干扰背景下煤机故障声音识别算法

Voice recognition of coal machine fault under interference background

摘要

煤机设备的稳定运行决定了煤矿生产效率与作业安全。然而,在实际工况中,复杂且强烈的背景噪声严重干扰故障声音特征的提取,导致传统识别方法准确率下降。针对这一问题,提出了一种融合生成对抗网络(generative adversarial network,RaGAN)与多通道轻量级移动端卷积神经网络(mobile convolutional neural network, Mc-MobileNet)的煤机故障声音识别方法。首先,针对原始故障声音信号信噪比较低的问题,引入改进的RaGAN对故障声音进行增强处理,通过构建生成器与判别器的对抗学习机制,有效抑制环境噪声,提高目标信号的清晰度与可分辨性。同时,在模型中引入辅助分类器以增强判别能力,并结合循环一致性损失约束生成过程,从而提升生成样本的真实性与稳定性。其次,将增强后的音频特征输入到改进的Mc-MobileNet模型中进行分类识别,该模型在保持较低计算复杂度的同时,具备较强的Mc-MobileNet能力和良好的实时性能,从而提高了系统在复杂环境下的鲁棒性。最终,构建了一个端到端的煤机故障声音识别系统,实现从数据增强到故障分类的整体优化。实验结果表明,与传统声音增强与识别方法相比,本文提出的RaGAN方法能够显著提高分类准确率约10%,在嘈杂环境下具有更好的性能表现,可有效用于煤矿等噪声环境下的故障检测。

Abstract

The stable operation of coal mining machinery equipment is of great significance to coal mine production efficiency and operational safety. However, in actual working conditions, complex and intense background noise severely interferes with the extraction of fault sound features, leading to a decrease in the accuracy of traditional recognition methods. To address this issue, this paper proposes a coal mining machinery fault sound recognition method that integrates the generative adversarial network (RaGAN) and the multi-channel lightweight mobile convolutional neural network (Mc-MobileNet). Firstly, to address the issue of low signal-to-noise ratio in the original fault sound signals, an improved RaGAN is introduced to enhance the fault sounds. By constructing an adversarial learning mechanism between the generator and the discriminator, environmental noise is effectively suppressed, enhancing the clarity and distinguishability of the target signals. Simultaneously, an auxiliary classifier is introduced into the model to enhance the discriminative ability, and the cyclic consistency loss is incorporated to constrain the generation process, thereby improving the authenticity and stability of the generated samples. Secondly, the enhanced audio features are input into an improved Mc-MobileNet model for classification and recognition. This model maintains low computational complexity while possessing strong feature extraction capabilities and excellent real-time performance, thus enhancing the system′s robustness in complex environments. Finally, an end-to-end coal mining machinery fault sound recognition system is constructed, achieving overall optimization from data enhancement to fault classification. Experimental results show that, compared to traditional sound enhancement and recognition methods, the proposed method significantly improves recognition performance in noisy environments, with a classification accuracy increase of approximately 10%. Additionally, it exhibits high computational efficiency and good generalization ability. This method provides an effective technical approach for intelligent detection of equipment faults in complex noise environments in coal mines, and holds strong engineering application value.

关键词

煤机故障声音 / 干扰背景 / 生成对抗网络 / Mc-MobileNet / 声音识别 / 故障检测

Key words

coal machine malfunction sound / interference background / generate adversarial networks / Mc-MobileNet / voice recognition / fault detection

[1] 李永成. 基于AI智能识别的带式输送机故障诊断技术研究[J]. 自动化应用, 2024, 65(7):33-35, 38.
[2] 彭嘉宁, 房娟, 杨家麒. 基于并行遗传算法的含风电场电力能源调度系统设计[J]. 自动化技术与应用, 2024, 43(12):75-79.
[3] 陈亦寒, 杨静泊, 俞家融, 等. 基于全卷积的井下作业安全自动监测方法研究[J]. 自动化技术与应用, 2025, 44(6):160-165.
[4] 问姝雅, 卓旭升, 吴尔夫. 基于聚类分析的磨煤机断煤过渡状态识别[J]. 自动化与仪表, 2021, 36(11):31-34.
[5] 姬剑中. 采煤机故障监测与识别系统设计研究[J]. 能源与节能, 2022(2):77-79.
[6] 冯小峰, 冯霞山, 张正峰, 等. 基于最大似然法和决策树的智能电能表计量误差检测方法[J]. 电测与仪表, 2024, 61(12):205-211.
[7] 汪志成, 王泽旺, 朱梦帆, 等. 基于卷积神经网络的局部放电声音识别研究[J]. 电子测量技术, 2023, 46(20):148155.
[8] 郎巨林, 郑晟. 基于CNN-RNN集成的隧道事故异常声音识别[J]. 电子测量技术, 2023, 46(20):164-169.
[9] 曹正远, 蒋伟, 方成辉. 基于双注意力生成对抗网络的煤流异物智能检测方法[J]. 工矿自动化, 2023, 49(12):56-62.
[10] 冯小峰, 冯霞山, 张正峰, 等. 基于最大似然法和决策树的智能电能表计量误差检测方法[J]. 电测与仪表,2024, 61(12):205-211.
[11] 胡嘉欣, 田军. 基于生成对抗网络的数字音频信号多声道增强方法[J]. 现代电子技术, 2023, 46(19):41-44.
[12] 陈习坤, 杨俊美. 基于离散小波包变换与胶囊生成对抗网络的语音超分辨率算法[J]. 电子学报, 2023, 51(4):1039-1049.
[13] 易强, 李成娟, 李宝清, 等. 基于改进Mobile-Net V1网络的野外车辆识别[J]. 工业控制计算机, 2020, 33(7):104-107.
[14] 邵伟平, 王兴, 曹昭睿, 等. 基于MobileNet与YOLOv3的轻量化卷积神经网络设计[J]. 计算机应用, 2020, 40(增刊1):8-13.
[15] 王红君, 刘紫宾, 赵辉, 等. 基于改进YOLOv5的苹果轻量化检测算法[J]. 农机化研究, 2025, 47(7):65-71.

基金资助

国家重点研发计划项目(2022YFB4703600)