Faulty data detection and classification for bridge structural health monitoring via statistical and deep‐learning approach

Abstract

Summary Over the last several decades, a lot of bridges have been equipped with the bridge structural health monitoring system, leading to an accumulation of voluminous monitoring data. Since the sensors and associated transmission hardware are subjected to harsh environments, the monitoring data frequently contains various faults, and it is laborious to cleanse the data manually. For the purpose of automatically detecting and classifying faulty monitoring data in large quantities, this paper proposes a novel method that uses the relative frequency distribution histograms (RFDH) of monitoring data as well as the one‐dimensional convolutional neural network (1‐D CNN). The overall procedure of this method can be described as follows: First, RFDHs are constructed from different classes of hour‐long data segments. Second, inverted envelopes of the RFDHs are labeled as the training data to train the 1‐D CNN. Third, a well‐trained 1‐D CNN is used to detect and classify long‐term monitoring data according to their RFDHs of hour‐long data segments. Comprehensive validation of the proposed method is conducted with selective acceleration data collected from two long‐span bridges. The validation yields satisfactory results, demonstrating the accuracy, efficiency, and generality of the method.

The application of deep learning in bridge health monitoring: a literature review

“…Anomaly identification for vibration signals can be disposed as time series classification, for which the effectiveness of one-dimensional (1D) CNN has been proved by some researchers (Jian et al 2021;Zhang and Lei 2021). Visualizing the time signals into images is an effective approach to leverage two-dimensional (2D) CNN for this mission.…”

“…Visualizing the time signals into images is an effective approach to leverage two-dimensional (2D) CNN for this mission. And drawing their curves directly is the most frequently adopted method, while other feature engineering approaches, like the spectrogram analysis, the probability density function and so on, were also employed to enhance the network's robustness (Jian et al 2021;Shajihan et al 2022). For example, Tang et al (2019) visualized time series data in time and frequency domain and stacked them as a single dual-channel image before it was inputted into a 2D CNN to classify data anomalies (see Fig.…”

A two‐stage data cleansing method for bridge global positioning system monitoring data based on bi‐direction long and short term memory anomaly identification and conditional generative adversarial networks data repair

“…Thus, for a training sample ranging over 1-h values, the feature extracted is represented by a 9 Â 60 matrix. The features are listed as follows: (1) mean value, (2) maximum value difference, (3) standard deviation, (4) root mean square level, (5) difference between the maximum and minimum, (6) maximum value divided by the 80th percentile, (7) mean value divided by the amplitude, (8) mean value divided by the 80th percentile, and ( 9) natural frequency estimated by Welch's power spectral density. The attempts in feature selection, including data dimensionality reduction and feature selection based on statistical indicators, and the construction of data set toward solving the above problems in deep learning can provide a certain reference for the classification tasks for deep learning.…”

“…The process of performing data anomaly identification and locating and repairing anomalous data segments is called data cleansing 6 . Therefore, data cleaning is the basic prerequisite for data mining and analysis by the monitoring system 7 …”

“…6 Therefore, data cleaning is the basic prerequisite for data mining and analysis by the monitoring system. 7 Data cleansing consists of two parts: data anomaly identification and data repair. For anomaly identification, traditional methods are mainly based on a statistical analysis of long-term historical data by finding sensitivity indicators and setting reasonable thresholds for specific indicators.…”