基于音频特征融合的振动筛故障诊断方法  

Fault Diagnosis Method of Vibrating Screen Based on Audio Feature Fusion

作　　者：李越 李敬兆[2] 何长林 王斌[3] 李彪[3] LI Yue;LI Jingzhao;HE Changlin;WANG Bin;LI Biao(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China;chool of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China;Huainan Mining Group Coal Preparation Branch,Huainan Anhui 232000,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001 [2]安徽理工大学电气与信息工程学院,安徽淮南232001 [3]淮南矿业集团选煤分公司,安徽淮南232000

出　　处：《兰州工业学院学报》2025年第1期60-67,共8页Journal of Lanzhou Institute of Technology

基　　金：国家自然科学基金(52374154)。

摘　　要：为及时发现振动筛的故障,提出一种融合改进梅尔频率倒谱系数(MFCC)、密集卷积神经网络(Dense-CNN)和双向长短期记忆网络(BiLSTM)的振动筛故障诊断模型(Dense-CNN-BiLSTM)。首先,利用固有时间尺度分解(ITD)对振动筛音频信号进行时频分析,提取其固有旋转分量(PRC);其次,提取由独立成分分析(ICA)改进的13维MFCC特征参数,并将特征参数输入Dense-CNN-BiLSTM模型,实现振动筛的故障诊断。结果表明:改进的MFCC特征参数能表示振动筛不同运行状态的音频信号特征,验证了基于音频特征融合实现振动筛故障诊断的可行性。In order to detect the faults of vibrating screen in time,a vibrating screen fault diagnosis model(Dense-CNN-BiLSTM)integrating the modified Mel frequency cepstrum coefficient(MFCC),dense convolutional neural network(Dense-CNN)and bidirectional long and short-term memory network(BiLSTM)is proposed.Firstly,the time-frequency analysis of the vibrating screen audio signal is carried out by using intrinsic time-scale decomposition(ITD)to extract its PRC components;secondly,the 13-dimensional MFCC feature parameters improved by independent component analysis(ICA)are extracted and the feature parameters are input into the Dense-CNN-BiLSTM model to realize the fault diagnosis of vibrating screen.The results show that the improved MFCC feature parameters can represent the audio signal characteristics of different operating states of the vibrating screen,which verifies the feasibility of realizing vibrating screen fault diagnosis based on audio feature fusion.

关 键 词：振动筛 梅尔频率倒谱系数 密集卷积神经网络 双向长短期记忆网络 

分 类 号：TD67[矿业工程—矿山机电]