基于决策树算法的桥梁应变传感器故障识别

Fault Identification of Bridge Strain Sensor Based on Decision Tree Algorithm

摘要:

针对桥梁健康监测系统中传感器故障经常触发报警并影响后续分析的问题，提出了一种基于决策树算法的传感器故障识别方法。根据应变传感器常见故障类型，分别建立了偏置、漂移、卡死等故障的数学模型，并提出分别采用对称位置传感器曼哈顿距离的平均值、前后半段数据平均值之差的绝对值、极差等作为各种故障的数据特征；给出了数据特征提取时最优信号时长的确定方法，并构建了以数据特征为输入、采用决策树算法进行机器学习的传感器故障识别模型。通过某斜拉桥应变监测数据的应用，验证模型对故障类型的识别能力；结果显示，该模型对于故障数据的识别精度达到92.7%。该方法构建的模型复杂度低、可解释性强，为桥梁健康监测系统的传感器故障诊断提供了可行的方法。

Abstract:

To address the issue of frequent false alarms caused by sensor fault in bridge health monitoring system that affects subsequent analysis, a sensor fault identification method based on decision tree algorithm is proposed. According to the common fault types of strain sensors, the mathematical models for biases, drifts and stuck faults are established respectively. It is proposed to respectively adopt the average value of the Manhattan distance of the symmetrical position sensor, the absolute value of the difference between the average values of the front and back halves of the data, the range and the others as the data characteristics of various faults. A method of determining the optimal signal duration for data feature extraction is given, and a sensor fault identification model is constructed, which takes data features as input and uses the decision tree algorithm for machine learning. The capability of the model to identify fault types is verified through the application of strain monitoring data of a cable-stayed bridge. The results show that the identification accuracy of this model for fault data reaches 92.7%. The model constructed by this method has low complexity and strong interpretability, which provides a feasible approach for sensor fault diagnosis in bridge health monitoring systems.