| Résumé | Regular monitoring of bridge and rail structural abnormalities is inseparable from the application of industrial sensors, especially the data analysis of new sensors, such as displacement, temperature, strain and other measurement sensors. The data-driven analysis model based on machine learning provides the possibility of long-distance, long-term, and abnormal state monitoring for early warning and effective management of important structural components. This paper presents a systematic data-driven analysis pipeline for anomaly monitoring. Firstly, data correlation analysis is performed to ensure that prediction models can benefit from strong correlations. Next, a density distribution analysis of statistical signals is established, enabling the reconstruction of global data signals within a certain bandwidth to compensate for data missingness and errors. Subsequently, a seasonal Auto-regressive Integrated Moving Average (ARIMA) model is developed for data prediction, providing reference values for abnormal behavior monitoring. Particularly, we propose the utilization of kurtosis to introduce long-term memory models. By performing cumulative prediction and comparing it with the measured values from continuous time series, an effective method for detecting anomalies can be achieved. Overall, our systematic approach effectively promotes the intelligent development of bridge health and safety monitoring. |
|---|
