基于声信号无线远传和声纹识别的制冷机运行及故障监控系统  

Refrigerator operation and fault monitoring system based on acoustic signal wireless remote transmission and voiceprint recognition

作　　者：陆正卿 方维岚 胡晓俊[1] 梁军汀[2] LU Zhengqing;FANG Weilan;HU Xiaojun;LIANG Junting(Shanghai Tobacco Group Co.,Ltd.,Shanghai 200082,China;Institute of Acoustics,Tongji University,Shanghai 200092,China)

机构地区：[1]上海烟草集团有限责任公司,上海200082 [2]同济大学声学研究所,上海200092

出　　处：《无损检测》2022年第2期73-77,共5页Nondestructive Testing

摘　　要：制冷机在工业企业中的使用极为广泛,为了能及时发现制冷机长期运行过程中因气缸、曲轴磨损及设备松动等引起的运行异常及故障,保证制冷机的长期、稳定运行,研制了一套基于声信号无线远传和声纹识别的制冷机运行及故障监控系统。系统采用通用分组无线业务(GPRS)移动通讯网络,将现场采集到的制冷机运转噪声信号无线远传给上位机,然后在上位机内采用Mel倒谱系数(MFCC)特征提取与长短时记忆(LSTM)神经网络相结合的声纹识别方法对接收到的声信号进行处理,进而对制冷机的运行状况及故障进行监控和识别。试验结果表明,采用MFCC与LSTM神经网络相结合的方法,可以有效地提高系统的识别率及诊断效果,具有良好的应用前景。Refrigerators are widely used in industrial enterprises. In order to timely find the operation abnormalities and faults caused by cylinder and crankshaft wear and equipment looseness during the long-term operation of the refrigerator, and ensure its long-term stable operation, a monitoring system for the refrigerator operation and fault based on wireless remote transmission of acoustic signal and voiceprint recognition is developed. The system uses GPRS mobile communication network to wirelessly transmit the refrigerator operation noise signal collected on site to the host computer, and then uses the voiceprint recognition method of Mel-frequency cepstral coefficient(MFCC) feature extraction and long short-term memory(LSTM) neural network to monitor and identify the operation status and faults of the refrigerator. The experimental results show that the combination of MFCC and LSTM neural network can effectively improve the recognition rate and diagnosis effect of the system, and has a good application prospect.

关 键 词：制冷机 MEL倒谱系数 长短时记忆 声纹识别 监控系统 

分 类 号：TG115.28[金属学及工艺—物理冶金]