煤矿采空区充填体的损伤监测直接关系到采矿安全问题，充填体的破裂裂隙定位技 术逐渐受到关注。传统的裂隙定位方法在计算复杂度、精度和实时性等方面存在明显局 限性，严重影响矿山开采进展。为了提高矿山开采安全性和掘进可行性，设计并实现一 款胶结充填体的声发射源定位器件。主要研究工作内容如下： （1）针对声发射信号微弱、采矿作业噪声大、岩体破裂噪声强等问题，开发了一种 基于双核 RISC-V 处理器的声发射定位系统。通过仿真实验设计了一种声发射信号处理 电路，该电路针对声发射信号具有反向比例运算电路的最优增益，运用双向 TVS 二极管 钳位输入信号幅度，将输入、输出调整为差分信号，实现了对上述充填试块中的声发射 信号低噪声、高精度放大的目的。 （2）针对传统拾取算法中拾取效率低、波形畸变严重、多源干扰强等问题，设计了 ResDAU-Net 网络实现对充填试块破裂过程中产生的声发射信号初至到时准确拾取的目 的。该网络基于 U-Net 网络架构，将采样部分用残差单元代替，并且在跳跃结构中嵌入 CAM 和 PAM 双注意力机制模块，形成 9 层网络结构。编码阶段提取声发射信号 P 波、 S 波的特征信息，解码阶段将提取结果恢复至原始数据大小。输入声发射信号波形数据， 输出 P 波、S 波和噪声的概率分布，通过最大概率点拾取初至时刻。该方法对比 AR-AIC、 参考阈值 STA/LTA，在满足不同绝对误差条件下，该算法初至到时数据量占比最高达到 93.5%。 （3）针对现有声发射源定位方法对传感器位置要求高、三维定位难、定位精度低等 问题，提出 HWOA-FAP 的定位算法，达到对不同种类充填试块裂隙定位的目的。分别 对比了 Beta 分布、Halton 序列、Logistic 混沌映射等初始化情况，Halton 序列初始化效 果最好。采用非线性收敛因子和自适应权重，提高前期收敛速度和后期局部收敛精准度。 实现了在边长为 20cm 立方体试块上平均误差小于 1cm 的声发射源三维定位。 （4）针对煤矿采空区充填体易破裂、易腐蚀和难监测等问题，配比出浓度为 72%抗 压强度在 4MPa 到 6MPa 的充填试块。通过在充填试块上折断铅笔芯模拟声发射源位置，采用 ResDAU-Net 网络对声发射信号进行拾取，其次通过 HWOA-FAP 算法反演计算出 声发射源位置坐标，达到了在不同形状的充填试块上对声发射源精准定位的目的。 本研究实现煤矿采空区充填体声发射信号实时拾取，充填体破裂裂隙定位，为矿山 开采过程提供有力的安全保障。

The damage monitoring of the filling body in the goaf of coal mines is directly related to the mining safety issue, and the fracture and fissure location technology of the filling body has gradually attracted attention. The traditional fracture location methods have obvious limitations in terms of computational complexity, accuracy and real-time performance, which seriously affect the progress of mine exploitation. In order to improve the safety of mine exploitation and the feasibility of tunneling, a localization device for acoustic emission sources of cemented filling bodies is designed and implemented. The main research work contents are as follows: (1)Aiming at the problems such as weak acoustic emission signals, large noise from mining operations, and strong noise from rock mass fracture, an acoustic emission positioning system based on a dual-core RISC-V processor has been developed. Through simulation experiments, an acoustic emission signal processing circuit was designed. This circuit has the optimal gain of the reverse proportional operation circuit for the acoustic emission signal. By using a bidirectional TVS diode to clamp the input signal amplitude, the input and output are adjusted to differential signals, achieving the purpose of low-noise and high-precision amplification of the acoustic emission signal in the above-mentioned filling test block. (2) Aiming at the problems such as low picking efficiency, severe waveform distortion and strong multi-source interference in the traditional picking algorithm, the ResDaO-NET network was designed to accurately pick up the initial arrival of the acoustic emission signal generated during the rupture process of the filling test block. This network is based on the U-Net network architecture. The sampling part is replaced by residual units, and the CAM and PAM dual attention mechanism modules are embedded in the jump structure to form a 9-layer network structure. In the encoding stage, the characteristic information of the P wave and S wave of the acoustic emission signal is extracted, and in the decoding stage, the extraction results arerestored to the original data size. Input the waveform data of the acoustic emission signal, output the probability distribution of P-waves, S-waves and noise, and pick up the initial arrival time through the maximum probability point. This method compares AR-AIC and the reference threshold STA/LTA. Under the condition of meeting different absolute errors, the proportion of data volume at the initial arrival of this algorithm can reach up to 93.5% at most. (3) Aiming at the problems of high requirements for sensor position, difficulty in threedimensional positioning and low positioning accuracy in the existing acoustic emission source positioning methods, a positioning method for optimizing the objective function by improving the whale optimization algorithm is proposed. The position coordinates of the acoustic emission source are inverted and calculated to achieve the purpose of locating the cracks of different types of filling test blocks. The initialization situations of Beta distribution, Halton sequence, Logistic chaotic map, etc. were compared respectively. The initialization effect of Halton sequence was the best. Nonlinear convergence factors and adaptive weights are adopted to improve the convergence speed in the early stage and the accuracy of local convergence in the later stage. Realize the three-dimensional positioning of acoustic emission sources with an average error within 1cm. (4) Aiming at issues such as susceptibility to fracture, corrosion, and monitoring challenges in backfill bodies within coal mine goafs, backfill specimens with 72% concentration and 4–6 MPa compressive strength were formulated. Artificial acoustic emission sources were simulated by snapping pencil leads on these specimens. The ResDAU-Net network was utilized to capture AE signals, followed by inversion calculation of AE source coordinates via the HWOA-FAP algorithm, thus attaining high-precision localization of acoustic emission sources on differently shaped backfill specimens. This study realizes the real-time pickup of acoustic emission signals of the filling body in the goaf of coal mines and the location of the fracture fissures of the filling body, providing a strong safety guarantee for the mining process of the mine.