基于最大李雅普诺夫指数异常感知和CatBoost识别的机械密封失效模式层次化诊断框架

Hierarchical Diagnostic Framework of Mechanical Seal Failure Modes Based on Maximum Lyapunov Exponent Anomaly Sensing and CatBoost Model Recognition

离心泵在现代工业生产中具有广泛的应用,其运行状况和健康程度直接影响着整个系统的能耗、效率和安全。机械密封泄漏或损坏是水力旋转机械最典型的故障之一,与机封失效相关的泵类设备故障问题直接影响系统总体的可靠性和安全性。为此,本论文研究了一种基于最大李雅普诺夫指数异常感知和CatBoost识别的机械密封失效模式层次化诊断框架。首先,对采集的机械密封处振动信号序列提取其最大李雅普诺夫指数,并基于模糊统计法和指派法设计Type-1模糊逻辑,从而实现对机械密封故障的异常检测和感知。接着,一旦检测到机封异常,再从原始振动信号中提取多尺度模糊熵,联同最大李雅普诺夫指数一起输入到CatBoost模型进行机械密封失效模式识别和诊断。最后,基于实际实验数据对所提出的层次化诊断框架进行了验证。结果表明,所提出的方法对机封故障的异常检测精度达到100%,CatBoost模型的机封失效模式识别率达到99.66%,其精度和鲁棒性均好于支持向量机、AdaBoost、深度神经网络等智能模型。

Centrifugal pump is widely used in the field of modern industrial production.Its operation and health statement directly affect the energy consumption,efficiency and safety of the whole system.mechanical seal leakage or damage is one of the most typical failures of hydraulic rotating machinery.The fault of pump equipment related to the mechanical seal failure directly affects the overall reliability and safety of the system.therefore,this paper studies a hierarchical diagnostic framework of mechanical seal failure modes based on maximum Lyapunov exponent anomaly sensing and CatBoost model recognition.Firstly,the maximum Lyapunov exponent of the vibration signal sequence collected at the mechanical seal is extracted,and the Type-1 fuzzy logic is designed based on the fuzzy statistical method and assignment method,so as to realize the abnormal detection and sensing of the mechanical seal fault.Then,once the mechanical seal abnormality is detected,the multi-scale fuzzy entropy is extracted from the original vibration signal and input into the catBoost model together with the maximum lyapunov exponent for mechanical seal failure pattern recognition and diagnosis.Finally,the proposed hierarchical diagnosis framework is verified based on the real-world experimental data.The results show that the proposed approach achieves an anomaly detection accuracy of 100%and a failure mode recognition rate of 99.66%for seal failure,whose accuracy and robustness are better than those of intelligent models such as support vector machine,AdaBoost and deep neural network.