输气管道微泄漏检测的深度神经网络模型研究  

Research on Deep Neural Network Model for Small Leakage Detection of Gas Transportation Pipes

作　　者：利节[1] 高铮[1] 陆猛[1] 李莉[1] 夏煦坤 

机构地区：[1]重庆科技学院电气与信息工程学院,重庆401331

出　　处：《重庆科技学院学报（自然科学版）》2016年第5期66-70,共5页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：重庆市教委科学技术研究项目"基于深度学习的加气站管道微泄漏检测方法研究"(KJ1501305);重庆科技学院校内科研基金"天然气管道输送安全事故的预测模型"(CK2015B17)

摘　　要：建立了基于声音信号的深度神经网络模型,用于解决半导体制造业气体输送管道的微泄漏检测问题;设计了一种声音信号的预处理方法,用于去噪和特征提取,将预处理后的声音信号作为模型输入,该模型由7层神经元构成,通过对声音信号的训练和学习,获得管道微泄漏的概率。大量仿真实验结果表明,与半导体企业现采用的流量和压力传感器检测方法相比,该模型能更精确、更快速地实现管道微泄漏检测;与常规神经网络相比,该模型能更有效且稳定地进行检测。A deep neural network model （DNN） based on sound signal is proposed to solve the small leakage detec- tion problem of gas transportation pipes in the semiconductor manufacturing. To avoid noise and gain the character, a pre -process model of sound signal is presented. The solved sound signal is the input of our DNN, where the structure is combined with seven layers of neurons. The sound signals are trained and learned for getting the correct probability of small leakage of pipes. The simulation results show our model is more precise and quick than the standard sensor detection. Meantime, it is more stable than the standard neural network.

关 键 词：微泄漏 半导体 声音信号 神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]