析,结果见表1

    表  1  不同断链监测网络模型性能对比
    Table  1.  Performance comparison of different network models for broken chain monitoring
    模型 mAP50/% 准确率/% 精确率/% 检测速度/(帧·s−1
    文献[18] 93.4 96.5 87.1 206.3
    文献[19] 97.6 97.2 87.8 67.2
    文献[20] 92.7 94.8 85.7 64.9
    文献[21] 92.8 95.1 88.0 93.6
    文献[22] 75.8 93.4 43.8
    ELM 94.1 96.8 88.2 173.5
    OSELM 98.6 99.3 91.7 205.6
    下载: 导出CSV 
    | 显示表格

    表1可看出:OSELM网络的mAP50、准确率和精确率均处于较高水平,分别达98.6%,99.3%,91.7%,较ELM网络分别提高了4.5%,2.5%,3.5%;与文献[18-21]中网络模型和ELM相比,OSELM网络整体监测性能更高,主要原因是OSELM网络不仅能够通过断链离线样本信息进行训练,还能在线实时学习当前复杂场景的链条状态样本信息,而其他模型仅能依靠离线样本进行网络训练,存在较高的样本局限性;文献[22]中网络模型的断链识别精确率较OSELM网络高1.7%,但准确率较OSELM网络低23.5%,存在较大的目标识别误差;OSELM网络的检测速度达205.6帧/s,仅略低于深度神经网络融合网络,验证了OSELM网络在刮板输送机断链监测方面的高效性。

    理论方面,文献[18-22]中网络模型和ELM网络的训练和参数调优过程复杂,训练时间长,在实时训练方面性能较差。OSELM网络对增加的样本能够进行实时学习训练,通过上一个状态的网络输出权重,结合新增加的离线样本和在线样本,对网络输出权重进行更新迭代,不断优化网络输出权重,从而达到优化目标识别网络的目的。该网络的迭代参数种类较少,且可随新样本数据的到来不断更新而无需重新训练网络,使得网络模型始终保持在最新状态。

    工业性试验结果验证了基于OSELM网络的刮板输送机断链智能监测系统能准确识别刮板输送机链条断裂故障,未发生漏报和误报情况,表明OSELM网络在煤矿井下复杂环境中具有较高的自主学习能力及较强的泛化性和鲁棒性。

    1) OSELM网络在ELM网络基础上增加了在线训练模块,不仅能学习离线样本信息,还能在煤矿井下复杂场景中进行样本的实时在线训练,提高了刮板输送机断链识别模型的可靠性和泛化性。

    2) 工业性试验结果表明,OSELM网络的mAP50、准确率和精确率分别达98.6%,99.3%,91.7%,高于深度神经网络融合网络、RT−DETR、YOLOv5、YOLOv8、ELM等对比模型;检测速度达205.6帧/s,可满足实时性检测要求。

    3) 未来将重点优化OSELM网络针对刮板输送机断链的检测速度,并研究该网络在刮板输送机上煤矸识别和转载机危险区域人员误入检测方面的应用。

  • 图  1   巷道模型

    Figure  1.   Roadway model

    图  2   附壁风筒出风条示意图

    Figure  2.   Wall-attached duct outlet air strip seam

    图  3   不同Lr条件下各断面内风速矢量分布

    Figure  3.   Wind speed vector distribution in each section under different radial air outlet distances Lr conditions

    图  4   不同φ条件下综掘工作面风流分布

    Figure  4.   Air flow distribution of excavation face under different ratios of radial air outlet φ conditions

    图  5   不同Lrφ条件下综掘机司机呼吸带处粉尘质量浓度分布

    Figure  5.   Distribution of dust mass concentration in the breathing zone of excavator driver under different Lr and φ conditions

    图  6   不同Lr条件下φLd之间拟合曲线及拟合公式

    Figure  6.   The fitting curve and formula between φ and Ld under different Lr conditions

    图  7   不同La条件下综掘机司机呼吸带处速度云图

    Figure  7.   Speed cloud at the breathing zone of excavator driver under different axial outlet distance La conditions

    图  8   不同La条件下综掘机司机呼吸带处粉尘质量浓度分布

    Figure  8.   Distribution of dust mass concentration at the breathing zone of the driver under different La conditions

