Structural Health Monitoring Data Research Articles

Transfer learning-based data anomaly detection for structural health monitoring

The structural health monitoring (SHM) data of civil infrastructure are inevitably contaminated due to sensor faults, environmental noise interference, and data transmission failures. Anomalous data severely disturb the subsequent structural modal identification, damage identification, and condition assessment. Therefore, it is critical to detect and clean SHM data before data analysis. This paper proposes a novel approach for data anomaly detection based on transfer learning, that makes full use of the similarity of the anomalous patterns across different bridges and shares the knowledge incorporated in a deep neural network to achieve high-accuracy data anomaly identification for bridge groups. In the proposed approach, first, a multivariate database for a source bridge is built by plotting and labeling the raw sequential data. Then, a convolutional neural network (CNN) for data anomaly classification is designed and trained with the database in different conditions. The original CNN with the highest accuracy is transferred to a new bridge with enhancement training using a small part of the target bridge data. To validate the performance of the proposed method, the multivariate SHM data for two real long-span bridges are employed, including the acceleration, strain, displacement, humidity, and temperature data. The results demonstrate that transfer learning leads to a better classification capacity for the case of scarce labeled training data compared with the original network.

Prediction for segment strain and opening of underwater shield tunnel using deep learning method

Predicting the future mechanical behaviors of tunnel structure is essential to prevent disasters and maintain the long-term stability. Due to the underwater shield tunnels are usually constructed in complicated geological conditions, it is a challenge for traditional models to accurately predict the future behaviors considering multiple influence factors comprehensively. In this study, a multi-learning model termed as GC-GRU was presented on the basis of deep learning algorithm to predict the future mechanical behaviors of tunnel structure, which was formalized on the structural health monitoring data obtained from the Wuhan Yangtze River tunnel. Based on GC-GRU, temporal dependencies of historical performance and the spatial correlations among different monitoring indicators were captured, and the segment strain and opening in next 45 days were predicted. In addition, a series of experiments were conducted to discuss the predictive capability of the presented model, including the comparison to single indictor prediction model and some widely used classical prediction models, such as GRU, LSTM, XGboost, LR, and RNN. The comparison results denoted that GC-GRU performed best among all models especially when the prediction time scale reaches 20 days. The predicted errors of GC-GRU deduce at least 0.02 mm and 8.62με for joint opening and segment strain respectively, and model learning capability improves 2.2% for both of them. Therefore, it is reliable to introduce GC-GRU model to predict the future mechanical behaviors of tunnel structure.

Understanding structural health monitoring data to support decision-making processes and service life management of mass timber buildings. A preliminary study on use of data scaffolding

ABSTRACT Structural health monitoring (SHM) can be used to support decision-making processes leading to improved building management plans. With advanced engineered wooden buildings on the rise, SHM has emerged as a critical tool to document behaviour of these structures. However, monitoring data need to be easily accessible and understandable to support informed decisions. This study explores the potential benefits of an intentionally designed data scaffolding to support interpretation of SHM data for timber structures. Results suggest that appropriate scaffolding (e.g. pinpointing critical parameter ranges) can enable lay users to better interpret monitoring data of timber structures. Results also suggest that, for more highly educated users, access to graphs linked to metadata can be beneficial. Overall, it appears that documentation related to the monitored phenomena and context, as well as more explicit reference to building performance requirements, could improve data-driven decision making for all knowledge levels for the building maintenance sector.

Correlation of local and global structural damage state for SHM

In long-term structural health monitoring (SHM), one of the main challenges is to correlate an observed variation of vibration period with a damage level for a structure. The aim of this paper is to propose a methodology to quantify the seismic damage in a building through SHM data. For this purpose, an existing reinforced concrete building has been selected as a case-study and non-linear dynamics analyses have been carried out to estimate local and global damage levels. Ranges of variation of the vibration period of the building are derived for each damage level, and are compared with those previously assessed by the authors for single columns.

