摘要
针对微电机质量检测水平低、故障识别困难等问题,设计一种基于声学特征的微电机故障诊断方法。通过声音采集装置获得微电机转动时的正常声音信号和三种故障信号;从声音信号中提取39维梅尔频率倒谱系数和短时能量,搭建一维卷积神经网络模型进行识别。将声音信号转化成语谱图,建立二维卷积神经网络模型并识别。利用多模型融合技术中的加权平均算法将两个模型融合,融合后模型的准确率为93.58%,比单个模型平均提高2.43%。
A fault diagnosis method based on acoustic characteristics is designed to solve the problems of low quality detection and difficulty in fault identification of needle micro-motor. The normal sound signal and three kinds of fault signals are obtained by sound acquisition device. 39 dimensions Mel frequency cestrum coefficient and short-time energy are extracted from sound signals to built one dimensional convolutional neural network for identification, and the sound signals are transformed into speech spectrum diagram to establish two dimensional convolutional neural network model for identification. The weighted average algorithm in the multi-model fusion technology is applied to fuse the two models with the accuracy of the fused models up to 93.58%, which is 2.43% higher than the single model on average.
作者
刘艳杰
陈炳发
丁力平
LIU Yanjie;CHEN Bingfa;DING Liping(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)
出处
《机械制造与自动化》
2022年第2期190-194,共5页
Machine Building & Automation
关键词
声学特征
模型融合
卷积神经网络
微电机
故障诊断
acoustic feature
model fusion
convolutional neural network
micro-motor
fault diagnosis
