基于动态主题模型的往复机械故障早期预警方法

An early warning method for reciprocating mechanical faults based on a dynamic topic model

往复机械的故障早期高预警准确率对于设备安全运维具有重要意义。目前企业广泛应用的单特征门限报警方法存在预警准确率普遍较低的问题,这主要是由往复机械运行过程中振动激励性信号的非平稳性导致。针对该问题,提出一种基于动态主题模型(dynamic topic model,DTM)的往复机械故障早期预警方法。该方法运用机器统计学习的主题模型建模方法,基于往复机械正常运行状态的数据生成正常运行数据模型分布,并将其作为基准模型,根据实时工况状态数据建立的混合模型与基准混合模型间的差异实现对往复机械的故障早期预警。最后,分别采用往复压缩机工程案例数据和故障模拟试验数据对所提方法进行验证,结果表明:所提方法可准确识别设备的异常状态,同时减少了预警分析计算时间,有效提升了往复机械故障早期预警的准确性和时效性。

Improving the early warning accuracy of reciprocating machinery is of great significance for the safe operation and maintenance of equipment.The single-feature threshold early warning alarm method that is currently widely used in enterprises suffers from generally low accuracy.In this paper,we propose an early warning method for reciprocating mechanical faults based on a dynamic topic model(DTM).The method uses dynamic thematic modeling to model the equipment vibration response signal time sequence dynamically and observe the dynamic changes of each excitation source signal with time in a targeted manner.This method can effectively reduce the influence of non-stationary signals caused by the mutual coupling of excitation sources on the warning index.By analyzing the difference between the real-time working state and the normal state of the mixed model,early warning of the failure of the reciprocating machinery can be realized.Engineering case data and reciprocating compressor fault simulation test data were used to verify the proposed method.The test results show that the method can accurately identify the abnormal state of the equipment,while reducing the calculation time of early warning analysis,which effectively improves the accuracy and timeliness of the early warning of reciprocating mechanical faults.