SHM of historical buildings: The case study of Santa Maria in Via church in Camerino (Italy)

Historical buildings constitute a huge heritage for the Italian country, and they deserve to be preserved to be handed over in good condition for future generations. As well know, Italy is a prone earthquake country and many earthquakes occurred over time producing awful consequences in terms of human losses and building damage. Historical buildings are typically the most damaged ones since they were built often with poor materials and conceived to sustain only gravitational loads. The preservation of these buildings is not a trivial task as structural retrofits have to be carried out not only in a proper way, but also timely in case the actual conditions of the construction may worse over time. Since large-scale rapid interventions after an earthquake are almost impossible, instrumenting damaged structures with appropriate sensors and analyzing measured data become crucial to control the damage evolution. At the same time, the monitoring of secured structures is essential to evaluate the effectiveness of the interventions and the possible evolution of damage. This paper presents the Structural Health Monitoring (SHM) system installed in the Santa Maria in Via church in Camerino (Central Italy). The church, severely damaged after the seismic sequence that stroke Central Italy in 2016, was secured with a steel structural system aimed at preventing possible collapse of some peculiar components, as the overturning of the massive façade body and the collapse of the slender elliptical drum. The SHM system permanently installed is presented; the latter consists of several sensors of different typologies placed in specific positions to monitor the effectiveness of the securing steel structure and the behavior of the church. Also, the Optimal Sensor Placement (OSP) methodology is adopted to determine the best sensor layout (number and position) to develop the SHM of the case study. Finally, the numerical model developed to support the interpretation of the SHM data, is described.

Wind load assessment with the JPDF of wind speed and direction based on SHM data

The calculation of basic wind speed (wind load) in return period is directly related to the wind resistance design of long-span bridge, high-rise building and etc. Nowadays, the calculation of basic wind speed in Chinese codes (GB50009-2012) mainly depends on the probability distribution function (PDF) of extreme wind speed, that is, univariate Gumbel PDF. However, this method doesn’t consider the influence of wind direction and the finite mixture (FM) of Gumbel PDF. In this paper, we use the Angular-linear (AL) model to characterize the extreme wind speed and wind direction, that is, joint PDF (JPDF). Specially, the FM Gumbel PDF is used to model the extreme wind speed and FM von Mises PDF is used to model the wind direction based on structural health monitoring (SHM) data. Then, the Genetic algorithm (GA) method is applied to solve the multi parameter solution in FM-AL model. Finally, the results of basic wind speed with return period of 100 years are compared based on the univariate Gumbel PDF (GB50009-2012 of Chinese code), FM Gumbel distribution, univariate Gumbel distribution + FM von Mises distribution and FM Gumbel distribution + FM von Mises distribution.

Time-dependent structural response estimation method for concrete structures using time information and convolutional neural networks

In this study, a structural response estimation method for evaluating the long-term strain, which can serve as the basis of assessing structural safety of concrete structures, is presented. Characteristics of short-term deformation caused by environmental influences and long-term inelastic deformation caused by combination of rheological effects and environmental influences are identified. To reflect these time-dependent short- and long-term deformation characteristics in the proposed method, the strain and corresponding time information collected from a structural health monitoring (SHM) are used. In the proposed method, the relationship between a strain response measured from a concrete structure and the corresponding time information including the year, month, day, and hour is defined by a convolutional neural network (CNN), which is a deep learning technique. The method assumes that the main sources of the strain in the structure are environmental influences, such as temperature and humidity, which have daily and seasonal periodicity, and rheological effects in concrete, such as creep and shrinkage, but it also assumes that the quantitative information on these sources is not available (e.g., environmental temperature and humidity, and rheological strain were not reliably measured). Thus, the CNN method developed in this study is trained only with strain and time data collected over several years, and used to estimate strain values within a specific timeframe, e.g., when the safety evaluation of the concrete structure is required or in case of SHM system failure or data loss. The presented method was developed and validated using SHM data from a real structure instrumented with fiber optic strain sensors. In addition, exploration of the constitution of the input of the CNN identified the type of time information that is the most effective in the long-term strain estimation of the developed method.

