基于核规范变量分析(KCVA)和长短期记忆网络(LSTM)的液压潜液泵故障诊断

Fault Diagnosis in Hydraulic Submerged Pumps Using Kernel Canonical Variate Analysis (KCVA) and Long Short-Term Memory (LSTM)

摘要: 本文针对液压潜液泵的电机过载、温度异常、液压系统泄漏及冷却系统效率低下等典型故障，开展故障诊断方法研究。传统方法多依赖振动信号与经验判断，难以应对多源高维数据中存在的强非线性、动态耦合及时序演化特性，导致故障识别能力不足。为此，本研究提出一种融合核规范变量分析(KCVA)与长短期记忆网络(LSTM)的两阶段诊断模型。该模型首先利用KCVA对原始监测数据进行非线性动态特征提取，挖掘变量间隐含的特征信息，进而将所提取的特征序列输入LSTM网络，对故障的长期演化建模，以实现对渐进性与交互性故障的精准辨识。本研究在特征提取与时序建模两个层面实现优化，提升了液压潜液泵故障诊断的准确性与可解释性，为复杂工业过程中的故障智能诊断提供了新的解决思路。

Abstract: This study focuses on the fault diagnosis methods of typical failures in hydraulic submerged pumps, such as motor overload, temperature anomalies, hydraulic system leakage, and cooling system inefficiency. Traditional methods, which largely rely on vibration signals and empirical judgment, struggle to handle the strong nonlinearity, dynamic coupling, and temporal evolution present in multi-source, high-dimensional data, resulting in insufficient fault identification capabilities. To address these limitations, this research proposes a two-stage diagnostic model that integrates Kernel Canonical Variate Analysis (KCVA) with Long Short-Term Memory (LSTM). The model first employs KCVA to perform nonlinear dynamic feature extraction from raw monitoring data, uncovering implicit characteristic information among variables. The extracted feature sequences are then fed into an LSTM network to model the long-term evolution of faults, enabling accurate identification of progressive and interactive failures. By optimizing both feature extraction and temporal modeling, this study enhances the accuracy and interpretability of fault diagnosis for hydraulic submerged pumps, offering a novel and effective solution for intelligent fault diagnosis in complex industrial processes.