基于卷积神经网络的发动机凸轮轴在线振动监测模型及异常信号处理算法研究  

Research on Online Vibration Monitoring Model of Engine Camshaft and Abnormal Signal Processing Algorithm Basedon Convolution Neural Network

作　　者：卢燃 庞博 胡勇 彭六保 LU Ran;PANG Bo;HU Yong;PENG Liubao(Equipment Maintenance Center of Guoneng Beidian Shengli Energy Co.,Ltd.,Xilinhot 026000,China;Aerospace Intelligent Control(Beijing)Monitoring Technology Co.,Ltd.,Beijing 100000,China)

机构地区：[1]国能北电胜利能源有限公司设备维修中心,内蒙古锡林浩特026000 [2]航天智控(北京)监测技术有限公司,北京100000

出　　处：《内燃机》2022年第6期19-24,共6页Internal Combustion Engines

摘　　要：当凸轮轴无法正常运行时,会导致大型机械的发动机也无法正常工作,造成严重的经济损失。提出了一种基于卷积神经网络的在线监测模型,该模型不仅可以对发动机凸轮轴的振动信号进行在线监测,还可以对凸轮轴的异常信号进行处理。首先利用传感器提取凸轮轴运行期间的磨损信号,并且以历史的信号数据作为样本进行训练,得到卷积神经网络的参数权重。通过传感器采集凸轮轴磨损情况的特征信号数据,并将时间序列分析方法带入到卷积神经网络建模过程中,提高凸轮轴振动信号监测的准确率。对异常信号进行数据处理以对凸轮轴做故障判定。结果表明,本文模型可以有效地监测出凸轮轴在不同转速下非正常运行振动信号波形;对3种路面情况下的凸轮轴振动信号进行监测,发现本文模型监测的正确率和F1值均为最高,分别达到94.51%和96.42%、 98.32%和94.55%、 92.972%和92.16%;而漏检率和误拦率均为最低,分别为3.41%和6.02%、 4.69%和6.34%、 9.31%和10.01%。因此,证实了本文提出的模型理想的监测性能。When the camshafts do not work properly, the engines of large machinery will not work properly, which will cause serious economic losses. Therefore, an on-line monitoring model based on convolution neural network is proposed, which can not only monitor the vibration signal of engine camshaft on-line, but also process the abnormal signal of camshaft. Firstly, the wear signal during camshaft operation is extracted by sensors and trained with historical signal data as samples to get the parameter weights of convolution neural network. The sensor collects the characteristic signal data of camshaft wear and brings the time series analysis method into the convolution neural network modeling process to improve the accuracy of camshaft vibration signal monitoring. The abnormal signal is processed to determine the camshaft fault. The results show that the model in this paper can effectively monitor the vibration signal waveform of camshaft abnormal operation at different speeds. By monitoring camshaft vibration signals under three road conditions, it is found that the correct rate and F1 value of the model monitoring in this paper are the highest, reaching 94.51% and 96. 42%, 98.32%and 94.55%, 92.972% and 92.16% respectively. The leak detection rate and false interception rate are the lowest, which are 3.41% and 6.02%, 4.69% and 6.34%, 9.31% and 10.01% respectively. Therefore, it can be proved that the monitoring performance of the model proposed in this paper is ideal.

关 键 词：发动机凸轮轴 在线监测 监测模型 卷积神经网络 异常信号处理 

分 类 号：TK402[动力工程及工程热物理—动力机械及工程]