Determining Strain Threshold Values for Bridge Condition Assessment Using Normal Mixture Models

Structural health monitoring (SHM) enables bridge owners to evaluate bridge conditions efficiently and accurately. When implementing SHM, a popular method to detect the abnormal response of a structure is statistical pattern recognition. This often involves unsupervised statistical analysis due to a lack of measured SHM data from abnormal conditions. In this study, a novel methodology to calculate strain threshold indices (STIs) and establish decision boundaries (DBs) was used to detect the abnormal responses of the Indian River Inlet Bridge (IRIB), Sussex County, Delaware. First, a series of statistical models were applied and compared. Gaussian three mixture distributions were the optimal statistical models for the heavy vehicle-induced strain peaks. Threshold values for the strain gauges were selected using 99% uppers limits (USLs), and these limits were used to detect the abnormal response of the IRIB. The outlier ratios (R) for the sensors were calculated based on the threshold values. Corresponding STIs were defined by analyzing R, and DBs were determined using a t-distribution. The abnormal responses of the IRIB were detected by comparing the STI and DBs. The validity and sensitivity of the proposed methodology were demonstrated through simulated data that was created by perturbing the actual collected SHM data. Varying degrees of simulated damage were successfully detected using the proposed DBs. The proposed methodology showed promise when short and long-term abnormal responses and could provide practical guidance for bridge owners when using SHM data in their decision-making process.

Intelligent health indicator construction for prognostics of composite structures utilizing a semi-supervised deep neural network and SHM data

A health indicator (HI) is a valuable index demonstrating the health level of an engineering system or structure, which is a direct intermediate connection between raw signals collected by structural health monitoring (SHM) methods and prognostic models for remaining useful life estimation. An appropriate HI should conform to prognostic criteria, i.e., monotonicity, trendability, and prognosability, that are commonly utilized to measure the HI’s quality. However, constructing such a HI is challenging, particularly for composite structures due to their vulnerability to complex damage scenarios. Data-driven models and deep learning are powerful mathematical tools that can be employed to achieve this purpose. Yet the availability of a large dataset with labels plays a crucial role in these fields, and the data collected by SHM methods can only be labeled after the structure fails. In this respect, semi-supervised learning can incorporate unlabeled data monitored from structures that have not yet failed. In the present work, a semi-supervised deep neural network is proposed to construct HI by SHM data fusion. For the first time, the prognostic criteria are used as targets of the network rather than employing them only as a measurement tool of HI’s quality. In this regard, the acoustic emission method was used to monitor composite panels during fatigue loading, and extracted features were used to construct an intelligent HI. Finally, the proposed roadmap is evaluated by the holdout method, which shows a 77.3% improvement in the HI’s quality, and the leave-one-out cross-validation method, which indicates the generalized model has at least an 81.77% score on the prognostic criteria. This study demonstrates that even when the true HI labels are unknown but the qualified HI pattern (according to the prognostic criteria) can be recognized, a model can still be built that provides HIs aligning with the desired degradation behavior.

Investigation of modal parameters of a 600-m-tall skyscraper based on two-year-long structural health monitoring data and five typhoons measurements

On the basis of two-year-long structural health monitoring data, this paper investigates the structural dynamic properties of a 600-m-high supertall building installed with an active tuned mass damper (ATMD) system. The long-term modal parameters such as natural frequencies and damping ratios of the monitored skyscraper are first determined by the stochastic subspace identification approach, and the results generally agree well with those obtained by the finite element method and forced vibration tests. The statistical analysis of the modal parameters indicates that the natural frequencies follow the normal distribution, while the damping ratios fit well with the Gumbel distribution. In addition, the two-year-long variations of the modal parameters with ambient environmental factors are investigated and it is found that both the ambient temperature and humidity have limited effects on the modal parameters of the skyscraper. Moreover, the impacts of the ATMD system on the modal parameters are explored based on the field measurements during five typhoons. The analyzed results demonstrate that the active damper system had significant effects on the modal parameters of the fundamental sway modes, while it had marginal influences on these of higher modes. This study aims to enhance the understanding of structural dynamic properties of supertall buildings and provide useful information for their wind-resistant design.

