神经网络和证据理论融合的尾轴承磨损故障诊断  

Diagnosis of Stern Bearing Wear Fault with Combination of Neural Network and Evidence Theory

作　　者：王永坚[1] 陈景锋[1] 杨小明 

机构地区：[1]集美大学轮机工程学院,福建厦门361021 [2]福建省厦门轮船公司,福建厦门361021

出　　处：《中国航海》2014年第4期20-24,共5页Navigation of China

基　　金：福建省自然科学基金(2012J01228);福建省教育厅资助项目(JA12203)

摘　　要：为克服铁谱、光谱、理化和颗粒计数等4种油液检测分析方法在船舶尾轴承磨损故障诊断中存在的准确性偏差等问题,提出运用神经网络和D-S证据理论对尾轴承磨损故障进行融合诊断。依据各分析方法的标准磨损界限值,将各检测原始数据预处理转换为布尔值,运用神经网络算法获取每种检测方法的故障域单项诊断结果。利用D-S证据理论融合各单项故障诊断结果,以获得更为准确的诊断结果,并通过具体的案例验证方法的准确性。With the aim of improving the accuracy of existing oil analysis techniques for diagnosing stern bearing wear fault, the method of fusing 4 commonly used diagnosis analysis （ ferrograph, spectroscopy, oil chemical-physics analysis and particle counting analysis） is designed based on neural networks and the D-S evidence theory. The test data from the 4 kinds of analysis of the oil sample are transformed into BOOL values by comparing them to respective wear limit values, and processed by means of neural network analysis method to find the tentative decisions given by each analysis method. These tentative decisions are processed according to the D-S evidence theory to get the final diagnosis decision. The method is verified by actual cases.

关 键 词：船舶工程 船舶尾轴承 油液分析 神经网络 D-S证据理论 磨损故障 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] U664.21[自动化与计算机技术—控制科学与工程]