基于熵理论和BP神经网络的船舶柴油机磨损故障识别  

Wear Fault Identification of Marine Diesel Engine Based on Entropy Theory and BP Neural Network

作　　者：石新发 邢广笑 贺石中 谢小鹏[3] SHI Xinfa;XING Guangxiao;HE Shizhong;XIE Xiaopeng(Guangzhou Mechanical Engineering Research Institute Co.,Ltd.,Guangzhou Guangdong 510530,China;The 92001 Troops of PLA,Qingdao Shandong 266000,China;School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou Guangdong 510640,China)

机构地区：[1]广州机械科学研究院有限公司,广东广州510530 [2]中国人民解放军92001部队,山东青岛266000 [3]华南理工大学机械与汽车工程学院,广东广州510640

出　　处：《润滑与密封》2022年第7期54-58,共5页Lubrication Engineering

基　　金：广东省科技计划项目(2020B1212070022);广州机械科学研究院有限公司博士后专项(17300065)。

摘　　要：磨损监测与故障诊断是保证船舶柴油机安全可靠运行的重要技术手段。随着船舶柴油机运行可靠性的要求增高,其磨损监测需要更加全面,数据呈高维化,无关数据和冗余数据增多,使故障诊断的复杂程度增大,且近年来,船舶柴油机故障诊断的智能化需求日益增高。针对以上问题和需求,基于信息熵理论,应用信息熵值与度量熵组合设计柴油机磨损监测与故障诊断特征属性约简算法,将某型柴油机润滑磨损故障诊断特征指标维度从16维降低至7维;应用设计的BP神经网络和磨损故障模式识别规则,以该型柴油机44个磨损故障诊断数据样本为对象,进行应用验证与研究分析。结果表明,构建的模型在保证数据集分类特性的基础上,有效实现其数据降维,且所构建的磨损故障识别BP神经网络在属性约简后,故障识别的准确性有明显提高。Wear monitoring and fault diagnosis are important technical methods to ensure the safe and reliable operation of marine diesel engines.With the increasing requirements for the operational reliability of marine diesel engines,the wear monitoring needs to be analyzed more comprehensively.High-dimension data,irrelevant data and increasing redundant data multiply the complexity of fault diagnosis.In recent year,the demand for intelligent fault diagnosis of marine diesel engines is increasing day by day.Aimed at the above problems and requirements,based on the information entropy theory,the unsupervised attribute reduction algorithm for wear monitoring and fault diagnosis features of diesel engine was designed by combining information entropy and measurement entropy.With this algorithm,the dimension of feature index for the diesel engine lubrication wear fault diagnosis can be reduced from 16 to 7.By applying the BP neural network and identification rules of wear fault patterns,44 wear fault diagnosis samples of the diesel engine were verified and analyzed.The results show that the model can effectively reduce the dimension of the data set on the basis of ensuring the classification characteristics of the data set,and the accuracy of wear fault identification by the constructed BP neural network is significantly improved by taking the reduced data set as the research object.

关 键 词：柴油机 油液监测 磨损故障识别 熵理论 BP神经网络 

分 类 号：TH117.1[机械工程—机械设计及理论]