旋转机械突发不平衡故障早期预警及诊断方法研究

Research on Incipient Warning and Diagnosis Method of Sudden Unbalance Fault for Rotating Machinery

旋转机械突发不平衡故障的早期预警是当前状态监测领域工程实践中的难题,对于避免设备非计划停机造成灾害事故发生具有重要的意义。针对实际工程应用中,正常数据丰富、故障数据缺乏的情况,提出一种旋转机械突发不平衡故障早期预警及诊断方法。提取正常运行工况历史数据的特征指标,构建支持矢量数据描述预警模型检测早期故障趋势,再利用包括强化特征提取、强化特征迁移、强化模式识别的强化故障诊断方法进行模式识别。应用两组石化企业实际工程故障案例作为测试数据,早期预警方法相较于传统的振动峰-峰值报警分别提前了2 400 min和4 330 min。强化故障诊断方法对两组测试数据的识别准确率分别为95%和90%,与其他故障识别方法相比诊断精度和F分数最高,标准差最低。结果表明,所提方法具有良好的早期故障检测性能,对不同设备和工况的跨域数据故障识别鲁棒性较好,领域泛化能力较强。

Incipient warning of sudden unbalance faults for rotating machinery is a difficult problem in engineering practice in the field of current condition monitoring,and it is of great significance to avoid disaster accidents caused by unplanned downtime of equipment.Aiming at the situation of abundant normal data and lack of fault data in practical engineering applications,an incipient warning and diagnosis method of sudden unbalance faults for rotating machinery is proposed.Characteristic indicators of historical data for normal operating conditions are extracted,and an incipient warning model based on support vector data description is built to detect incipient fault trends.Then,the mode identification is carried out applying reinforced fault diagnosis method including reinforced feature extraction,reinforced feature transfer,and reinforced pattern recognition three steps.Two groups actual engineering failure cases of petrochemical enterprises are used as test data,the incipient warning method is 2400 min and 4330 min earlier than the traditional vibration peak-to-peak alarm,respectively.The identification accuracy of the reinforced fault diagnosis method for two groups test data is 95%and 90%,respectively.Compared with other fault identification methods,the diagnosis accuracy and F-Score are the highest,and the standard deviation is the lowest.The results indicate that the proposed method possesses more excellent incipient fault detection performance,good robustness to cross-domain data fault identification of different equipment and working conditions,and strong domain generalization ability.