基于MFCC-IMFCC混合倒谱的托辊轴承故障诊断
Fault diagnosis of idler bearings based on MFCC-IMFCC hybrid cepstral coefficients
陶瀚宇 1陈换过 1彭程程 2高祥冲 1杨磊1
作者信息
- 1. 浙江理工大学 浙江省机电产品可靠性技术研究重点实验室,浙江 杭州 310018
- 2. 杭州聆为信息技术有限公司 技术部,浙江 杭州 311215
- 折叠
摘要
针对梅尔倒谱系数(MFCC)对托辊轴承高频特征提取能力不足的问题,提出了一种基于梅尔倒谱系数和翻转梅尔倒谱系数(MFCC-IMFCC)的混合倒谱以及长短时记忆(LSTM)网络的托辊轴承故障诊断方法.首先,分析了三种状态下的托辊声音信号,明确了托辊轴承故障信息主要分布在中高频区域;然后,为有效保留高频信息,提取了MFCC-IMFCC,以帧级串联的方式组成了混合倒谱特征;最后,将混合倒谱特征输入到双层LSTM模型中进行了训练,建立了托辊轴承故障诊断模型.研究结果表明:针对托辊正常、滚动体故障和偏心旋转故障三种状态,LSTM结合混合倒谱特征的平均识别准确率达到 96.72%,相比于单一的MFCC和IMFCC特征,准确率分别提升3.94%和7.41%,凸显了混合倒谱特征在表征托辊轴承故障信息方面的显著优势.
Abstract
Addressing the insufficient capability of Mel-frequency cepstral coefficient(MFCC)in extracting high-frequency features of idler bearing faults,a novel fault diagnosis method for idler bearings based on Mel-frequency cepstral coefficient and inverse-Mel-frequency cepstral coefficient(MFCC-IMFCC)hybrid cepstral coefficients and long short-term memory(LSTM)networks was proposed.Firstly,the acoustic signals of idler under three states were analyzed,revealing that the bearing fault information mainly resided in the mid-to-high-frequency range.Then,to effectively retain high-frequency information,MFCC-IMFCC were extracted and combined in a frame-level concatenation to form hybrid cepstral features.Finally,the hybrid cepstral features were input into a two-layer LSTM model for training,establishing a diagnostic model for idler bearing faults.The research results indicate that,for normal state,rolling element fault state,and eccentric rotation fault state,the average recognition accuracy of LSTM combined with hybrid cepstral features reaches96.72% .Comparing to using individual MFCC and IMFCC features,the accuracy is improved by 3.94% and 7.41%,highlighting the significant advantage of hybrid cepstral features in representing information about idler bearing faults.
关键词
托辊轴承/轴承故障声音信号/高频信息/梅尔倒谱系数/翻转梅尔倒谱系数/混合倒谱系数/长短时记忆网络Key words
idler bearings/bearing fault acoustic signal/high frequency information/Mel-frequency cepstral coefficient(MFCC)/inverse-Mel-frequency cepstral coefficient(IMFCC)/hybrid cepstral coefficients/long short-term memory(LSTM)networks引用本文复制引用
基金项目
国家自然科学基金资助项目(51975535)
国家重点研发计划项目(2021YFB3301601)
出版年
2024
NETL
NSTL
万方数据