Analysis of Structural Health Monitoring Data with Correlated Measurement Error by Bayesian System Identification: Theory and Application.

Measurement error is non-negligible and crucial in SHM data analysis. In many applications of SHM, measurement errors are statistically correlated in space and/or in time for data from sensor networks. Existing works solely consider spatial correlation for measurement error. When both spatial and temporal correlation are considered simultaneously, the existing works collapse, as they do not possess a suitable form describing spatially and temporally correlated measurement error. In order to tackle this burden, this paper generalizes the form of correlated measurement error from spatial correlation only or temporal correlation only to spatial-temporal correlation. A new form of spatial-temporal correlation and the corresponding likelihood function are proposed, and multiple candidate model classes for the measurement error are constructed, including no correlation, spatial correlation, temporal correlation, and the proposed spatial-temporal correlation. Bayesian system identification is conducted to achieve not only the posterior probability density function (PDF) for the model parameters, but also the posterior probability of each candidate model class for selecting the most suitable/plausible model class for the measurement error. Examples are presented with applications to model updating and modal frequency prediction under varying environmental conditions, ensuring the necessity of considering correlated measurement error and the capability of the proposed Bayesian system identification in the uncertainty quantification at the parameter and model levels.

An output‐only structural condition assessment method for civil structures by the stochastic gradient descent method

The interesting to assess the condition of a structure with structural health monitoring data has gained many attentions. Most of the existing methods require the measurement at the force location. This paper presents a novel output-only condition assessment method that does not require measurement at the force location. The unknown structural damage indices and input force can be identified with the stochastic gradient descent method. The dynamic acceleration response sensitivities with respect to the unknown structural damage indices and input force are derived analytically. Both unknown damage indices and unknown input force can be identified by minimizing the discrepancy between the measured and calculated vibration data. Numerical studies on a two-dimensional truss and seven-floor frame and experimental studies on a steel frame structure are presented to verify the accuracy and efficiency of the proposed method. Results demonstrate that the damage severity, location, and unknown input force can be identified. Also, the measurement at the force location is not required. Furthermore, when 20% measurement noise is considered, the identified error is less than 5%.

Guidelines for effective unsupervised guided wave compression and denoising in long-term guided wave structural health monitoring

This paper studies the effectiveness of joint compression and denoising strategies with realistic, long-term guided wave structural health monitoring data. We leverage the high correlation between nearby collections of guided waves in time to create sparse and low-rank representations. While compression and denoising schemes are not new, they are almost exclusively designed and studied with relatively simple datasets. In contrast, guided wave structural health monitoring datasets have much more complex operational and environmental conditions, such as temperature, that distort data and for which the requirements to achieve effective compression and denoising are not well understood. The paper studies how to optimize our data collection and algorithms to best utilize guided wave data for compression, denoising, and damage detection based on seven million guided wave measurements collected over 2 years.

LSTM approach for condition assessment of suspension bridges based on time-series deflection and temperature data

Deflection data provides important information about the mechanical characteristics and structural health condition of bridges. The study presented here pertains to development of a deep learning based approach for structural health monitoring by employing the bridge deflections. The method presented herein uses the long short-term memory (LSTM) framework in detecting the state of damage by tracking the feature changes of time-series deflection and temperature data. Deflection and temperature data of Chongqing Egongyan Rail Transit Suspension Bridge was employed over a period of 15 months to develop the proposed method. The concept of square error index (SE) is introduced as an assessment tool for estimation of the bridge damage level. Results from the present study indicated that the statistical characteristics of SE index are proportional to the level of damage, and are only sensitive to abnormal changes in deflection. Structural health monitoring data over the period of 15 months indicated that the proposed approach has the capability to detect cable damages as low as 0.5%.

