基于频域特征的航空轴承智能诊断
Intelligent diagnosis of aviation bearings based on frequency domain features
李宏宇 1苏越 1陈康 1王俨剀1
作者信息
- 1. 西北工业大学动力与能源学院,西安 710072
- 折叠
摘要
针对航空发动机滚动轴承的故障诊断,提出一种基于频域特征的故障诊断模型.将原始振动信号进行包络解调预处理,仅取每段数据处理后的512个点作为故障特征,将其作为双向循环长短期记忆网络(BiLSTM)模型的输入,可对内圈故障、外圈故障、滚动体故障及每种故障所对应3种不同的故障程度进行诊断.该模型不仅弥补完全由原始振动信号输入导致输入数据冗长,特征不明显等缺点,也弥补由人工提取振动特征来进行故障诊断的不确定性.在滚动轴承公开数据集上进行实验,结果表明故障识别的准确度达到99.8％以上.搭建航空轴承实验器来对方法与模型进行检验.基于频域特征的双向循环长短期记忆网络模型能够更准确地对轴承进行故障诊断,所提方法对于航空发动机滚动轴承故障诊断具有重要工程价值.
Abstract
A fault diagnosis model based on feature extraction was proposed for aeroengine rolling bearing fault diagnosis.The original vibration signals were preprocessed by envelope demodulation,and only 512 points of each segment of data were taken as fault features,and used as input of bidirectional long short-term memory(BiLSTM)model to diagnose the inner ring faults,outer ring faults,rolling body faults and three different fault degrees corresponding to each fault.The model made up for not only the disadvantages of long input data and obscure features caused by the original vibration signal input,but also the uncertainty of fault diagnosis by extracting vibration features manually.Experiments on the open data set of rolling bearings showed that the accuracy of fault identification was above 99.8％.An aero-bearing tester was built to verify the method and model.BiLSTM based on feature extraction can diagnose the bearing faults more accurately.The proposed method has important engineering value for the fault diagnosis of aeroengine rolling bearings.
关键词
故障诊断/特征提取/神经网络/滚动轴承/双向循环长短期记忆网络Key words
fault diagnosis/feature extraction/neural network/rolling bearing/bidirectional long short-term memory引用本文复制引用
基金项目
国家科技重大专项(2017-Ⅰ-0006-0007)
出版年
2024
NETL
NSTL
万方数据