    表  1   不同Lrφ条件下粉尘扩散距离Ld

    Table  1   Dust diffusion distance Ld under different Lr and φ condition

    Lr/m Ld/m
    φ=0.6 φ=0.7 φ=0.8 φ=0.9
    10 9.1 8.9 9.8 10.5
    15 8.3 7.4 7.1 8.2
    20 9.4 8.1 7.0 6.4
    25 12.7 11.6 10.1 6.8
    下载: 导出CSV

    表  2   各断面测点风速

    Table  2   Wind speed at measuring points of each section

    距工作面距离/m A点风速/(m·s−1 B点风速/(m·s−1
    实测值 模拟值 实测值 模拟值
    5 0.48 0.54 0.39 0.45
    10 0.43 0.51 0.36 0.43
    20 1.16 1.32 0.87 0.97
    下载: 导出CSV

    表  3   各断面测点粉尘质量浓度

    Table  3   Dust concentration measuring value of each section

    距工作面距离/m 粉尘质量浓度/(mg·m−3
    3 208.8
    5 63.3
    7 33.6
    下载: 导出CSV
  • [1] 龚晓燕,赵晓莹,杨富强,等. 综掘面尘源动态变化下粉尘场优化的风流调控研究[J]. 煤炭工程,2021,53(2):122-126.

    GONG Xiaoyan,ZHAO Xiaoying,YANG Fuqiang,et al. Wind flow control for optimization of dust field distribution under dynamic change of dust source in fully mechanized heading face[J]. Coal Engineering,2021,53(2):122-126.

    [2] 秦翥. 带式输送机转载点粉尘浓度分布数值模拟[J]. 工矿自动化,2018,44(7):70-74.

    QIN Zhu. Numerical simulation of dust concentration distribution in transfer site of belt conveyor[J]. Industry and Mine Automation,2018,44(7):70-74.

    [3] 龚晓燕,张浩,陈龙,等. 综掘面抽风口调控下的旋流风幕降尘优化[J]. 中国安全科学学报,2023,33(6):56-63.

    GONG Xiaoyan,ZHANG Hao,CHEN Long,et al. Optimization of cyclone air curtain dust control under control of exhaust outlet on fully mechanized excavation face[J]. China Safety Science Journal,2023,33(6):56-63.

    [4] 程卫民,周刚,陈连军,等. 我国煤矿粉尘防治理论与技术20年研究进展及展望[J]. 煤炭科学技术,2020,48(2):1-20.

    CHENG Weimin,ZHOU Gang,CHEN Lianjun,et al. Research progress and prospect of dust control theory and technology in China's coal mines in the past 20 years[J]. Coal Science and Technology,2020,48(2):1-20.

    [5] 吕英华,察兴鹏,聂文,等. 基于数值模拟的喷雾装置粉尘抑制研究[J]. 煤矿机械,2022,43(11):34-37.

    LYU Yinghua,CHA Xingpeng,NIE Wen,et al. Study on dust suppression of spray device based on numerical simulation[J]. Coal Mine Machinery,2022,43(11):34-37.

    [6] 张义坤. 煤矿综掘面泡沫降尘技术研究与实施[J]. 中国安全科学学报,2012,22(2):151-156.

    ZHANG Yikun. Study on dust control technology with foam in fully mechanized workface of Huoerxinhe Coal Mine[J]. China Safety Science Journal,2012,22(2):151-156.

    [7] 蒋仲安,杨斌,张国梁,等. 高原矿井掘进工作面截割粉尘污染效应及通风控尘参数分析[J]. 煤炭学报,2021,46(7):2146-2157.

    JIANG Zhong'an,YANG Bin,ZHANG Guoliang,et al. Analysis of dust pollution effect of cutting dust and ventilation control parameters at the heading face in plateau mines[J]. Journal of China Coal Society,2021,46(7):2146-2157.

    [8] 秦跃平,张苗苗,崔丽洁,等. 综掘工作面粉尘运移的数值模拟及压风分流降尘方式研究[J]. 北京科技大学学报,2011,33(7):790-794.

