多参数优化深度置信网络的滚动轴承外圈损伤程度识别

Damage Severity Identification for Outer Rings of Rolling Bearings Using Multi-Parameter Optimized Deep Belief Network

针对滚动轴承振动信号故障特征提取依赖于专家经验引起的不确定性影响识别准确率的问题,提出一种基于多参数优化深度置信网络的滚动轴承外圈损伤程度识别方法。首先,考虑外圈滚道损伤分布弧长的变化对滚动轴承动力学特性的影响,建立五自由度滚动轴承损伤动力学模型,仿真求解外圈不同损伤程度的滚动轴承响应信号;然后,基于模拟退火算法优化DBN的多个结构参数,利用仿真数据的原始时域波形直接进行损伤程度的识别。不同信噪比外圈损伤仿真数据与齿轮箱轴承试验数据的分析结果表明,该方法无需先验知识提取外圈损伤特征,可直接利用轴承原始时域数据自学习地提取不同损伤程度的特征信息,且识别准确性和稳定性更高,具有工程应用价值。

The uncertainties caused by expertise for extracted fault features of vibration signals of rolling bearings affect identification accuracy.A method is proposed for identifying damage severity for outer rings of rolling bearings based on multi-parameter optimized deep belief network(DBN).Firstly,the five degree of freedom damage dynamic model of rolling bearings is established with the consideration of influence of variable arc lengths of damage distribution of outer rings on dynamic characteristics of rolling bearings.The response signals of rolling bearings with different damage severity of outer rings is then simulated and solved.According to simulated annealing algorithm(SAA),the structural parameters of DBN are optimized,and the identification of damage severity is carried out directly by using original time-domain waveform of simulation data.The analysis results of simulation data of outer ring damage under different SNR and test data of gearbox bearings show that the method is able to extract outer ring damage feature without prior knowledge,the feature information of different damage severity is extracted directly by using original time domain data of bearings with high identification accuracy and stability,which is of great engineering significance.