基于ResNet-18的燃气发电机组故障声音识别技术研究  

Research on Fault Sound Recognition Technology of Gas Generator Sets Based on ResNet-18

作　　者：李奇 杨春雨 王立婷 陈洪刚 LI Qi;YANG Chunyu;WANG Liting;CHEN Honggang(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou,Jiangsu 221116,China;Zibo Zichai New Energy Co.,Ltd.,Zibo,Shandong 255000,China)

机构地区：[1]中国矿业大学信息与控制工程学院,江苏徐州221116 [2]淄博淄柴新能源有限公司,山东淄博255000

出　　处：《自动化应用》2025年第2期40-44,49,共6页Automation Application

摘　　要：提出一种燃气发电机组故障声音识别技术,首先对麦克风实时采集的声音信号进行时频域分析,生成时频图,然后利用训练过的ResNet-18卷积神经网络对时频图进行分类,判断其属于正常、嘈杂、不稳定、敲击这4种状态中的哪种。其实时性强,可用于实现在线的报警保护。A fault sound recognition technology for gas generator sets is proposed.Firstly,the real-time sound signal collected by the microphone is analyzed in the time-frequency domain to generate a time-frequency map.Then,a trained ResNet-18 convolutional neural network is used to classify the time-frequency map and determine which of the four states it belongs to:normal,noisy,unstable,and knocking.It has strong real-time performance and can be used to achieve online alarm protection.

关 键 词：ResNet-18 卷积神经网络 燃气发电机故障 声音识别 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]