【摘要】 针对单传感器所测数据存在的不全面性及离心泵故障信号的非线性非平稳性等问题，提出多传感器数据下基于多重分形去趋势波动分析法（multi-fractal detrended fluctuation analysis, MFDFA）与BP（back propagation, BP）神经网络的离心泵空化致振故障分析方法。采用MFDFA法对离心泵5种不同工况下的8类实测信号进行分析，提取多重分形谱特征参数Δf、α₀、Δα、α_min及α_max作为故障特征向量，并结合BP神经网络进行单传感器信号故障诊断，优选识别率较佳的信号拼接构成多传感器特征向量，进而开展多传感器离心泵空化致振故障研究。结果表明：MFDFA法提取的多重分形谱特征参数能准确反映泵的运行状态，其中Δf、Δα以及α_max等参数对故障分类效果更好；泵振动、扭矩及电机振动等信号对故障本质的反映更准确；在此基础上形成的多传感器故障诊断模型准确率比单传感器提升了13%以上，为离心泵不同程度空化故障的状态识别提供了一种新的方法。更多还原

【Abstract】 Here, aiming at incompleteness of data measured by a single sensor and non-linearity and non-stationarity of centrifugal pump fault signals, a centrifugal pump cavitation induced vibration fault analysis method based on the multi-fractal detrended fluctuation analysis（MFDFA）and the back propagation（BP） neural network under multi-sensor data was proposed.Firstly, MFDFA method was used to analyze 8 types measured signals of centrifugal pump under 5 different working conditions, and extract multifractal spectrum feature parameters of Δf, α₀, Δα, α_min and α_max for forming fault characteristic vector. Combining with BP neural network, fault diagnosis of single sensor signalswas performed, and signals with better recognition rate were optimized to form the multi-sensor characteristic vector, and conduct the study on cavitation induced vibration faults of multi-sensor centrifugal pump. The results showed that the multifractal spectrum feature parameters extracted with MFDFA can correctly reflect operation state of centrifugal pump, andparameters of Δf, Δα and α_max and have a better effect on fault classification; signals, such as, pump vibration, torque and motor vibration reflect fault essence more correctly; the accuracy rate of multi-sensor fault diagnosis model is more than 13% higher than that of a single sensor to provide a new method for state identification of different levels cavitation faultsof centrifugal pump.更多还原