Cumulative displacement-based detection of damper malfunction in bridges using data-driven isolation forest algorithm

Long-span cable-supported bridges are flexible and prone to significant deformations under temperature, wind, and vehicle loads. Such large deformations induce frequent longitudinal movement at the end of the bridge girder, which can cause malfunction or failure in the components such as expansion joints, bearings, or longitudinal viscous dampers. This paper develops a condition assessment approach using the girder end displacement measurements, which primarily includes four steps, namely, data preprocessing, feature characterization, datasets preparation and splitting, and anomaly detection. The data preprocessing step considers the physics-based correlation between the temperature and displacement, which is used to pre-check noisy data issues in the displacement monitoring data. Subsequently, hourly cumulative displacement is calculated as the feature to feed into the machine learning model. The third step prepares the datasets and splits them into training and testing datasets. Then the isolation forest algorithm is implemented for anomaly detection with well-tuned parameters of the number of decision trees and the number of samples in each decision tree. Structural health monitoring data of a suspension bridge are collected to verify the proposed approach, and the detection targets are specified as viscous damper malfunction and a specific holiday period. The proposed approach demonstrates its ability to detect both events successfully, which can shed light on the predictive and preventive maintenance of the bridge.

Long-term missing wind data recovery using free access databases and deep learning for bridge health monitoring

Wind environment monitoring at the bridge site is essential in structural health monitoring (SHM). However, as the performance of the electronic equipment for bridge SHM system deteriorates, wind monitoring data often suffer from long-term continuous data missing. The duration of data missing may be as long as several months, creating barriers to safety monitoring and monitoring data mining of the bridge structures. The conventional interpolation techniques hardly achieve the desired data recovery. Therefore, a framework was proposed for long-term missing wind data recovery based on a deep neural network (DNN) utilizing a free access database (European Center for Medium-Range Weather Forecasts, ECMWF). This framework consisted of one regression task (Task 1) and one temporal super-resolution task (Task 2). In Task 1, the hourly wind data provided by ECMWF were first learned to the hourly ones of the bridge SHM system. In Task 2, the low-resolution wind data (hourly averages) were upsampled to high-resolution ones (10-min averages). The U-net architecture provided the basis for the DNNs in both tasks. Unlike the conventional time-domain loss function used in Task 1, Task 2 adopted a time-frequency cross-domain loss function for training, which innovatively employed a spectrum magnitude balance strategy to enhance the reconstruction of the high-frequency components of wind speed signals. The proposed framework's feasibility and effectiveness were verified through a case study of recovering long-term continuous missing wind data in the SHM system of Sutong Bridge, China. The proposed methodology provides a new perspective for recovering long-term continuous missing SHM data.

Model calibration of a long-span concrete cable-stayed bridge based on structural health monitoring data: Influence of concrete variability

Structural Health Monitoring (SHM) systems, in combination with controlled load tests, can provide valuable data for calibrating high fidelity bridge models, which can then be used for evaluating the long-term performance of the bridge, improved load ratings, and permit vehicle evaluation. The objective of this research was to calibrate a 3D model of the Indian River Inlet (IRIB) cable-stayed bridge, using strains recorded by the bridge SHM system during a controlled load test. The bridge was modeled in STAAD-Pro and calibrated using a pre-commercialized software platform that uses a Generic Algorithm to minimize the error between the measured and predicted strains. The calibration parameters were the elastic modulus of groups of the main longitudinal edge girder/deck elements, which once calibrated, could be related to the measured concrete strength of the members. Four different models were investigated, using 6, 10, 14, and 18 parameter element groups of the edge girder members. Of the different models, the 14 and 18 parameter models yielded the best results. The “design” model yielded errors as high as 42% when compared to the measured strains; the error was less than 10% for the majority of measurements for the 14-parameter model. Including the effect of the traffic barriers in the model, the weighted average concrete strength of the calibrated model was within 4% of the measured weighted strength. The calibration was shown to be insensitive to measurement noise and was validated using several unique single and multi-vehicle load cases that were heavier and more offset from the centerline of the bridge. The calibration procedure was able to capture the variability in flexural stiffness of the edge girders due to the variability of the concrete, resulting in significantly better agreement between the live load measured strains and the model predicted strains.

