基于核密度估计的旋转机械损伤贝叶斯智能评价方法  

Kernel Density Estimation Based Bayesian Method for Fault Identification of Rotating Machinery

作　　者：赵海心 姜孝谟 徐胜利[1] 林琳[1] 宫云庆 Hai-xin Zhao;Xiao-mo Jiang;Sheng-li Xu;Lin Lin;Yun-qing Gong(Dalian University of Technology;Shenyang Blower Group Co.,Ltd)

机构地区：[1]大连理工大学 [2]沈阳鼓风机集团股份有限公司

出　　处：《风机技术》2020年第3期69-76,共8页Chinese Journal of Turbomachinery

基　　金：国家重点研发计划(2018YFB0606104)。

摘　　要：神经网络预测和贝叶斯假设检验常用于旋转机械的损伤识别。神经网络用于预测外激励作用下旋转机械的动力响应,贝叶斯方法使用传感数据和模型预测之间的残差数据获得旋转机械的健康概率。然而传统的贝叶斯方法是从数据的高斯分布假设中得出的。在实际工程中经常违反此假设,影响损伤识别精度。针对这一问题,本文提出了一种基于核密度估计的非参数贝叶斯评估方法。贝叶斯置信度是从核密度估计中直接得出的,而无需对数据进行任何分布假设,并结合长短时记忆神经网络深度学习预测模型,实现旋转机械的故障损伤识别。通过使用实际运行的压缩机数据,与传统方法进行比较,验证了新方法对旋转机械故障损伤识别的有效性。Bayesian hypothesis test and neural network prediction are commonly used for damage identification of rotating machinery.The neural network model is used to predict the dynamic response of the system under external excitation.Bayesian method uses residual data between sensing data and model predictions to obtain the health probability of rotating machinery.However,the conventional Bayesian method is derived from the Gaussian distribution assumption of the data.This assumption is often violated in practical engineering,which leads to inaccurate fault detection.This paper proposes a nonparametric Bayesian assessment method based on kernel density estimation to address this issue.Bayesian confidence on the system condition is explicitly derived from kernel density estimation,without need of any distribution assumption on the data.A Long Short-Term Memory neural network model is developed to realize fault damage identification of rotating machinery.Using the data collected from real-world centrifugal compressors,a comparison study with the traditional Bayesian hypothesis testing method is conducted to demonstrate the effectiveness and feasibility of the proposed methodology.

关 键 词：数据不确定性 旋转机械 核密度估计 贝叶斯方法 深度学习 故障识别 

分 类 号：TH17[机械工程—机械制造及自动化] TK05[动力工程及工程热物理]