Anomaly detection of structural health monitoring data using the maximum likelihood estimation-based Bayesian dynamic linear model

Enormous data are continuously collected by the structural health monitoring system of civil infrastructures. The structural health monitoring data inevitably involve anomalies caused by sensors, transmission errors, or abnormal structural behaviors. It is important to identify the anomalies and find their origin (e.g. sensor fault or structural damage) to make correct interventions. Moreover, online anomaly identification of the structural health monitoring data is critical for timely structural condition assessment and decision-making. This study proposes an online approach for detecting anomalies of the structural health monitoring data based on the Bayesian dynamic linear model. In particular, Bayesian dynamic linear model, consisting of various components, is implemented to characterize the feature of real-time measurements. Expectation maximization algorithm and Kalman smoother are combined to estimate the Bayesian dynamic linear model parameters and generate log-likelihood functions. The subspace identification method is introduced to overcome the initialization issue of the expectation maximization algorithm. The log-likelihood difference of consecutive time steps is then used to determine thresholds without introducing extra anomaly detectors. The proposed Bayesian dynamic linear model-based approach is first illustrated by the simulation data and then applied to the structural health monitoring data collected from two long-span bridges. The results indicate that the proposed method exhibits good accuracy and high computational efficiency and also allows for reconstructing the strain measurements to replace anomalies.

Improvement of Bayesian Dynamic Linear Model for Predicting Missing Data of Bridges

The missing data in bridge operation will lead to the decline of the reliability of data analysis results. In this paper, the Bayesian dynamic linear model is improved by changing the parameter matrix of hidden state variables, and the model is optimized under the condition that the predefined variables are unchanged. The frequency of a strain measuring point of the bridge is taken as the observed value, and the collected frequency value of one month is used as the training set (the collection time interval is 30 minutes) to predict the data of the next week. By comparing the predicted result with the observed value, it is found that the absolute error is less than 14.05Hz and the relative error is less than 1.82% when the training frequency value varies from 756 Hz to 773.4 Hz.