一种输油泵机组故障诊断与健康评估方法

输油泵机组是长距离油品传输的关键设备，有效预防其出现突发故障、减小故障造成的损失至关重要。然而目前针对输油泵的故障诊断方法在现场应用时普适性不佳，且缺乏针对机组一体的监测诊断研究，不利于计划性维修。此外，受现场可提供数据的限制，现有的输油泵状态评估方法在很多现场无法使用。针对上述问题，提出一种输油泵机组故障诊断与健康评估方法，利用迁移学习提高输油泵故障诊断在工业现场应用时的准确率；通过搭建实验台并对电机运行状态进行监测、分析，构建电机机械类故障诊断模型；构建基于卷积-长短期记忆神经网络(CNN-LSTM)的状态评估模型，并以此为基础利用时序卷积网络(TCN)结合注意力机制进行状态趋势预测。在现场试运行的结果表明，本文提出的故障诊断及状态评估方法可以及时发现设备的早期故障，为设备运维提供有效的数据参考。

Fault diagnosis and health assessment of oil pump units

Oil pump units are critical components in long-distance oil product transmission, and it is very important to prevent losses resulting from their sudden failure. Fault diagnosis and assessment involve online monitoring of the pumps. However, current methods of fault diagnosis in oil pumps do not perform so well in actual industrial sites. In addition, the monitoring and fault diagnosis mainly focuses on the pump itself, and does not include the motor, which is not conducive to planned maintenance. Furthermore, due to the limited data available in industrial sites, many of the existing pump assessment methods cannot be used. We propose a fault diagnosis and health assessment approach for oil pump units, which improves the accuracy of oil pump fault diagnosis in industrial applications by means of transfer learning, and builds a motor fault diagnosis model by establishing an experimental platform for the motor and analyzing the monitoring data. In addition, we construct a health assessment model using CNN-LSTM and a TCN network combined with an attention mechanism to predict trends in pump health. When employed in industrial sites, the fault diagnosis and health assessment approach proposed in this paper can find faults at an early stage and provide a useful data reference for equipment operation and maintenance.