基于机器学习的海洋平台往复式压缩机故障诊断方法

Fault diagnosis method for reciprocating compressor of offshore platform based on machine learning

针对深远海平台往复式压缩机智能在线监测以及故障诊断预警需求,提出以往复压缩机示功图为基础的智能诊断方法。将正常示功曲线与实测示功曲线置于同一示功图中对比识别,依靠一种卷积神经网络模型的特征提取和自学习能力进行示功图分类和往复式压缩机智能故障判别。通过使用Fluent软件数值模拟压缩机故障并提取气缸内压力变化生成示功图,为智能机器学习模型提供数据库,进行训练和测试,结果显示所提出的基于卷积神经网络故障识别方法准确率在95%以上,最高准确率可达99.14%,为实际压缩机的预测性智能维护提供了理论支撑。应用于深海平台往复压缩机故障诊断,智能分类识别了压缩机排气过程中的压力振荡现象,初步诊断为压缩机气阀排气量不匹配。

For the requirements of intelligent on-line monitoring and early warning of reciprocating compressor on deep-sea platform,an intelligent diagnosis method based on indicator diagram of reciprocating compressor was proposed.The normal indicator curve and the measured indicator curve were placed in the same indicator diagram for comparison and identification.The indicator diagram classification and intelligent fault discrimination of reciprocating compressor were carried out by relying on the feature extraction and self-learning ability of a convolutional neural network model.The software FLUENT was used to simulate the compressor fault and extract internal pressure change of the cylinder to generate the indicator diagram so as to provide the database for training and testing the intelligent machine learning model.The results show that the accuracy of the proposed fault identification method based on convolutional neural network is more than 95%,and the highest accuracy can reach 99.14%,which provides a theoretical support for the predictive intelligent maintenance of the actual compressor.When applied to the fault diagnosis of reciprocating compressor on deep-sea platform,the pressure oscillation phenomenon in the process of compressor exhaust was intelligently classified and identified,and was preliminarily diagnosed that the exhaust volume of compressor valve does not match.