Physics-Informed Data-Driven Prediction of 2D Normal Strain Field in Concrete Structures.

Concrete exhibits time-dependent long-term behavior driven by creep and shrinkage. These rheological effects are difficult to predict due to their stochastic nature and dependence on loading history. Existing empirical models used to predict rheological effects are fitted to databases composed largely of laboratory tests of limited time span and that do not capture differential rheological effects. A numerical model is typically required for application of empirical constitutive models to real structures. Notwithstanding this, the optimal parameters for the laboratory databases are not necessarily ideal for a specific structure. Data-driven approaches using structural health monitoring data have shown promise towards accurate prediction of long-term time-dependent behavior in concrete structures, but current approaches require different model parameters for each sensor and do not leverage geometry and loading. In this work, a physics-informed data-driven approach for long-term prediction of 2D normal strain field in prestressed concrete structures is introduced. The method employs a simplified analytical model of the structure, a data-driven model for prediction of the temperature field, and embedding of neural networks into rheological time-functions. In contrast to previous approaches, the model is trained on multiple sensors at once and enables the estimation of the strain evolution at any point of interest in the longitudinal section of the structure, capturing differential rheological effects.

Separation of the Temperature Effect on Structure Responses via LSTM—Particle Filter Method Considering Outlier from Remote Cloud Platforms

Structural health monitoring (SHM) has been widely applied in the field of Mechanical and Civil Engineering in recent years. It is very hard to detect damage, however, using the measured data directly from the remote cloud platform of on-site structure, owing to changing environmental conditions. At the same time, outlier data from the remote cloud platform often occurs due to the harsh environmental conditions, interferences in the wireless medium, and the usage of low-quality sensors, which can greatly reduce the accuracy of structural health monitoring. In this paper, a novel temperature compensation method based on a long-short term memory (LSTM) network and the particle filter (PF) is proposed to separate the temperature effect from long-term structural health monitoring data. This method takes LSTMs as the state equation of PF, which solves the problem whereby PF cannot accurately derive the state equation for complex structures. A feedback model using the probability distribution generated by PF is developed to filter the observed value, thus measurement outliers can be successfully reduced. A numerical simulation and the measured deflection data from an SHM system are utilized to verify the proposed method. Results from the numerical simulation show that the LSTM-PF method can satisfactorily compensate for the temperature effect even when the nonlinear temperature effect is considered. Moreover, outputs from the SHM system of a large-scale suspension bridge indicate the temperature effect can be compensated and outliers can be appropriately reduced at the same time using the measured deflection data.

Structural health monitoring data cleaning based on Bayesian robust tensor learning

Data cleaning is necessary to obtain high-quality monitoring data, which consists of two parts: anomaly identification and repair. A high-quality dataset is always used to repair data anomalies in current research, which may be difficult to obtain in practice because of the large number of data anomalies existing in real data. Besides, outliers, baseline shifts, and noise in dataset can lead to wrong missing restoration results. Therefore, the anomaly identification and location are needed to remove outliers, baseline shifts, and noise to regard the repair of these anomalies as missing data restoration. However, the identification of multiple data anomalies is time consuming and challenging. To this end, we propose a Bayesian robust tensor learning model to reconstruct the monitoring data tensor by extracting potential data features of the spatiotemporal tensor. A mixture of Laplace distribution with generalized inverse Gaussian distribution is used to model noise to enhance robustness. The tensor formed from the data containing anomalies can be input directly into the model without the identification and location of the data anomalies in advance. All data anomalies are repaired during the tensor reconstruction process. In addition, outliers, baseline shifts, and noise do not interfere with the missing restoration results owing to the robustness of the model. Furthermore, different cleaning strategies are proposed for dynamic and static response data. The data performance is verified using the real monitoring data of a concrete box girder bridge. The results show that the proposed method exhibits good data-cleaning performance when dealing with data containing a large number of data anomalies. In addition, as the abnormal rate increases, the proposed method still exhibits good data-cleaning performance.