    QIN Yueping,ZHANG Miaomiao,CUI Lijie,et al. Numerical simulation of dust migration and study on dust removal modes with the forced ventilation shunt in a fully mechanized workface[J]. Journal of University of Science and Technology Beijing,2011,33(7):790-794.

    [9] 聂文,程卫民,周刚. 综掘工作面压风气幕形成机理与阻尘效果分析[J]. 煤炭学报,2015,40(3):609-615.

    NIE Wen,CHENG Weimin,ZHOU Gang. Formation mechanism of pressure air curtain and analysis of dust suppression's effects in mechanized excavation face[J]. Journal of China Coal Society,2015,40(3):609-615.

    [10]

    LIU Qiang,NIE Wen,HUA Yun,et al. The effects of the installation position of a multi-radial swirling air-curtain generator on dust diffusion and pollution rules in a fully-mechanized excavation face:a case study[J]. Powder Technology,2018,329:371-385. DOI: 10.1016/j.powtec.2018.01.064

    [11] 陈芳,张设计,马威,等. 综掘工作面压风分流控除尘技术研究与应用[J]. 煤炭学报,2018,43(增刊2):483-489.

    CHEN Fang,ZHANG Sheji,MA Wei,et al. Research and application of the technology of forced ventilation diversion to control and reduce dust in fully mechanized excavation face[J]. Journal of China Coal Society,2018,43(S2):483-489.

    [12] 王建国,王康,樊亦洋. 附壁风筒条缝参数对综掘工作面控尘效果的影响[J]. 煤矿安全,2021,52(6):206-210.

    WANG Jianguo,WANG Kang,FAN Yiyang. Influence of slot parameters of wall-attached air duct on dust control effect of fully mechanized mining face[J]. Safety in Coal Mines,2021,52(6):206-210.

    [13] 王昊,张永亮,撒占友,等. 综掘工作面压抽风量比对多径向涡旋气幕阻尘效果的影响[J]. 矿业安全与环保,2022,49(2):117-121,126.

    WANG Hao,ZHANG Yongliang,SA Zhanyou,et al. Effect of the forced-suction airflow volume ratio on the dust suppression of swirling air curtain in fully mechanized excavation face[J]. Mining Safety & Environmental Protection,2022,49(2):117-121,126.

    [14] 龚晓燕,彭高高,宋涛,等. 掘进工作面长压短抽通风出风口风流调控参数研究[J]. 工矿自动化,2021,47(9):45-52.

    GONG Xiaoyan,PENG Gaogao,SONG Tao,et al. Study on air flow control parameters of long-pressure and short-extraction ventilation air outlets in heading face[J]. Industry and Mine Automation,2021,47(9):45-52.

    [15] 陈芳. 综掘工作面分段控风控尘技术研究与应用[J]. 煤炭工程,2019,51(12):106-110.

    CHEN Fang. Research and application of sectional airflow and dust control technology in fully mechanized tunnelling face[J]. Coal Engineering,2019,51(12):106-110.

    [16] 程卫民,王昊,孙彪,等. 综掘面径向分风与压风配比对风幕阻尘的影响[J]. 中国矿业大学学报,2017,46(5):1014-1023.

    CHENG Weimin,WANG Hao,SUN Biao,et al. Effects of ratio of radial partial air volume to pressure air volume on air curtain dust control at fully mechanized working face[J]. Journal of China University of Mining & Technology,2017,46(5):1014-1023.

    [17] 韩敏,王建国,王康. 多抽风筒对综掘面除尘的影响研究[J]. 矿业安全与环保,2022,49(5):114-118.

    HAN Min,WANG Jianguo,WANG Kang. Study on the influence of multiple exhaust pipes on dust removal in fully mechanized excavation face[J]. Mining Safety & Environmental Protection,2022,49(5):114-118.

    [18] 王昊,撒占友,王春源,等. 综掘工作面通风条件对径向旋流风幕阻尘效果的影响[J]. 煤矿安全,2022,53(3):186-192.

    WANG Hao,SA Zhanyou,WANG Chunyuan,et al. Influences of ventilation conditions of fully mechanized excavation face on dust-resistance effect of radial swirling air curtain[J]. Safety in Coal Mines,2022,53(3):186-192.

    [19] 杨泽安,王义亮. 综掘工作面湿式除尘条件下尘雾耦合数值模拟[J]. 煤矿安全,2021,52(11):170-175.

