基于对抗自编码模型的高速泵异常检测

Anomaly Detection of High Speed Pump Based on Adversarial Auto-Encoder Model

针对传统大型旋转机械健康状态评估中过分依赖人工经验和对复杂信号的处理通用性较差的问题,基于对抗自编码模型提出一种误差阈值异常检测方法。直接利用设备振动信号进行特征提取与运行状态建模,利用正常状态下设备的振动状态数据建立分布模型;通过深度学习的方式学习振动数据的内在特征,并引入误差阈值作为故障预警的决策准则,实现设备运行状态的高效评估;以一台高转速离心泵为测试对象验证所提方法。结果表明:对抗自编码模型对异常数据的判断准确率能达到100%,该方法能够基于监测数据对旋转设备运行状态进行有效检测;相比于传统自编码神经网络,该方法的诊断准确度和精度大幅提高。

To address the problems of over-reliance on manual experience and poor generality of processing complex signals in the traditional rotating machinery health state assessment methods, an error threshold anomaly detection method was proposed based on the adversarial autoencoders model(AAE).Feature extraction and operating state modeling were carried out by using the vibration signal of equipment directly, and distribution model was established by using vibration state data of the equipment in normal state;deep learning was used to learn the inherent characteristics of vibration data, and error threshold was introduced as the decision criterion of fault warning to achieve the efficient evaluation of the equipment operating state;a high speed centrifugal pump was used to verify the proposed method.The results show that the accuracy of the counteracting self-coding model can reach 100% for the judgment of abnormal data, by using this method, the operation state of rotating equipment can be effectively detected based on the monitoring data;compared with the traditional auto-encoder, the method has a significant improvement in the accuracy and precision of diagnosis.