Abstract: It is difficult to track the occurrence and development of the performance degradation of rotating machinery by the conventional fixed threshold alarm method in the in-service condition monitoring system.To solve this problem,a data-driven rotating equipment performance degradation trend prediction model was constructed by using the raw normal vibration monitoring data and real-time monitoring data,and a method for predicting performance degradation trend based on the spectral distance index operating reliability curve l1 trend filtering was proposed.By dynamically tracking the point-by-point slope change of the trend filtering curve,the occurrence and development of performance degradation of rotating equipment could be detected.Using the Cincinnati Intelligent Maintenance Information System (IMS) center bearing experimental data and the rotor unbalance fault case data of a Chinese petrochemical company's centrifugal compressor to verify the model,the results showed that the data-driven performance degradation prediction model of rotating equipment only needed raw vibration data in normal operating conditions without relying on the prior knowledge of external experts,and could accurately predict and track the occurrence and development of the performance degradation trend of rotating equipment.

Key words: data-driven, performance degradation, trend prediction, l1 trend filtering, predictive maintenance

摘要： 为解决在役状态监测系统采用常规固定阈值报警方法难以追踪旋转机械性能退化发生和发展的问题,应用振动监测原始数据和实时监测原始数据构建了数据驱动的旋转设备性能退化趋势预测模型,提出一种基于谱距离指标运行可靠性曲线l1趋势滤波的旋转设备性能退化趋势预测方法。应用美国辛辛那提智能维修信息系统(IMS)中心轴承实验数据和中国某石化公司离心压缩机转子不平衡故障案例数据验证了所构建的旋转设备性能退化预测模型。结果表明,数据驱动的旋转设备性能退化预测模型只需要运行正常状态振动原始数据,无需依赖外部专家先验知识,能够准确预测和追踪旋转设备性能退化趋势的发生和发展。

关键词: 数据驱动, 性能退化, 趋势预测, l1趋势滤波, 预测性维修