    YANG Ze'an,WANG Yiliang. Numerical simulation of dust and spray coupling under wet dust removal condition in fully mechanized excavation face[J]. Safety in Coal Mines,2021,52(11):170-175.

    [20] 魏星,高丹红,张国宝. 旋流气幕系统对综掘面控尘流场的影响研究[J]. 矿业研究与开发,2019,39(12):109-115.

    WEI Xing,GAO Danhong,ZHANG Guobao. The impact of swirling air curtain system on the dust control flow field in the fully-mechanized excavation face[J]. Mining Research and Development,2019,39(12):109-115.

    [21] 王磊,程煜,徐荣萧. 基于EDEM仿真的煤岩截割产尘规律研究[J]. 煤矿安全,2023,54(1):22-28.

    WANG Lei,CHENG Yu,XU Rongxiao. Mechanism of dust production during coal and rock cutting based on EDEM[J]. Safety in Coal Mines,2023,54(1):22-28.

    [22] 邹常富. 岩巷产尘规律及控风除尘技术研究[J]. 煤炭技术,2021,40(4):77-79.

    ZOU Changfu. Research on dust generation law of rock drift and wind control and dust removal technology[J]. Coal Technology,2021,40(4):77-79.

    [23] 侯树宏,郝军,李腾龙,等. 综掘工作面通风控尘参数匹配关系研究及应用[J]. 煤炭技术,2022,41(1):166-169.

    HOU Shuhong,HAO Jun,LI Tenglong,et al. Research and application of ventilation and dust control parameters matching in fully mechanized driving face[J]. Coal Technology,2022,41(1):166-169.

    [24] 张义坤,蒋仲安,孙雅茹. 附壁风筒对掘进工作面通风除尘的促进作用[J]. 煤矿安全,2017,48(12):161-163.

    ZHANG Yikun,JIANG Zhong'an,SUN Yaru. Promotion effect of dust removal in heading face by ventilation duct with coanda effect[J]. Safety in Coal Mines,2017,48(12):161-163.

  • 期刊类型引用(9)

    1. 毛清华,胡鑫,王孟寒,张旭辉,薛旭升. 改进YOLOv5s的采煤机滚筒与支架护帮板干涉状态智能识别. 煤炭科学技术. 2024(02): 253-263 . 百度学术
    2. 王新航. 煤矿安全目标检测技术研究进展. 能源与节能. 2023(04): 118-120 . 百度学术
    3. 陈立烨,党浚哲,崔子航,陈思妍,段琦锋. 基于FocalLoss的煤矿井下人员安全视觉监测方法. 现代信息科技. 2023(13): 96-100 . 百度学术
    4. 刘丁源,冯锋. 矿井下人员安全检测技术分析. 信息与电脑(理论版). 2022(15): 41-43+47 . 百度学术
    5. 魏智锋,肖书浩,蒋国璋,伍世虔,程国飞. 基于深度学习的人造板表面缺陷检测研究. 林产工业. 2021(02): 21-26 . 百度学术
    6. 董昕宇,师杰,张国英. 基于参数轻量化的井下人体实时检测算法. 工矿自动化. 2021(06): 71-78 . 本站查看
    7. 杨建伟,涂兴子,梅峰漳,李亚宁,范鑫杰. 基于深度学习优化YOLOV3算法的芳纶带检测算法研究. 中国矿业. 2020(04): 67-72 . 百度学术
    8. 孙杰臣,李敬兆,王继伟,许志. 煤矿井下人车联动控制系统. 工矿自动化. 2020(12): 7-12 . 本站查看
    9. 张春堂,管利聪. 基于SSD-MobileNet的矿工安保穿戴设备检测系统. 工矿自动化. 2019(06): 96-100 . 本站查看

    其他类型引用(18)

图(8)  /  表(3)
计量
  • 文章访问数:  286
  • HTML全文浏览量:  83
  • PDF下载量:  61
  • 被引次数: 27
出版历程
  • 收稿日期:  2023-06-03
  • 修回日期:  2023-12-24
  • 网络出版日期:  2024-01-30
  • 刊出日期:  2024-01-30

目录

/

返回文章
返回