高压釜泄漏声音的高频高阶空间交互识别算法研究  

High Frequency and High-Order Spatial Interaction Recognition Algorithm for Autoclave Leaking Voice

作　　者：李衍志 郭丽敏 张维国 古健 宗井彬 张凯 刘君[1] LI Yanzhi;GUO Limin;ZHANG Weiguo;GU Jian;ZONG Jingbing;ZHANG Kai;LIU Jun(China Enfi Engineering Corporation,Beijing 100038,China)

机构地区：[1]中国恩菲工程技术有限公司,北京100038

出　　处：《计算机测量与控制》2024年第10期169-174,共6页Computer Measurement &Control

基　　金：国家重点研发计划(2022YFB3304901)。

摘　　要：高压釜是湿法冶金领域常用的重要设备,存在危险气体泄漏的风险;同时,泄漏会导致高压釜内压不稳,严重时甚至引起爆炸,威胁设备和生产安全;针对高压釜泄漏导致的危险和威胁,提出了一种高压釜泄漏声音的高频高阶空间交互识别算法,用于监测高压釜泄漏发生时的声音,从而及时发现泄漏并采取相应的措施,确保设备和生产安全;该算法首先通过高通滤波器消除低频噪声对于识别结果的干扰,然后利用递归门控卷积块实现高频分量在高阶空间的交互,最后使用全卷积层识别高压釜泄漏的声音;实验结果表明,所提算法具有较好的高压釜泄漏识别效果,平均置信度达到0.93,以0.65作为置信度阈值时,识别准确率可达到99.5%;在处理速度上,算法能够每秒识别60个5秒长的音频文件,满足实时性的需求。An autoclave is important equipment commonly applied in the field of hydrometallurgy,which has a risk of hazardous gas leaks.Additionally,this leak leads to the pressure unstable in the autoclave,potentially causing explosions that threaten both the equipment and production safety.To address this issue,a high frequency high-order spatial interaction recognition algorithm based on the sound of high-pressure autoclave leaks is proposed.This algorithm is used to monitor the sounds associated with autoclave leaks,thus timely detect leaks,and make a decision to ensure equipment and production safety.Firstly,the low-frequency noise interference of the recognition results is eliminated through a high-pass filter.Then,a recursive gated convolutional block is used to achieve the interaction of high-frequency components in high-order spatial dimensions.Finally,a fully convolutional layer is utilized to recognize the sound of autoclave leaks.Experimental results demonstrate that the proposed algorithm achieves good recognition results for the leakage of the autoclave,with an average confidence level of 0.93.The recognition accuracy can reach up to 99.5%,with a confidence threshold of 0.65.During the processing speed,the algorithm can recognize 60 audio files per second,with each file lasting 5 seconds,meeting its real-time requirement.

关 键 词：湿-法冶金 高压釜 泄漏声音识别 递归门控卷积 

分 类 号：TG115.28[金属学及工艺—物理冶金]