基于统计域指数的压力类传感器故障检测方法

打开文本图片集
中图分类号: TB9; TH81 文献标志码: A文章编号: 1674–5124(2025)05–0110–07
Abstract: A method for sensor fault detection based on statistical domain indices has been proposed to address the issue of pressure sensor failures in industrial processes due to aging and environmental disturbances. The method first constructs a time-series model predicting the normal output of sensors using a long short-term memory (LSTM) neural network, generating residual signals from the model's predicted values and actual measurements. Subsequently, it calculates the moving average index (MAI), moving root mean square index (MRI), moving variance index (MVI), and moving energy index (MEI) of the residual signals. Thresholds are designed using the interquartile range (IQR) method for sensor fault detection. Finally, the method was experimentally validated using historical operational data from a 320MW coal-fired unit's induced draft fan outlet flue gas pressure sensor, and compared with traditional residual analysis. The results showed that this method improved accuracy, precision, recall, and F -value by 11.88% , 3.16% , 22.15% , and 14.06% respectively. It has significant advantages in pressure sensor fault detection. Keywords: fault detection; statistical domain indices; pressure sensor; residual analysis; neural network
0 引 言
压力传感器可用于测定介质的压力、液位和流量等参数,是工业过程中重要的组成部分。(剩余9855字)